04_multilingual-ner.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Uncomment and run this cell if you're on Colab or Kaggle\n",
    "# !git clone https://github.com/nlp-with-transformers/notebooks.git\n",
    "# %cd notebooks\n",
    "# from install import *\n",
    "# install_requirements()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using transformers v4.11.3\n",
      "Using datasets v1.13.0\n",
      "Using accelerate v0.5.1\n"
     ]
    }
   ],
   "source": [
    "#hide\n",
    "from utils import *\n",
    "setup_chapter()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Multilingual Named Entity Recognition"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## The Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>Jeff</td>\n",
       "      <td>Dean</td>\n",
       "      <td>is</td>\n",
       "      <td>a</td>\n",
       "      <td>computer</td>\n",
       "      <td>scientist</td>\n",
       "      <td>at</td>\n",
       "      <td>Google</td>\n",
       "      <td>in</td>\n",
       "      <td>California</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Tags</th>\n",
       "      <td>B-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>O</td>\n",
       "      <td>B-LOC</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#id jeff-dean-ner\n",
    "#caption An example of a sequence annotated with named entities\n",
    "#hide_input\n",
    "import pandas as pd\n",
    "toks = \"Jeff Dean is a computer scientist at Google in California\".split()\n",
    "lbls = [\"B-PER\", \"I-PER\", \"O\", \"O\", \"O\", \"O\", \"O\", \"B-ORG\", \"O\", \"B-LOC\"]\n",
    "df = pd.DataFrame(data=[toks, lbls], index=['Tokens', 'Tags'])\n",
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "XTREME has 183 configurations\n"
     ]
    }
   ],
   "source": [
    "from datasets import get_dataset_config_names\n",
    "\n",
    "xtreme_subsets = get_dataset_config_names(\"xtreme\")\n",
    "print(f\"XTREME has {len(xtreme_subsets)} configurations\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['PAN-X.af', 'PAN-X.ar', 'PAN-X.bg']"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "panx_subsets = [s for s in xtreme_subsets if s.startswith(\"PAN\")]\n",
    "panx_subsets[:3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e36aea61c9bc4620961a8a553452c716",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    validation: Dataset({\n",
       "        features: ['tokens', 'ner_tags', 'langs'],\n",
       "        num_rows: 10000\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['tokens', 'ner_tags', 'langs'],\n",
       "        num_rows: 10000\n",
       "    })\n",
       "    train: Dataset({\n",
       "        features: ['tokens', 'ner_tags', 'langs'],\n",
       "        num_rows: 20000\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# hide_output\n",
    "from datasets import load_dataset\n",
    "\n",
    "load_dataset(\"xtreme\", name=\"PAN-X.de\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e72fc7257d6a4b72b0d75edc220ace3c",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading and preparing dataset xtreme/PAN-X.fr (download: 223.17 MiB, generated: 6.37 MiB, post-processed: Unknown size, total: 229.65 MiB) to /Users/lewtun/.cache/huggingface/datasets/xtreme/PAN-X.fr/1.0.0/fb182342ff5c7a211ebf678cde070463acd29524b30b87f8f38c617948c2826a...\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset xtreme downloaded and prepared to /Users/lewtun/.cache/huggingface/datasets/xtreme/PAN-X.fr/1.0.0/fb182342ff5c7a211ebf678cde070463acd29524b30b87f8f38c617948c2826a. Subsequent calls will reuse this data.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "cbdd4421d79e4645804ee0e293bcd27c",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading and preparing dataset xtreme/PAN-X.it (download: 223.17 MiB, generated: 7.35 MiB, post-processed: Unknown size, total: 230.63 MiB) to /Users/lewtun/.cache/huggingface/datasets/xtreme/PAN-X.it/1.0.0/fb182342ff5c7a211ebf678cde070463acd29524b30b87f8f38c617948c2826a...\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset xtreme downloaded and prepared to /Users/lewtun/.cache/huggingface/datasets/xtreme/PAN-X.it/1.0.0/fb182342ff5c7a211ebf678cde070463acd29524b30b87f8f38c617948c2826a. Subsequent calls will reuse this data.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "df69d9802e33407587c275a6594a66b6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading and preparing dataset xtreme/PAN-X.en (download: 223.17 MiB, generated: 7.30 MiB, post-processed: Unknown size, total: 230.59 MiB) to /Users/lewtun/.cache/huggingface/datasets/xtreme/PAN-X.en/1.0.0/fb182342ff5c7a211ebf678cde070463acd29524b30b87f8f38c617948c2826a...\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset xtreme downloaded and prepared to /Users/lewtun/.cache/huggingface/datasets/xtreme/PAN-X.en/1.0.0/fb182342ff5c7a211ebf678cde070463acd29524b30b87f8f38c617948c2826a. Subsequent calls will reuse this data.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c6a47e1860db4c2ea599939049a63141",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# hide_output\n",
    "from collections import defaultdict\n",
    "from datasets import DatasetDict\n",
    "\n",
    "langs = [\"de\", \"fr\", \"it\", \"en\"]\n",
    "fracs = [0.629, 0.229, 0.084, 0.059]\n",
    "# Return a DatasetDict if a key doesn't exist\n",
    "panx_ch = defaultdict(DatasetDict)\n",
    "\n",
    "for lang, frac in zip(langs, fracs):\n",
    "    # Load monolingual corpus\n",
    "    ds = load_dataset(\"xtreme\", name=f\"PAN-X.{lang}\")\n",
    "    # Shuffle and downsample each split according to spoken proportion\n",
    "    for split in ds:\n",
    "        panx_ch[lang][split] = (\n",
    "            ds[split]\n",
    "            .shuffle(seed=0)\n",
    "            .select(range(int(frac * ds[split].num_rows))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>de</th>\n",
       "      <th>fr</th>\n",
       "      <th>it</th>\n",
       "      <th>en</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Number of training examples</th>\n",
       "      <td>12580</td>\n",
       "      <td>4580</td>\n",
       "      <td>1680</td>\n",
       "      <td>1180</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                de    fr    it    en\n",
       "Number of training examples  12580  4580  1680  1180"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "\n",
    "pd.DataFrame({lang: [panx_ch[lang][\"train\"].num_rows] for lang in langs},\n",
    "             index=[\"Number of training examples\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "langs: ['de', 'de', 'de', 'de', 'de', 'de', 'de', 'de', 'de', 'de', 'de', 'de']\n",
      "ner_tags: [0, 0, 0, 0, 5, 6, 0, 0, 5, 5, 6, 0]\n",
      "tokens: ['2.000', 'Einwohnern', 'an', 'der', 'Danziger', 'Bucht', 'in', 'der',\n",
      "'polnischen', 'Woiwodschaft', 'Pommern', '.']\n"
     ]
    }
   ],
   "source": [
    "element = panx_ch[\"de\"][\"train\"][0]\n",
    "for key, value in element.items():\n",
    "    print(f\"{key}: {value}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tokens: Sequence(feature=Value(dtype='string', id=None), length=-1, id=None)\n",
      "ner_tags: Sequence(feature=ClassLabel(num_classes=7, names=['O', 'B-PER',\n",
      "'I-PER', 'B-ORG', 'I-ORG', 'B-LOC', 'I-LOC'], names_file=None, id=None),\n",
      "length=-1, id=None)\n",
      "langs: Sequence(feature=Value(dtype='string', id=None), length=-1, id=None)\n"
     ]
    }
   ],
   "source": [
    "for key, value in panx_ch[\"de\"][\"train\"].features.items():\n",
    "    print(f\"{key}: {value}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ClassLabel(num_classes=7, names=['O', 'B-PER', 'I-PER', 'B-ORG', 'I-ORG',\n",
      "'B-LOC', 'I-LOC'], names_file=None, id=None)\n"
     ]
    }
   ],
   "source": [
    "tags = panx_ch[\"de\"][\"train\"].features[\"ner_tags\"].feature\n",
    "print(tags)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "def create_tag_names(batch):\n",
    "    return {\"ner_tags_str\": [tags.int2str(idx) for idx in batch[\"ner_tags\"]]}\n",
    "\n",
    "panx_de = panx_ch[\"de\"].map(create_tag_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>10</th>\n",
       "      <th>11</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>2.000</td>\n",
       "      <td>Einwohnern</td>\n",
       "      <td>an</td>\n",
       "      <td>der</td>\n",
       "      <td>Danziger</td>\n",
       "      <td>Bucht</td>\n",
       "      <td>in</td>\n",
       "      <td>der</td>\n",
       "      <td>polnischen</td>\n",
       "      <td>Woiwodschaft</td>\n",
       "      <td>Pommern</td>\n",
       "      <td>.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Tags</th>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>O</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           0           1   2    3         4      5   6    7           8   \\\n",
       "Tokens  2.000  Einwohnern  an  der  Danziger  Bucht  in  der  polnischen   \n",
       "Tags        O           O   O    O     B-LOC  I-LOC   O    O       B-LOC   \n",
       "\n",
       "                  9        10 11  \n",
       "Tokens  Woiwodschaft  Pommern  .  \n",
       "Tags           B-LOC    I-LOC  O  "
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# hide_output\n",
    "de_example = panx_de[\"train\"][0]\n",
    "pd.DataFrame([de_example[\"tokens\"], de_example[\"ner_tags_str\"]],\n",
    "['Tokens', 'Tags'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>ORG</th>\n",
       "      <th>LOC</th>\n",
       "      <th>PER</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>validation</th>\n",
       "      <td>2683</td>\n",
       "      <td>3172</td>\n",
       "      <td>2893</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>test</th>\n",
       "      <td>2573</td>\n",
       "      <td>3180</td>\n",
       "      <td>3071</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>train</th>\n",
       "      <td>5366</td>\n",
       "      <td>6186</td>\n",
       "      <td>5810</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             ORG   LOC   PER\n",
       "validation  2683  3172  2893\n",
       "test        2573  3180  3071\n",
       "train       5366  6186  5810"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from collections import Counter\n",
    "\n",
    "split2freqs = defaultdict(Counter)\n",
    "for split, dataset in panx_de.items():\n",
    "    for row in dataset[\"ner_tags_str\"]:\n",
    "        for tag in row:\n",
    "            if tag.startswith(\"B\"):\n",
    "                tag_type = tag.split(\"-\")[1]\n",
    "                split2freqs[split][tag_type] += 1\n",
    "pd.DataFrame.from_dict(split2freqs, orient=\"index\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Multilingual Transformers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## A Closer Look at Tokenization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1324109509bb4056b1d2ab8e9af0f9e4",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/29.0 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "70e765f2ded04031b1b09e6f6bdb8fd5",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/570 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2b04031c091e458cbe33613a4a71f26b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/208k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "cfc04444ed064ef183309691d7300553",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/426k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "acd2b13e44d641a4bef1d839a6106376",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/512 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2fed8c75130c48159d6d015af9840fe0",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/4.83M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "dce3206a657a43e3a66429554a870996",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/8.68M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# hide_output\n",
    "from transformers import AutoTokenizer\n",
    "\n",
    "bert_model_name = \"bert-base-cased\"\n",
    "xlmr_model_name = \"xlm-roberta-base\"\n",
    "bert_tokenizer = AutoTokenizer.from_pretrained(bert_model_name)\n",
    "xlmr_tokenizer = AutoTokenizer.from_pretrained(xlmr_model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "text = \"Jack Sparrow loves New York!\"\n",
    "bert_tokens = bert_tokenizer(text).tokens()\n",
    "xlmr_tokens = xlmr_tokenizer(text).tokens()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>BERT</th>\n",
       "      <td>[CLS]</td>\n",
       "      <td>Jack</td>\n",
       "      <td>Spa</td>\n",
       "      <td>##rrow</td>\n",
       "      <td>loves</td>\n",
       "      <td>New</td>\n",
       "      <td>York</td>\n",
       "      <td>!</td>\n",
       "      <td>[SEP]</td>\n",
       "      <td>None</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>XLM-R</th>\n",
       "      <td>&lt;s&gt;</td>\n",
       "      <td>▁Jack</td>\n",
       "      <td>▁Spar</td>\n",
       "      <td>row</td>\n",
       "      <td>▁love</td>\n",
       "      <td>s</td>\n",
       "      <td>▁New</td>\n",
       "      <td>▁York</td>\n",
       "      <td>!</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#hide_input\n",
    "df = pd.DataFrame([bert_tokens, xlmr_tokens], index=[\"BERT\", \"XLM-R\"])\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### The Tokenizer Pipeline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img alt=\"Tokenizer pipeline\" caption=\"The steps in the tokenization pipeline\" src=\"images/chapter04_tokenizer-pipeline.png\" id=\"toknizer-pipeline\"/>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### The SentencePiece Tokenizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'<s> Jack Sparrow loves New York!</s>'"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\"\".join(xlmr_tokens).replace(u\"\\u2581\", \" \")"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "[[train_ner_tagger]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Transformers for Named Entity Recognition"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img alt=\"Architecture of a transformer encoder for classification.\" caption=\"Fine-tuning an encoder-based transformer for sequence classification\" src=\"images/chapter04_clf-architecture.png\" id=\"clf-arch\"/>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img alt=\"Architecture of a transformer encoder for named entity recognition. The wide linear layer shows that the same linear layer is applied to all hidden states.\" caption=\"Fine-tuning an encoder-based transformer for named entity recognition\" src=\"images/chapter04_ner-architecture.png\" id=\"ner-arch\"/>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## The Anatomy of the Transformers Model Class"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Bodies and Heads"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img alt=\"bert-body-head\" caption=\"The `BertModel` class only contains the body of the model, while the `BertFor&lt;Task&gt;` classes combine the body with a dedicated head for a given task\" src=\"images/chapter04_bert-body-head.png\" id=\"bert-body-head\"/>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Creating a Custom Model for Token Classification"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.nn as nn\n",
    "from transformers import XLMRobertaConfig\n",
    "from transformers.modeling_outputs import TokenClassifierOutput\n",
    "from transformers.models.roberta.modeling_roberta import RobertaModel\n",
    "from transformers.models.roberta.modeling_roberta import RobertaPreTrainedModel\n",
    "\n",
    "class XLMRobertaForTokenClassification(RobertaPreTrainedModel):\n",
    "    config_class = XLMRobertaConfig\n",
    "\n",
    "    def __init__(self, config):\n",
    "        super().__init__(config)\n",
    "        self.num_labels = config.num_labels\n",
    "        # Load model body\n",
    "        self.roberta = RobertaModel(config, add_pooling_layer=False)\n",
    "        # Set up token classification head\n",
    "        self.dropout = nn.Dropout(config.hidden_dropout_prob)\n",
    "        self.classifier = nn.Linear(config.hidden_size, config.num_labels)\n",
    "        # Load and initialize weights\n",
    "        self.init_weights()\n",
    "\n",
    "    def forward(self, input_ids=None, attention_mask=None, token_type_ids=None, \n",
    "                labels=None, **kwargs):\n",
    "        # Use model body to get encoder representations\n",
    "        outputs = self.roberta(input_ids, attention_mask=attention_mask,\n",
    "                               token_type_ids=token_type_ids, **kwargs)\n",
    "        # Apply classifier to encoder representation\n",
    "        sequence_output = self.dropout(outputs[0])\n",
    "        logits = self.classifier(sequence_output)\n",
    "        # Calculate losses\n",
    "        loss = None\n",
    "        if labels is not None:\n",
    "            loss_fct = nn.CrossEntropyLoss()\n",
    "            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))\n",
    "        # Return model output object\n",
    "        return TokenClassifierOutput(loss=loss, logits=logits, \n",
    "                                     hidden_states=outputs.hidden_states, \n",
    "                                     attentions=outputs.attentions)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loading a Custom Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "index2tag = {idx: tag for idx, tag in enumerate(tags.names)}\n",
    "tag2index = {tag: idx for idx, tag in enumerate(tags.names)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "from transformers import AutoConfig\n",
    "\n",
    "xlmr_config = AutoConfig.from_pretrained(xlmr_model_name, \n",
    "                                         num_labels=tags.num_classes,\n",
    "                                         id2label=index2tag, label2id=tag2index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "import torch\n",
    "\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "xlmr_model = (XLMRobertaForTokenClassification\n",
    "              .from_pretrained(xlmr_model_name, config=xlmr_config)\n",
    "              .to(device))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>&lt;s&gt;</td>\n",
       "      <td>▁Jack</td>\n",
       "      <td>▁Spar</td>\n",
       "      <td>row</td>\n",
       "      <td>▁love</td>\n",
       "      <td>s</td>\n",
       "      <td>▁New</td>\n",
       "      <td>▁York</td>\n",
       "      <td>!</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Input IDs</th>\n",
       "      <td>0</td>\n",
       "      <td>21763</td>\n",
       "      <td>37456</td>\n",
       "      <td>15555</td>\n",
       "      <td>5161</td>\n",
       "      <td>7</td>\n",
       "      <td>2356</td>\n",
       "      <td>5753</td>\n",
       "      <td>38</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             0      1      2      3      4  5     6      7   8     9\n",
       "Tokens     <s>  ▁Jack  ▁Spar    row  ▁love  s  ▁New  ▁York   !  </s>\n",
       "Input IDs    0  21763  37456  15555   5161  7  2356   5753  38     2"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# hide_output\n",
    "input_ids = xlmr_tokenizer.encode(text, return_tensors=\"pt\")\n",
    "pd.DataFrame([xlmr_tokens, input_ids[0].numpy()], index=[\"Tokens\", \"Input IDs\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of tokens in sequence: 10\n",
      "Shape of outputs: torch.Size([1, 10, 7])\n"
     ]
    }
   ],
   "source": [
    "outputs = xlmr_model(input_ids.to(device)).logits\n",
    "predictions = torch.argmax(outputs, dim=-1)\n",
    "print(f\"Number of tokens in sequence: {len(xlmr_tokens)}\")\n",
    "print(f\"Shape of outputs: {outputs.shape}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>&lt;s&gt;</td>\n",
       "      <td>▁Jack</td>\n",
       "      <td>▁Spar</td>\n",
       "      <td>row</td>\n",
       "      <td>▁love</td>\n",
       "      <td>s</td>\n",
       "      <td>▁New</td>\n",
       "      <td>▁York</td>\n",
       "      <td>!</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Tags</th>\n",
       "      <td>O</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>O</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>B-LOC</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          0      1      2      3      4      5     6      7      8      9\n",
       "Tokens  <s>  ▁Jack  ▁Spar    row  ▁love      s  ▁New  ▁York      !   </s>\n",
       "Tags      O  I-LOC  B-LOC  B-LOC      O  I-LOC     O      O  I-LOC  B-LOC"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds = [tags.names[p] for p in predictions[0].cpu().numpy()]\n",
    "pd.DataFrame([xlmr_tokens, preds], index=[\"Tokens\", \"Tags\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tag_text(text, tags, model, tokenizer):\n",
    "    # Get tokens with special characters\n",
    "    tokens = tokenizer(text).tokens()\n",
    "    # Encode the sequence into IDs\n",
    "    input_ids = xlmr_tokenizer(text, return_tensors=\"pt\").input_ids.to(device)\n",
    "    # Get predictions as distribution over 7 possible classes\n",
    "    outputs = model(input_ids)[0]\n",
    "    # Take argmax to get most likely class per token\n",
    "    predictions = torch.argmax(outputs, dim=2)\n",
    "    # Convert to DataFrame\n",
    "    preds = [tags.names[p] for p in predictions[0].cpu().numpy()]\n",
    "    return pd.DataFrame([tokens, preds], index=[\"Tokens\", \"Tags\"])\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Tokenizing Texts for NER"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "words, labels = de_example[\"tokens\"], de_example[\"ner_tags\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenized_input = xlmr_tokenizer(de_example[\"tokens\"], is_split_into_words=True)\n",
    "tokens = xlmr_tokenizer.convert_ids_to_tokens(tokenized_input[\"input_ids\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>...</th>\n",
       "      <th>15</th>\n",
       "      <th>16</th>\n",
       "      <th>17</th>\n",
       "      <th>18</th>\n",
       "      <th>19</th>\n",
       "      <th>20</th>\n",
       "      <th>21</th>\n",
       "      <th>22</th>\n",
       "      <th>23</th>\n",
       "      <th>24</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>&lt;s&gt;</td>\n",
       "      <td>▁2.000</td>\n",
       "      <td>▁Einwohner</td>\n",
       "      <td>n</td>\n",
       "      <td>▁an</td>\n",
       "      <td>▁der</td>\n",
       "      <td>▁Dan</td>\n",
       "      <td>zi</td>\n",
       "      <td>ger</td>\n",
       "      <td>▁Buch</td>\n",
       "      <td>...</td>\n",
       "      <td>▁Wo</td>\n",
       "      <td>i</td>\n",
       "      <td>wod</td>\n",
       "      <td>schaft</td>\n",
       "      <td>▁Po</td>\n",
       "      <td>mmer</td>\n",
       "      <td>n</td>\n",
       "      <td>▁</td>\n",
       "      <td>.</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1 rows × 25 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "         0       1           2  3    4     5     6   7    8      9   ...   15  \\\n",
       "Tokens  <s>  ▁2.000  ▁Einwohner  n  ▁an  ▁der  ▁Dan  zi  ger  ▁Buch  ...  ▁Wo   \n",
       "\n",
       "       16   17      18   19    20 21 22 23    24  \n",
       "Tokens  i  wod  schaft  ▁Po  mmer  n  ▁  .  </s>  \n",
       "\n",
       "[1 rows x 25 columns]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#hide_output\n",
    "pd.DataFrame([tokens], index=[\"Tokens\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>...</th>\n",
       "      <th>15</th>\n",
       "      <th>16</th>\n",
       "      <th>17</th>\n",
       "      <th>18</th>\n",
       "      <th>19</th>\n",
       "      <th>20</th>\n",
       "      <th>21</th>\n",
       "      <th>22</th>\n",
       "      <th>23</th>\n",
       "      <th>24</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>&lt;s&gt;</td>\n",
       "      <td>▁2.000</td>\n",
       "      <td>▁Einwohner</td>\n",
       "      <td>n</td>\n",
       "      <td>▁an</td>\n",
       "      <td>▁der</td>\n",
       "      <td>▁Dan</td>\n",
       "      <td>zi</td>\n",
       "      <td>ger</td>\n",
       "      <td>▁Buch</td>\n",
       "      <td>...</td>\n",
       "      <td>▁Wo</td>\n",
       "      <td>i</td>\n",
       "      <td>wod</td>\n",
       "      <td>schaft</td>\n",
       "      <td>▁Po</td>\n",
       "      <td>mmer</td>\n",
       "      <td>n</td>\n",
       "      <td>▁</td>\n",
       "      <td>.</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Word IDs</th>\n",
       "      <td>None</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>3</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>5</td>\n",
       "      <td>...</td>\n",
       "      <td>9</td>\n",
       "      <td>9</td>\n",
       "      <td>9</td>\n",
       "      <td>9</td>\n",
       "      <td>10</td>\n",
       "      <td>10</td>\n",
       "      <td>10</td>\n",
       "      <td>11</td>\n",
       "      <td>11</td>\n",
       "      <td>None</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 25 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "            0       1           2  3    4     5     6   7    8      9   ...  \\\n",
       "Tokens     <s>  ▁2.000  ▁Einwohner  n  ▁an  ▁der  ▁Dan  zi  ger  ▁Buch  ...   \n",
       "Word IDs  None       0           1  1    2     3     4   4    4      5  ...   \n",
       "\n",
       "           15 16   17      18   19    20  21  22  23    24  \n",
       "Tokens    ▁Wo  i  wod  schaft  ▁Po  mmer   n   ▁   .  </s>  \n",
       "Word IDs    9  9    9       9   10    10  10  11  11  None  \n",
       "\n",
       "[2 rows x 25 columns]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# hide_output\n",
    "word_ids = tokenized_input.word_ids()\n",
    "pd.DataFrame([tokens, word_ids], index=[\"Tokens\", \"Word IDs\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>...</th>\n",
       "      <th>15</th>\n",
       "      <th>16</th>\n",
       "      <th>17</th>\n",
       "      <th>18</th>\n",
       "      <th>19</th>\n",
       "      <th>20</th>\n",
       "      <th>21</th>\n",
       "      <th>22</th>\n",
       "      <th>23</th>\n",
       "      <th>24</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>&lt;s&gt;</td>\n",
       "      <td>▁2.000</td>\n",
       "      <td>▁Einwohner</td>\n",
       "      <td>n</td>\n",
       "      <td>▁an</td>\n",
       "      <td>▁der</td>\n",
       "      <td>▁Dan</td>\n",
       "      <td>zi</td>\n",
       "      <td>ger</td>\n",
       "      <td>▁Buch</td>\n",
       "      <td>...</td>\n",
       "      <td>▁Wo</td>\n",
       "      <td>i</td>\n",
       "      <td>wod</td>\n",
       "      <td>schaft</td>\n",
       "      <td>▁Po</td>\n",
       "      <td>mmer</td>\n",
       "      <td>n</td>\n",
       "      <td>▁</td>\n",
       "      <td>.</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Word IDs</th>\n",
       "      <td>None</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>3</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>5</td>\n",
       "      <td>...</td>\n",
       "      <td>9</td>\n",
       "      <td>9</td>\n",
       "      <td>9</td>\n",
       "      <td>9</td>\n",
       "      <td>10</td>\n",
       "      <td>10</td>\n",
       "      <td>10</td>\n",
       "      <td>11</td>\n",
       "      <td>11</td>\n",
       "      <td>None</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Label IDs</th>\n",
       "      <td>-100</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>-100</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>-100</td>\n",
       "      <td>-100</td>\n",
       "      <td>6</td>\n",
       "      <td>...</td>\n",
       "      <td>5</td>\n",
       "      <td>-100</td>\n",
       "      <td>-100</td>\n",
       "      <td>-100</td>\n",
       "      <td>6</td>\n",
       "      <td>-100</td>\n",
       "      <td>-100</td>\n",
       "      <td>0</td>\n",
       "      <td>-100</td>\n",
       "      <td>-100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Labels</th>\n",
       "      <td>IGN</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>IGN</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>...</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "      <td>O</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4 rows × 25 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "             0       1           2     3    4     5      6     7     8   \\\n",
       "Tokens      <s>  ▁2.000  ▁Einwohner     n  ▁an  ▁der   ▁Dan    zi   ger   \n",
       "Word IDs   None       0           1     1    2     3      4     4     4   \n",
       "Label IDs  -100       0           0  -100    0     0      5  -100  -100   \n",
       "Labels      IGN       O           O   IGN    O     O  B-LOC   IGN   IGN   \n",
       "\n",
       "              9   ...     15    16    17      18     19    20    21  22    23  \\\n",
       "Tokens     ▁Buch  ...    ▁Wo     i   wod  schaft    ▁Po  mmer     n   ▁     .   \n",
       "Word IDs       5  ...      9     9     9       9     10    10    10  11    11   \n",
       "Label IDs      6  ...      5  -100  -100    -100      6  -100  -100   0  -100   \n",
       "Labels     I-LOC  ...  B-LOC   IGN   IGN     IGN  I-LOC   IGN   IGN   O   IGN   \n",
       "\n",
       "             24  \n",
       "Tokens     </s>  \n",
       "Word IDs   None  \n",
       "Label IDs  -100  \n",
       "Labels      IGN  \n",
       "\n",
       "[4 rows x 25 columns]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#hide_output\n",
    "previous_word_idx = None\n",
    "label_ids = []\n",
    "\n",
    "for word_idx in word_ids:\n",
    "    if word_idx is None or word_idx == previous_word_idx:\n",
    "        label_ids.append(-100)\n",
    "    elif word_idx != previous_word_idx:\n",
    "        label_ids.append(labels[word_idx])\n",
    "    previous_word_idx = word_idx\n",
    "    \n",
    "labels = [index2tag[l] if l != -100 else \"IGN\" for l in label_ids]\n",
    "index = [\"Tokens\", \"Word IDs\", \"Label IDs\", \"Labels\"]\n",
    "\n",
    "pd.DataFrame([tokens, word_ids, label_ids, labels], index=index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tokenize_and_align_labels(examples):\n",
    "    tokenized_inputs = xlmr_tokenizer(examples[\"tokens\"], truncation=True, \n",
    "                                      is_split_into_words=True)\n",
    "    labels = []\n",
    "    for idx, label in enumerate(examples[\"ner_tags\"]):\n",
    "        word_ids = tokenized_inputs.word_ids(batch_index=idx)\n",
    "        previous_word_idx = None\n",
    "        label_ids = []\n",
    "        for word_idx in word_ids:\n",
    "            if word_idx is None or word_idx == previous_word_idx:\n",
    "                label_ids.append(-100)\n",
    "            else:\n",
    "                label_ids.append(label[word_idx])\n",
    "            previous_word_idx = word_idx\n",
    "        labels.append(label_ids)\n",
    "    tokenized_inputs[\"labels\"] = labels\n",
    "    return tokenized_inputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def encode_panx_dataset(corpus):\n",
    "    return corpus.map(tokenize_and_align_labels, batched=True, \n",
    "                      remove_columns=['langs', 'ner_tags', 'tokens'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "panx_de_encoded = encode_panx_dataset(panx_ch[\"de\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Performance Measures"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "        MISC       0.00      0.00      0.00         1\n",
      "         PER       1.00      1.00      1.00         1\n",
      "\n",
      "   micro avg       0.50      0.50      0.50         2\n",
      "   macro avg       0.50      0.50      0.50         2\n",
      "weighted avg       0.50      0.50      0.50         2\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from seqeval.metrics import classification_report\n",
    "\n",
    "y_true = [[\"O\", \"O\", \"O\", \"B-MISC\", \"I-MISC\", \"I-MISC\", \"O\"],\n",
    "          [\"B-PER\", \"I-PER\", \"O\"]]\n",
    "y_pred = [[\"O\", \"O\", \"B-MISC\", \"I-MISC\", \"I-MISC\", \"I-MISC\", \"O\"],\n",
    "          [\"B-PER\", \"I-PER\", \"O\"]]\n",
    "print(classification_report(y_true, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "def align_predictions(predictions, label_ids):\n",
    "    preds = np.argmax(predictions, axis=2)\n",
    "    batch_size, seq_len = preds.shape\n",
    "    labels_list, preds_list = [], []\n",
    "\n",
    "    for batch_idx in range(batch_size):\n",
    "        example_labels, example_preds = [], []\n",
    "        for seq_idx in range(seq_len):\n",
    "            # Ignore label IDs = -100\n",
    "            if label_ids[batch_idx, seq_idx] != -100:\n",
    "                example_labels.append(index2tag[label_ids[batch_idx][seq_idx]])\n",
    "                example_preds.append(index2tag[preds[batch_idx][seq_idx]])\n",
    "\n",
    "        labels_list.append(example_labels)\n",
    "        preds_list.append(example_preds)\n",
    "\n",
    "    return preds_list, labels_list"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Fine-Tuning XLM-RoBERTa"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "from transformers import TrainingArguments\n",
    "\n",
    "num_epochs = 3\n",
    "batch_size = 24\n",
    "logging_steps = len(panx_de_encoded[\"train\"]) // batch_size\n",
    "model_name = f\"{xlmr_model_name}-finetuned-panx-de\"\n",
    "training_args = TrainingArguments(\n",
    "    output_dir=model_name, log_level=\"error\", num_train_epochs=num_epochs, \n",
    "    per_device_train_batch_size=batch_size, \n",
    "    per_device_eval_batch_size=batch_size, evaluation_strategy=\"epoch\", \n",
    "    save_steps=1e6, weight_decay=0.01, disable_tqdm=False, \n",
    "    logging_steps=logging_steps, push_to_hub=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "96bce6906b5d4713a064d40b8808c234",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(HTML(value=\"<center>\\n<img src=https://huggingface.co/front/assets/huggingface_logo-noborder.sv…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#hide_output\n",
    "from huggingface_hub import notebook_login\n",
    "\n",
    "notebook_login()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from seqeval.metrics import f1_score\n",
    "\n",
    "def compute_metrics(eval_pred):\n",
    "    y_pred, y_true = align_predictions(eval_pred.predictions, \n",
    "                                       eval_pred.label_ids)\n",
    "    return {\"f1\": f1_score(y_true, y_pred)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import DataCollatorForTokenClassification\n",
    "\n",
    "data_collator = DataCollatorForTokenClassification(xlmr_tokenizer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def model_init():\n",
    "    return (XLMRobertaForTokenClassification\n",
    "            .from_pretrained(xlmr_model_name, config=xlmr_config)\n",
    "            .to(device))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "env: TOKENIZERS_PARALLELISM=false\n"
     ]
    }
   ],
   "source": [
    "#hide\n",
    "%env TOKENIZERS_PARALLELISM=false"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "CODECARBON : No CPU tracking mode found. Falling back on CPU constant mode.\n",
      "CODECARBON : Failed to match CPU TDP constant. Falling back on a global constant.\n",
      "/home/lewtun/miniconda3/envs/book/lib/python3.7/site-packages/apscheduler/util.py:95: PytzUsageWarning: The zone attribute is specific to pytz's interface; please migrate to a new time zone provider. For more details on how to do so, see https://pytz-deprecation-shim.readthedocs.io/en/latest/migration.html\n",
      "  if obj.zone == 'local':\n"
     ]
    }
   ],
   "source": [
    "# hide_output\n",
    "from transformers import Trainer\n",
    "\n",
    "trainer = Trainer(model_init=model_init, args=training_args, \n",
    "                  data_collator=data_collator, compute_metrics=compute_metrics,\n",
    "                  train_dataset=panx_de_encoded[\"train\"],\n",
    "                  eval_dataset=panx_de_encoded[\"validation\"], \n",
    "                  tokenizer=xlmr_tokenizer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#hide_input\n",
    "trainer.train()\n",
    "trainer.push_to_hub(commit_message=\"Training completed!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>Epoch</th>\n",
       "      <th>Training Loss</th>\n",
       "      <th>Validation Loss</th>\n",
       "      <th>F1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.2652</td>\n",
       "      <td>0.160244</td>\n",
       "      <td>0.822974</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.1314</td>\n",
       "      <td>0.137195</td>\n",
       "      <td>0.852747</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.0806</td>\n",
       "      <td>0.138774</td>\n",
       "      <td>0.864591</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# hide_input\n",
    "df = pd.DataFrame(trainer.state.log_history)[['epoch','loss' ,'eval_loss', 'eval_f1']]\n",
    "df = df.rename(columns={\"epoch\":\"Epoch\",\"loss\": \"Training Loss\", \"eval_loss\": \"Validation Loss\", \"eval_f1\":\"F1\"})\n",
    "df['Epoch'] = df[\"Epoch\"].apply(lambda x: round(x))\n",
    "df['Training Loss'] = df[\"Training Loss\"].ffill()\n",
    "df[['Validation Loss', 'F1']] = df[['Validation Loss', 'F1']].bfill().ffill()\n",
    "df.drop_duplicates()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>10</th>\n",
       "      <th>11</th>\n",
       "      <th>12</th>\n",
       "      <th>13</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>&lt;s&gt;</td>\n",
       "      <td>▁Jeff</td>\n",
       "      <td>▁De</td>\n",
       "      <td>an</td>\n",
       "      <td>▁ist</td>\n",
       "      <td>▁ein</td>\n",
       "      <td>▁Informati</td>\n",
       "      <td>ker</td>\n",
       "      <td>▁bei</td>\n",
       "      <td>▁Google</td>\n",
       "      <td>▁in</td>\n",
       "      <td>▁Kaliforni</td>\n",
       "      <td>en</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Tags</th>\n",
       "      <td>O</td>\n",
       "      <td>B-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>O</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>O</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         0      1      2      3     4     5           6    7     8        9   \\\n",
       "Tokens  <s>  ▁Jeff    ▁De     an  ▁ist  ▁ein  ▁Informati  ker  ▁bei  ▁Google   \n",
       "Tags      O  B-PER  I-PER  I-PER     O     O           O    O     O    B-ORG   \n",
       "\n",
       "         10          11     12    13  \n",
       "Tokens  ▁in  ▁Kaliforni     en  </s>  \n",
       "Tags      O       B-LOC  I-LOC     O  "
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# hide_output\n",
    "text_de = \"Jeff Dean ist ein Informatiker bei Google in Kalifornien\"\n",
    "tag_text(text_de, tags, trainer.model, xlmr_tokenizer)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Error Analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.nn.functional import cross_entropy\n",
    "\n",
    "def forward_pass_with_label(batch):\n",
    "    # Convert dict of lists to list of dicts suitable for data collator\n",
    "    features = [dict(zip(batch, t)) for t in zip(*batch.values())]\n",
    "    # Pad inputs and labels and put all tensors on device\n",
    "    batch = data_collator(features)\n",
    "    input_ids = batch[\"input_ids\"].to(device)\n",
    "    attention_mask = batch[\"attention_mask\"].to(device)\n",
    "    labels = batch[\"labels\"].to(device)\n",
    "    with torch.no_grad():\n",
    "        # Pass data through model  \n",
    "        output = trainer.model(input_ids, attention_mask)\n",
    "        # Logit.size: [batch_size, sequence_length, classes]\n",
    "        # Predict class with largest logit value on classes axis\n",
    "        predicted_label = torch.argmax(output.logits, axis=-1).cpu().numpy()\n",
    "    # Calculate loss per token after flattening batch dimension with view\n",
    "    loss = cross_entropy(output.logits.view(-1, 7), \n",
    "                         labels.view(-1), reduction=\"none\")\n",
    "    # Unflatten batch dimension and convert to numpy array\n",
    "    loss = loss.view(len(input_ids), -1).cpu().numpy()\n",
    "\n",
    "    return {\"loss\":loss, \"predicted_label\": predicted_label}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5695f889b4c54f0599afc6daa6dffd38",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/197 [00:00<?, ?ba/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# hide_output\n",
    "valid_set = panx_de_encoded[\"validation\"]\n",
    "valid_set = valid_set.map(forward_pass_with_label, batched=True, batch_size=32)\n",
    "df = valid_set.to_pandas()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>attention_mask</th>\n",
       "      <th>input_ids</th>\n",
       "      <th>labels</th>\n",
       "      <th>loss</th>\n",
       "      <th>predicted_label</th>\n",
       "      <th>input_tokens</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[1, 1, 1, 1, 1, 1, 1]</td>\n",
       "      <td>[0, 10699, 11, 15, 16104, 1388, 2]</td>\n",
       "      <td>[IGN, B-ORG, IGN, I-ORG, I-ORG, I-ORG, IGN]</td>\n",
       "      <td>[0.0, 0.014679872, 0.0, 0.009469474, 0.0103934...</td>\n",
       "      <td>[I-ORG, B-ORG, I-ORG, I-ORG, I-ORG, I-ORG, I-ORG]</td>\n",
       "      <td>[&lt;s&gt;, ▁Ham, a, ▁(, ▁Unternehmen, ▁), &lt;/s&gt;]</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          attention_mask                           input_ids  \\\n",
       "0  [1, 1, 1, 1, 1, 1, 1]  [0, 10699, 11, 15, 16104, 1388, 2]   \n",
       "\n",
       "                                        labels  \\\n",
       "0  [IGN, B-ORG, IGN, I-ORG, I-ORG, I-ORG, IGN]   \n",
       "\n",
       "                                                loss  \\\n",
       "0  [0.0, 0.014679872, 0.0, 0.009469474, 0.0103934...   \n",
       "\n",
       "                                     predicted_label  \\\n",
       "0  [I-ORG, B-ORG, I-ORG, I-ORG, I-ORG, I-ORG, I-ORG]   \n",
       "\n",
       "                                 input_tokens  \n",
       "0  [<s>, ▁Ham, a, ▁(, ▁Unternehmen, ▁), </s>]  "
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# hide_output\n",
    "index2tag[-100] = \"IGN\"\n",
    "df[\"input_tokens\"] = df[\"input_ids\"].apply(\n",
    "    lambda x: xlmr_tokenizer.convert_ids_to_tokens(x))\n",
    "df[\"predicted_label\"] = df[\"predicted_label\"].apply(\n",
    "    lambda x: [index2tag[i] for i in x])\n",
    "df[\"labels\"] = df[\"labels\"].apply(\n",
    "    lambda x: [index2tag[i] for i in x])\n",
    "df['loss'] = df.apply(\n",
    "    lambda x: x['loss'][:len(x['input_ids'])], axis=1)\n",
    "df['predicted_label'] = df.apply(\n",
    "    lambda x: x['predicted_label'][:len(x['input_ids'])], axis=1)\n",
    "df.head(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>attention_mask</th>\n",
       "      <th>input_ids</th>\n",
       "      <th>labels</th>\n",
       "      <th>loss</th>\n",
       "      <th>predicted_label</th>\n",
       "      <th>input_tokens</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>10699</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>0.01</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>▁Ham</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>15</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>0.01</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>▁(</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>16104</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>0.01</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>▁Unternehmen</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>1388</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>0.01</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>▁)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>56530</td>\n",
       "      <td>O</td>\n",
       "      <td>0.00</td>\n",
       "      <td>O</td>\n",
       "      <td>▁WE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>83982</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>0.34</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>▁Luz</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>10</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>0.45</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>▁a</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  attention_mask input_ids labels  loss predicted_label  input_tokens\n",
       "0              1     10699  B-ORG  0.01           B-ORG          ▁Ham\n",
       "0              1        15  I-ORG  0.01           I-ORG            ▁(\n",
       "0              1     16104  I-ORG  0.01           I-ORG  ▁Unternehmen\n",
       "0              1      1388  I-ORG  0.01           I-ORG            ▁)\n",
       "1              1     56530      O  0.00               O           ▁WE\n",
       "1              1     83982  B-ORG  0.34           B-ORG          ▁Luz\n",
       "1              1        10  I-ORG  0.45           I-ORG            ▁a"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# hide_output\n",
    "df_tokens = df.apply(pd.Series.explode)\n",
    "df_tokens = df_tokens.query(\"labels != 'IGN'\")\n",
    "df_tokens[\"loss\"] = df_tokens[\"loss\"].astype(float).round(2)\n",
    "df_tokens.head(7)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>input_tokens</th>\n",
       "      <td>▁</td>\n",
       "      <td>▁der</td>\n",
       "      <td>▁in</td>\n",
       "      <td>▁von</td>\n",
       "      <td>▁/</td>\n",
       "      <td>▁und</td>\n",
       "      <td>▁(</td>\n",
       "      <td>▁)</td>\n",
       "      <td>▁''</td>\n",
       "      <td>▁A</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>6066</td>\n",
       "      <td>1388</td>\n",
       "      <td>989</td>\n",
       "      <td>808</td>\n",
       "      <td>163</td>\n",
       "      <td>1171</td>\n",
       "      <td>246</td>\n",
       "      <td>246</td>\n",
       "      <td>2898</td>\n",
       "      <td>125</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.03</td>\n",
       "      <td>0.1</td>\n",
       "      <td>0.14</td>\n",
       "      <td>0.14</td>\n",
       "      <td>0.64</td>\n",
       "      <td>0.08</td>\n",
       "      <td>0.3</td>\n",
       "      <td>0.29</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>sum</th>\n",
       "      <td>200.71</td>\n",
       "      <td>138.05</td>\n",
       "      <td>137.33</td>\n",
       "      <td>114.92</td>\n",
       "      <td>104.28</td>\n",
       "      <td>99.15</td>\n",
       "      <td>74.49</td>\n",
       "      <td>72.35</td>\n",
       "      <td>59.31</td>\n",
       "      <td>54.48</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   0       1       2       3       4      5      6      7  \\\n",
       "input_tokens       ▁    ▁der     ▁in    ▁von      ▁/   ▁und     ▁(     ▁)   \n",
       "count           6066    1388     989     808     163   1171    246    246   \n",
       "mean            0.03     0.1    0.14    0.14    0.64   0.08    0.3   0.29   \n",
       "sum           200.71  138.05  137.33  114.92  104.28  99.15  74.49  72.35   \n",
       "\n",
       "                  8      9  \n",
       "input_tokens    ▁''     ▁A  \n",
       "count          2898    125  \n",
       "mean           0.02   0.44  \n",
       "sum           59.31  54.48  "
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(\n",
    "    df_tokens.groupby(\"input_tokens\")[[\"loss\"]]\n",
    "    .agg([\"count\", \"mean\", \"sum\"])\n",
    "    .droplevel(level=0, axis=1)  # Get rid of multi-level columns\n",
    "    .sort_values(by=\"sum\", ascending=False)\n",
    "    .reset_index()\n",
    "    .round(2)\n",
    "    .head(10)\n",
    "    .T\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>labels</th>\n",
       "      <td>B-ORG</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>B-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>O</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>2683</td>\n",
       "      <td>1462</td>\n",
       "      <td>3820</td>\n",
       "      <td>3172</td>\n",
       "      <td>2893</td>\n",
       "      <td>4139</td>\n",
       "      <td>43648</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.66</td>\n",
       "      <td>0.64</td>\n",
       "      <td>0.48</td>\n",
       "      <td>0.35</td>\n",
       "      <td>0.26</td>\n",
       "      <td>0.18</td>\n",
       "      <td>0.03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>sum</th>\n",
       "      <td>1769.47</td>\n",
       "      <td>930.94</td>\n",
       "      <td>1850.39</td>\n",
       "      <td>1111.03</td>\n",
       "      <td>760.56</td>\n",
       "      <td>750.91</td>\n",
       "      <td>1354.46</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              0       1        2        3       4       5        6\n",
       "labels    B-ORG   I-LOC    I-ORG    B-LOC   B-PER   I-PER        O\n",
       "count      2683    1462     3820     3172    2893    4139    43648\n",
       "mean       0.66    0.64     0.48     0.35    0.26    0.18     0.03\n",
       "sum     1769.47  930.94  1850.39  1111.03  760.56  750.91  1354.46"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(\n",
    "    df_tokens.groupby(\"labels\")[[\"loss\"]] \n",
    "    .agg([\"count\", \"mean\", \"sum\"])\n",
    "    .droplevel(level=0, axis=1)\n",
    "    .sort_values(by=\"mean\", ascending=False)\n",
    "    .reset_index()\n",
    "    .round(2)\n",
    "    .T\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import ConfusionMatrixDisplay, confusion_matrix\n",
    "\n",
    "def plot_confusion_matrix(y_preds, y_true, labels):\n",
    "    cm = confusion_matrix(y_true, y_preds, normalize=\"true\")\n",
    "    fig, ax = plt.subplots(figsize=(6, 6))\n",
    "    disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=labels)\n",
    "    disp.plot(cmap=\"Blues\", values_format=\".2f\", ax=ax, colorbar=False)\n",
    "    plt.title(\"Normalized confusion matrix\")\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/pdf": "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\n",
      "image/svg+xml": "<?xml version=\"1.0\" encoding=\"utf-8\" standalone=\"no\"?>\n<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n  \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n<svg height=\"390.3305pt\" version=\"1.1\" viewBox=\"0 0 390.86 390.3305\" width=\"390.86pt\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n <metadata>\n  <rdf:RDF xmlns:cc=\"http://creativecommons.org/ns#\" xmlns:dc=\"http://purl.org/dc/elements/1.1/\" xmlns:rdf=\"http://www.w3.org/1999/02/22-rdf-syntax-ns#\">\n   <cc:Work>\n    <dc:type rdf:resource=\"http://purl.org/dc/dcmitype/StillImage\"/>\n    <dc:date>2021-12-07T12:19:00.917718</dc:date>\n    <dc:format>image/svg+xml</dc:format>\n    <dc:creator>\n     <cc:Agent>\n      <dc:title>Matplotlib v3.4.3, https://matplotlib.org/</dc:title>\n     </cc:Agent>\n    </dc:creator>\n   </cc:Work>\n  </rdf:RDF>\n </metadata>\n <defs>\n  <style type=\"text/css\">*{stroke-linecap:butt;stroke-linejoin:round;}</style>\n </defs>\n <g id=\"figure_1\">\n  <g id=\"patch_1\">\n   <path d=\"M 0 390.3305 \nL 390.86 390.3305 \nL 390.86 0 \nL 0 0 \nz\n\" style=\"fill:none;\"/>\n  </g>\n  <g id=\"axes_1\">\n   <g id=\"patch_2\">\n    <path d=\"M 57.5 349.9305 \nL 383.66 349.9305 \nL 383.66 23.7705 \nL 57.5 23.7705 \nz\n\" style=\"fill:#ffffff;\"/>\n   </g>\n   <g clip-path=\"url(#pb35a488033)\">\n    <image height=\"327\" id=\"image5cb95b4217\" transform=\"scale(1 -1)translate(0 -327)\" width=\"327\" x=\"57.5\" xlink:href=\"data:image/png;base64,\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\" y=\"-22.9305\"/>\n   </g>\n   <g id=\"matplotlib.axis_1\">\n    <g id=\"xtick_1\">\n     <g id=\"line2d_1\">\n      <defs>\n       <path d=\"M 0 0 \nL 0 3.5 \n\" id=\"m2df1bab93a\" style=\"stroke:#000000;stroke-width:0.8;\"/>\n      </defs>\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"80.797143\" xlink:href=\"#m2df1bab93a\" y=\"349.9305\"/>\n      </g>\n     </g>\n     <g id=\"text_1\">\n      <!-- O -->\n      <g transform=\"translate(77.38933 365.73925)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 1818 -64 \nC 2886 -64 3398 736 3398 2054 \nL 3398 2374 \nC 3398 3814 2842 4499 1818 4499 \nC 768 4499 237 3725 237 2362 \nL 237 2067 \nC 237 576 819 -64 1818 -64 \nz\nM 1824 384 \nC 1056 384 806 1037 806 2074 \nL 806 2477 \nC 806 3398 1082 4051 1805 4051 \nC 2573 4051 2829 3418 2829 2374 \nL 2829 1965 \nC 2829 998 2534 384 1824 384 \nz\n\" id=\"GuardianSansCond-Regular-4f\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_2\">\n     <g id=\"line2d_2\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"127.391429\" xlink:href=\"#m2df1bab93a\" y=\"349.9305\"/>\n      </g>\n     </g>\n     <g id=\"text_2\">\n      <!-- B-PER -->\n      <g transform=\"translate(114.382679 365.73925)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 416 0 \nL 1562 0 \nC 2528 0 2944 493 2944 1254 \nC 2944 1965 2618 2253 2112 2323 \nL 2112 2349 \nC 2605 2483 2790 2842 2790 3334 \nC 2790 4026 2464 4435 1517 4435 \nL 416 4435 \nL 416 0 \nz\nM 954 454 \nL 954 2074 \nL 1517 2074 \nC 2086 2074 2381 1875 2381 1293 \nC 2381 736 2157 454 1517 454 \nL 954 454 \nz\nM 954 2515 \nL 954 3974 \nL 1453 3974 \nC 1978 3974 2240 3814 2240 3264 \nC 2240 2707 1920 2515 1402 2515 \nL 954 2515 \nz\n\" id=\"GuardianSansCond-Regular-42\" transform=\"scale(0.015625)\"/>\n        <path d=\"M 333 1536 \nL 1408 1536 \nL 1408 1990 \nL 333 1990 \nL 333 1536 \nz\n\" id=\"GuardianSansCond-Regular-2d\" transform=\"scale(0.015625)\"/>\n        <path d=\"M 416 0 \nL 954 0 \nL 954 1766 \nL 1555 1766 \nC 2362 1766 2854 2246 2854 3117 \nC 2854 4026 2413 4435 1536 4435 \nL 416 4435 \nL 416 0 \nz\nM 954 2221 \nL 954 3974 \nL 1478 3974 \nC 2061 3974 2285 3750 2285 3098 \nC 2285 2496 2054 2221 1472 2221 \nL 954 2221 \nz\n\" id=\"GuardianSansCond-Regular-50\" transform=\"scale(0.015625)\"/>\n        <path d=\"M 416 0 \nL 2650 0 \nL 2650 461 \nL 954 461 \nL 954 2067 \nL 2266 2067 \nL 2266 2522 \nL 954 2522 \nL 954 3974 \nL 2630 3974 \nL 2630 4435 \nL 416 4435 \nL 416 0 \nz\n\" id=\"GuardianSansCond-Regular-45\" transform=\"scale(0.015625)\"/>\n        <path d=\"M 416 0 \nL 954 0 \nL 954 1888 \nL 1574 1888 \nL 2483 0 \nL 3066 0 \nL 2093 1990 \nL 2093 2022 \nC 2528 2182 2810 2502 2810 3168 \nC 2810 4058 2400 4435 1485 4435 \nL 416 4435 \nL 416 0 \nz\nM 954 2330 \nL 954 3974 \nL 1446 3974 \nC 2035 3974 2240 3744 2240 3149 \nC 2240 2586 2016 2330 1472 2330 \nL 954 2330 \nz\n\" id=\"GuardianSansCond-Regular-52\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-42\"/>\n       <use x=\"48.999985\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"76.199982\" xlink:href=\"#GuardianSansCond-Regular-50\"/>\n       <use x=\"123.199966\" xlink:href=\"#GuardianSansCond-Regular-45\"/>\n       <use x=\"168.09996\" xlink:href=\"#GuardianSansCond-Regular-52\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_3\">\n     <g id=\"line2d_3\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"173.985714\" xlink:href=\"#m2df1bab93a\" y=\"349.9305\"/>\n      </g>\n     </g>\n     <g id=\"text_3\">\n      <!-- I-PER -->\n      <g transform=\"translate(162.614777 365.73925)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 429 0 \nL 966 0 \nL 966 4435 \nL 429 4435 \nL 429 0 \nz\n\" id=\"GuardianSansCond-Regular-49\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-49\"/>\n       <use x=\"21.699997\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"48.899994\" xlink:href=\"#GuardianSansCond-Regular-50\"/>\n       <use x=\"95.899979\" xlink:href=\"#GuardianSansCond-Regular-45\"/>\n       <use x=\"140.799973\" xlink:href=\"#GuardianSansCond-Regular-52\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_4\">\n     <g id=\"line2d_4\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"220.58\" xlink:href=\"#m2df1bab93a\" y=\"349.9305\"/>\n      </g>\n     </g>\n     <g id=\"text_4\">\n      <!-- B-ORG -->\n      <g transform=\"translate(206.48 365.73925)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 1933 -64 \nC 2451 -64 2842 51 3098 166 \nL 3098 2234 \nL 1837 2234 \nL 1837 1779 \nL 2566 1779 \nL 2566 486 \nC 2432 435 2208 397 1978 397 \nC 1056 397 806 1011 806 2099 \nL 806 2406 \nC 806 3398 1088 4038 1990 4038 \nC 2438 4038 2714 3930 2944 3821 \nL 2944 4275 \nC 2758 4390 2477 4499 2022 4499 \nC 909 4499 237 3853 237 2342 \nL 237 2099 \nC 237 493 864 -64 1933 -64 \nz\n\" id=\"GuardianSansCond-Regular-47\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-42\"/>\n       <use x=\"48.999985\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"76.199982\" xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n       <use x=\"132.999969\" xlink:href=\"#GuardianSansCond-Regular-52\"/>\n       <use x=\"181.699966\" xlink:href=\"#GuardianSansCond-Regular-47\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_5\">\n     <g id=\"line2d_5\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"267.174286\" xlink:href=\"#m2df1bab93a\" y=\"349.9305\"/>\n      </g>\n     </g>\n     <g id=\"text_5\">\n      <!-- I-ORG -->\n      <g transform=\"translate(254.712098 365.73925)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-49\"/>\n       <use x=\"21.699997\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"48.899994\" xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n       <use x=\"105.699982\" xlink:href=\"#GuardianSansCond-Regular-52\"/>\n       <use x=\"154.399979\" xlink:href=\"#GuardianSansCond-Regular-47\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_6\">\n     <g id=\"line2d_6\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"313.768571\" xlink:href=\"#m2df1bab93a\" y=\"349.9305\"/>\n      </g>\n     </g>\n     <g id=\"text_6\">\n      <!-- B-LOC -->\n      <g transform=\"translate(300.436384 365.73925)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 416 0 \nL 2541 0 \nL 2541 461 \nL 954 461 \nL 954 4435 \nL 416 4435 \nL 416 0 \nz\n\" id=\"GuardianSansCond-Regular-4c\" transform=\"scale(0.015625)\"/>\n        <path d=\"M 1882 -64 \nC 2304 -64 2637 38 2874 186 \nL 2874 640 \nC 2643 518 2342 397 1946 397 \nC 1114 397 806 986 806 2042 \nL 806 2445 \nC 806 3648 1267 4038 1920 4038 \nC 2278 4038 2541 3955 2822 3840 \nL 2822 4288 \nC 2618 4410 2336 4499 1952 4499 \nC 1005 4499 237 3994 237 2387 \nL 237 2042 \nC 237 563 851 -64 1882 -64 \nz\n\" id=\"GuardianSansCond-Regular-43\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-42\"/>\n       <use x=\"48.999985\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"76.199982\" xlink:href=\"#GuardianSansCond-Regular-4c\"/>\n       <use x=\"117.999969\" xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n       <use x=\"174.799957\" xlink:href=\"#GuardianSansCond-Regular-43\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_7\">\n     <g id=\"line2d_7\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"360.362857\" xlink:href=\"#m2df1bab93a\" y=\"349.9305\"/>\n      </g>\n     </g>\n     <g id=\"text_7\">\n      <!-- I-LOC -->\n      <g transform=\"translate(348.668482 365.73925)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-49\"/>\n       <use x=\"21.699997\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"48.899994\" xlink:href=\"#GuardianSansCond-Regular-4c\"/>\n       <use x=\"90.699982\" xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n       <use x=\"147.499969\" xlink:href=\"#GuardianSansCond-Regular-43\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"text_8\">\n     <!-- Predicted label -->\n     <g transform=\"translate(189.075313 380.83925)scale(0.12 -0.12)\">\n      <defs>\n       <path d=\"M 346 0 \nL 870 0 \nL 870 2605 \nC 1062 2752 1306 2842 1587 2842 \nC 1638 2842 1728 2835 1786 2829 \nL 1786 3302 \nC 1747 3315 1677 3328 1606 3328 \nC 1306 3328 1030 3142 870 2938 \nL 845 2938 \nL 845 3283 \nL 346 3283 \nL 346 0 \nz\n\" id=\"GuardianSansCond-Regular-72\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1555 -51 \nC 1984 -51 2317 102 2470 243 \nL 2470 627 \nC 2240 499 1965 397 1613 397 \nC 1037 397 742 698 723 1466 \nL 2554 1466 \nL 2554 1798 \nC 2554 2893 2176 3334 1402 3334 \nC 659 3334 186 2810 186 1722 \nL 186 1549 \nC 186 499 634 -51 1555 -51 \nz\nM 1382 2906 \nC 1830 2906 2010 2611 2010 1894 \nL 723 1894 \nC 736 2630 992 2906 1382 2906 \nz\n\" id=\"GuardianSansCond-Regular-65\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1357 -51 \nC 1690 -51 1965 77 2125 243 \nL 2150 243 \nL 2195 0 \nL 2637 0 \nL 2637 4698 \nL 2118 4698 \nL 2118 3174 \nC 1978 3264 1734 3334 1466 3334 \nC 634 3334 198 2739 198 1670 \nL 198 1523 \nC 198 339 723 -51 1357 -51 \nz\nM 1504 378 \nC 992 378 742 749 742 1542 \nL 742 1728 \nC 742 2541 979 2893 1536 2893 \nC 1779 2893 2003 2835 2118 2758 \nL 2118 595 \nC 1984 493 1773 378 1504 378 \nz\n\" id=\"GuardianSansCond-Regular-64\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 358 0 \nL 883 0 \nL 883 3283 \nL 358 3283 \nL 358 0 \nz\nM 352 3910 \nL 883 3910 \nL 883 4474 \nL 352 4474 \nL 352 3910 \nz\n\" id=\"GuardianSansCond-Regular-69\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1498 -51 \nC 1869 -51 2125 51 2298 173 \nL 2298 595 \nC 2163 518 1907 390 1555 390 \nC 1005 390 723 704 723 1542 \nL 723 1798 \nC 723 2464 941 2893 1510 2893 \nC 1856 2893 2054 2803 2259 2714 \nL 2259 3149 \nC 2093 3258 1875 3334 1542 3334 \nC 768 3334 186 2886 186 1741 \nL 186 1536 \nC 186 416 685 -51 1498 -51 \nz\n\" id=\"GuardianSansCond-Regular-63\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1222 -38 \nC 1466 -38 1619 26 1722 90 \nL 1722 442 \nC 1632 416 1517 390 1382 390 \nC 1107 390 1005 480 1005 781 \nL 1005 2848 \nL 1683 2848 \nL 1683 3283 \nL 1005 3283 \nL 1005 4051 \nL 486 4051 \nL 486 3283 \nL 96 3283 \nL 96 2848 \nL 486 2848 \nL 486 710 \nC 486 160 774 -38 1222 -38 \nz\n\" id=\"GuardianSansCond-Regular-74\" transform=\"scale(0.015625)\"/>\n       <path id=\"GuardianSansCond-Regular-20\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 358 0 \nL 877 0 \nL 877 4698 \nL 358 4698 \nL 358 0 \nz\n\" id=\"GuardianSansCond-Regular-6c\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1082 -51 \nC 1395 -51 1638 77 1779 230 \nL 1811 230 \nL 1862 0 \nL 2291 0 \nL 2291 2413 \nC 2291 3027 1958 3334 1293 3334 \nC 915 3334 634 3258 410 3155 \nL 410 2771 \nC 576 2829 806 2886 1094 2886 \nC 1600 2886 1766 2739 1766 2278 \nL 1766 1958 \nL 1088 1856 \nC 480 1766 160 1466 160 883 \nC 160 250 525 -51 1082 -51 \nz\nM 1222 358 \nC 877 358 704 544 704 915 \nC 704 1274 877 1440 1242 1485 \nL 1766 1549 \nL 1766 525 \nC 1658 435 1459 358 1222 358 \nz\n\" id=\"GuardianSansCond-Regular-61\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1382 -64 \nC 2317 -64 2784 531 2784 1619 \nL 2784 1843 \nC 2784 2995 2285 3334 1702 3334 \nC 1338 3334 1075 3219 896 3104 \nL 870 3104 \nL 870 4698 \nL 346 4698 \nL 346 102 \nC 557 19 1005 -64 1382 -64 \nz\nM 1408 378 \nC 1165 378 979 416 870 454 \nL 870 2733 \nC 992 2790 1216 2893 1491 2893 \nC 2003 2893 2246 2650 2246 1818 \nL 2246 1549 \nC 2246 627 1875 378 1408 378 \nz\n\" id=\"GuardianSansCond-Regular-62\" transform=\"scale(0.015625)\"/>\n      </defs>\n      <use xlink:href=\"#GuardianSansCond-Regular-50\"/>\n      <use x=\"46.999985\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n      <use x=\"75.799973\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"118.499969\" xlink:href=\"#GuardianSansCond-Regular-64\"/>\n      <use x=\"165.09996\" xlink:href=\"#GuardianSansCond-Regular-69\"/>\n      <use x=\"184.399948\" xlink:href=\"#GuardianSansCond-Regular-63\"/>\n      <use x=\"222.199936\" xlink:href=\"#GuardianSansCond-Regular-74\"/>\n      <use x=\"251.199921\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"293.899918\" xlink:href=\"#GuardianSansCond-Regular-64\"/>\n      <use x=\"340.499908\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"356.299896\" xlink:href=\"#GuardianSansCond-Regular-6c\"/>\n      <use x=\"375.599884\" xlink:href=\"#GuardianSansCond-Regular-61\"/>\n      <use x=\"416.499878\" xlink:href=\"#GuardianSansCond-Regular-62\"/>\n      <use x=\"463.099869\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"505.799866\" xlink:href=\"#GuardianSansCond-Regular-6c\"/>\n     </g>\n    </g>\n   </g>\n   <g id=\"matplotlib.axis_2\">\n    <g id=\"ytick_1\">\n     <g id=\"line2d_8\">\n      <defs>\n       <path d=\"M 0 0 \nL -3.5 0 \n\" id=\"m588538198e\" style=\"stroke:#000000;stroke-width:0.8;\"/>\n      </defs>\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"57.5\" xlink:href=\"#m588538198e\" y=\"47.067643\"/>\n      </g>\n     </g>\n     <g id=\"text_9\">\n      <!-- O -->\n      <g transform=\"translate(43.684375 51.472018)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_2\">\n     <g id=\"line2d_9\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"57.5\" xlink:href=\"#m588538198e\" y=\"93.661929\"/>\n      </g>\n     </g>\n     <g id=\"text_10\">\n      <!-- B-PER -->\n      <g transform=\"translate(24.4825 98.066304)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-42\"/>\n       <use x=\"48.999985\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"76.199982\" xlink:href=\"#GuardianSansCond-Regular-50\"/>\n       <use x=\"123.199966\" xlink:href=\"#GuardianSansCond-Regular-45\"/>\n       <use x=\"168.09996\" xlink:href=\"#GuardianSansCond-Regular-52\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_3\">\n     <g id=\"line2d_10\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"57.5\" xlink:href=\"#m588538198e\" y=\"140.256214\"/>\n      </g>\n     </g>\n     <g id=\"text_11\">\n      <!-- I-PER -->\n      <g transform=\"translate(27.758125 144.660589)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-49\"/>\n       <use x=\"21.699997\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"48.899994\" xlink:href=\"#GuardianSansCond-Regular-50\"/>\n       <use x=\"95.899979\" xlink:href=\"#GuardianSansCond-Regular-45\"/>\n       <use x=\"140.799973\" xlink:href=\"#GuardianSansCond-Regular-52\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_4\">\n     <g id=\"line2d_11\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"57.5\" xlink:href=\"#m588538198e\" y=\"186.8505\"/>\n      </g>\n     </g>\n     <g id=\"text_12\">\n      <!-- B-ORG -->\n      <g transform=\"translate(22.3 191.254875)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-42\"/>\n       <use x=\"48.999985\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"76.199982\" xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n       <use x=\"132.999969\" xlink:href=\"#GuardianSansCond-Regular-52\"/>\n       <use x=\"181.699966\" xlink:href=\"#GuardianSansCond-Regular-47\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_5\">\n     <g id=\"line2d_12\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"57.5\" xlink:href=\"#m588538198e\" y=\"233.444786\"/>\n      </g>\n     </g>\n     <g id=\"text_13\">\n      <!-- I-ORG -->\n      <g transform=\"translate(25.575625 237.849161)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-49\"/>\n       <use x=\"21.699997\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"48.899994\" xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n       <use x=\"105.699982\" xlink:href=\"#GuardianSansCond-Regular-52\"/>\n       <use x=\"154.399979\" xlink:href=\"#GuardianSansCond-Regular-47\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_6\">\n     <g id=\"line2d_13\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"57.5\" xlink:href=\"#m588538198e\" y=\"280.039071\"/>\n      </g>\n     </g>\n     <g id=\"text_14\">\n      <!-- B-LOC -->\n      <g transform=\"translate(23.835625 284.443446)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-42\"/>\n       <use x=\"48.999985\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"76.199982\" xlink:href=\"#GuardianSansCond-Regular-4c\"/>\n       <use x=\"117.999969\" xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n       <use x=\"174.799957\" xlink:href=\"#GuardianSansCond-Regular-43\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_7\">\n     <g id=\"line2d_14\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"57.5\" xlink:href=\"#m588538198e\" y=\"326.633357\"/>\n      </g>\n     </g>\n     <g id=\"text_15\">\n      <!-- I-LOC -->\n      <g transform=\"translate(27.11125 331.037732)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-49\"/>\n       <use x=\"21.699997\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n       <use x=\"48.899994\" xlink:href=\"#GuardianSansCond-Regular-4c\"/>\n       <use x=\"90.699982\" xlink:href=\"#GuardianSansCond-Regular-4f\"/>\n       <use x=\"147.499969\" xlink:href=\"#GuardianSansCond-Regular-43\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"text_16\">\n     <!-- True label -->\n     <g transform=\"translate(16.00875 207.616125)rotate(-90)scale(0.12 -0.12)\">\n      <defs>\n       <path d=\"M 1152 0 \nL 1690 0 \nL 1690 3974 \nL 2746 3974 \nL 2746 4435 \nL 90 4435 \nL 90 3974 \nL 1152 3974 \nL 1152 0 \nz\n\" id=\"GuardianSansCond-Regular-54\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1158 -51 \nC 1523 -51 1824 109 2042 262 \nL 2067 262 \nL 2118 0 \nL 2566 0 \nL 2566 3283 \nL 2042 3283 \nL 2042 602 \nC 1888 512 1626 397 1350 397 \nC 947 397 858 557 858 838 \nL 858 3283 \nL 333 3283 \nL 333 826 \nC 333 301 576 -51 1158 -51 \nz\n\" id=\"GuardianSansCond-Regular-75\" transform=\"scale(0.015625)\"/>\n      </defs>\n      <use xlink:href=\"#GuardianSansCond-Regular-54\"/>\n      <use x=\"44.499985\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n      <use x=\"73.299973\" xlink:href=\"#GuardianSansCond-Regular-75\"/>\n      <use x=\"118.799957\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"161.499954\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"177.299942\" xlink:href=\"#GuardianSansCond-Regular-6c\"/>\n      <use x=\"196.59993\" xlink:href=\"#GuardianSansCond-Regular-61\"/>\n      <use x=\"237.499924\" xlink:href=\"#GuardianSansCond-Regular-62\"/>\n      <use x=\"284.099915\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"326.799911\" xlink:href=\"#GuardianSansCond-Regular-6c\"/>\n     </g>\n    </g>\n   </g>\n   <g id=\"patch_3\">\n    <path d=\"M 57.5 349.9305 \nL 57.5 23.7705 \n\" style=\"fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;\"/>\n   </g>\n   <g id=\"patch_4\">\n    <path d=\"M 383.66 349.9305 \nL 383.66 23.7705 \n\" style=\"fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;\"/>\n   </g>\n   <g id=\"patch_5\">\n    <path d=\"M 57.5 349.9305 \nL 383.66 349.9305 \n\" style=\"fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;\"/>\n   </g>\n   <g id=\"patch_6\">\n    <path d=\"M 57.5 23.7705 \nL 383.66 23.7705 \n\" style=\"fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;\"/>\n   </g>\n   <g id=\"text_17\">\n    <!-- 0.88 -->\n    <g style=\"fill:#f7fbff;\" transform=\"translate(70.362768 50.326393)scale(0.12 -0.12)\">\n     <defs>\n      <path d=\"M 1722 -70 \nC 2656 -70 3155 646 3155 2061 \nL 3155 2406 \nC 3155 3891 2598 4499 1715 4499 \nC 819 4499 269 3834 269 2368 \nL 269 2042 \nC 269 499 845 -70 1722 -70 \nz\nM 1728 384 \nC 1069 384 845 979 845 2016 \nL 845 2490 \nC 845 3494 1088 4051 1709 4051 \nC 2336 4051 2592 3514 2592 2458 \nL 2592 1978 \nC 2592 934 2374 384 1728 384 \nz\n\" id=\"GuardianSansCond-Regular-30\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 397 0 \nL 986 0 \nL 986 621 \nL 397 621 \nL 397 0 \nz\n\" id=\"GuardianSansCond-Regular-2e\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 1581 -64 \nC 2445 -64 2931 467 2931 1139 \nC 2931 1760 2592 2074 2106 2330 \nC 2490 2541 2822 2899 2822 3424 \nC 2822 4077 2394 4499 1606 4499 \nC 819 4499 384 4013 384 3398 \nC 384 2848 678 2509 1069 2285 \nC 602 2054 230 1683 230 1082 \nC 230 384 710 -64 1581 -64 \nz\nM 1587 390 \nC 1082 390 768 678 768 1126 \nC 768 1581 1037 1894 1414 2080 \nC 2029 1792 2387 1574 2387 1082 \nC 2387 614 2054 390 1587 390 \nz\nM 1754 2522 \nC 1210 2778 928 3021 928 3430 \nC 928 3821 1171 4051 1606 4051 \nC 2067 4051 2285 3782 2285 3398 \nC 2285 3008 2099 2765 1754 2522 \nz\n\" id=\"GuardianSansCond-Regular-38\" transform=\"scale(0.015625)\"/>\n     </defs>\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-38\"/>\n     <use x=\"124.499969\" xlink:href=\"#GuardianSansCond-Regular-38\"/>\n    </g>\n   </g>\n   <g id=\"text_18\">\n    <!-- 0.06 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(116.741429 50.326393)scale(0.12 -0.12)\">\n     <defs>\n      <path d=\"M 1632 384 \nC 1082 384 845 915 845 1837 \nL 845 2054 \nC 992 2150 1254 2278 1594 2278 \nC 2086 2278 2323 1984 2323 1357 \nC 2323 742 2086 384 1632 384 \nz\nM 1632 -70 \nC 2438 -70 2886 544 2886 1408 \nC 2886 2234 2490 2746 1754 2746 \nC 1363 2746 1056 2611 845 2445 \nC 877 3853 1440 4038 2029 4038 \nC 2291 4038 2554 3974 2688 3930 \nL 2688 4352 \nC 2547 4442 2272 4499 1958 4499 \nC 1043 4499 275 3981 275 2208 \nL 275 1843 \nC 275 691 717 -70 1632 -70 \nz\n\" id=\"GuardianSansCond-Regular-36\" transform=\"scale(0.015625)\"/>\n     </defs>\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-36\"/>\n    </g>\n   </g>\n   <g id=\"text_19\">\n    <!-- 0.02 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(163.720089 50.326393)scale(0.12 -0.12)\">\n     <defs>\n      <path d=\"M 173 0 \nL 2470 0 \nL 2470 461 \nL 800 461 \nL 800 480 \nL 1574 1421 \nC 2067 2035 2419 2528 2419 3264 \nC 2419 4026 2029 4499 1222 4499 \nC 806 4499 499 4410 269 4282 \nL 269 3853 \nC 467 3949 730 4032 1056 4032 \nC 1619 4032 1850 3750 1850 3213 \nC 1850 2630 1542 2189 1152 1670 \nL 173 397 \nL 173 0 \nz\n\" id=\"GuardianSansCond-Regular-32\" transform=\"scale(0.015625)\"/>\n     </defs>\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-32\"/>\n    </g>\n   </g>\n   <g id=\"text_20\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(211.136563 50.326393)scale(0.12 -0.12)\">\n     <defs>\n      <path d=\"M 832 0 \nL 1363 0 \nL 1363 4474 \nL 1171 4474 \nL 141 4083 \nL 141 3776 \nL 832 3846 \nL 832 0 \nz\n\" id=\"GuardianSansCond-Regular-31\" transform=\"scale(0.015625)\"/>\n     </defs>\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_21\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(257.730848 50.326393)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_22\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(302.842946 50.326393)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_23\">\n    <!-- 0.02 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(350.097232 50.326393)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-32\"/>\n    </g>\n   </g>\n   <g id=\"text_24\">\n    <!-- 0.05 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(70.555893 96.920679)scale(0.12 -0.12)\">\n     <defs>\n      <path d=\"M 1094 -64 \nC 1971 -64 2445 512 2445 1363 \nC 2445 2298 1997 2707 1165 2707 \nC 1011 2707 890 2694 794 2675 \nL 877 3974 \nL 2342 3974 \nL 2342 4435 \nL 429 4435 \nL 294 2170 \nC 467 2208 710 2246 966 2246 \nC 1587 2246 1875 1990 1875 1363 \nC 1875 717 1587 397 954 397 \nC 627 397 365 474 154 582 \nL 154 147 \nC 352 26 672 -64 1094 -64 \nz\n\" id=\"GuardianSansCond-Regular-35\" transform=\"scale(0.015625)\"/>\n     </defs>\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-35\"/>\n    </g>\n   </g>\n   <g id=\"text_25\">\n    <!-- 0.85 -->\n    <g style=\"fill:#f7fbff;\" transform=\"translate(117.395804 96.920679)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-38\"/>\n     <use x=\"124.499969\" xlink:href=\"#GuardianSansCond-Regular-35\"/>\n    </g>\n   </g>\n   <g id=\"text_26\">\n    <!-- 0.03 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(163.737902 96.920679)scale(0.12 -0.12)\">\n     <defs>\n      <path d=\"M 1094 -64 \nC 1920 -64 2432 416 2432 1248 \nC 2432 1939 2112 2253 1606 2362 \nL 1606 2381 \nC 2042 2515 2310 2803 2310 3405 \nC 2310 4096 1882 4499 1158 4499 \nC 787 4499 486 4429 275 4320 \nL 275 3891 \nC 461 3968 678 4038 998 4038 \nC 1478 4038 1754 3866 1754 3302 \nC 1754 2797 1478 2547 1018 2547 \nL 717 2547 \nL 717 2086 \nL 1056 2086 \nC 1600 2086 1869 1837 1869 1267 \nC 1869 646 1517 403 947 403 \nC 646 403 358 493 147 608 \nL 147 166 \nC 358 32 640 -64 1094 -64 \nz\n\" id=\"GuardianSansCond-Regular-33\" transform=\"scale(0.015625)\"/>\n     </defs>\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-33\"/>\n    </g>\n   </g>\n   <g id=\"text_27\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(209.654375 96.920679)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_28\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(257.730848 96.920679)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_29\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(302.842946 96.920679)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_30\">\n    <!-- 0.06 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(349.712857 96.920679)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-36\"/>\n    </g>\n   </g>\n   <g id=\"text_31\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(71.353705 143.514964)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_32\">\n    <!-- 0.03 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(117.143616 143.514964)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-33\"/>\n    </g>\n   </g>\n   <g id=\"text_33\">\n    <!-- 0.94 -->\n    <g style=\"fill:#f7fbff;\" transform=\"translate(163.701339 143.514964)scale(0.12 -0.12)\">\n     <defs>\n      <path d=\"M 1171 -70 \nC 2176 -70 2854 448 2854 2221 \nL 2854 2586 \nC 2854 3795 2432 4499 1523 4499 \nC 691 4499 243 3917 243 3021 \nC 243 2202 653 1683 1389 1683 \nC 1766 1683 2074 1818 2285 1984 \nC 2266 608 1798 397 1107 397 \nC 851 397 563 467 429 518 \nL 429 96 \nC 563 0 858 -70 1171 -70 \nz\nM 1542 2150 \nC 1056 2150 806 2445 806 3072 \nC 806 3712 1043 4045 1523 4045 \nC 2048 4045 2285 3558 2285 2592 \nL 2285 2374 \nC 2138 2272 1882 2150 1542 2150 \nz\n\" id=\"GuardianSansCond-Regular-39\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 1805 0 \nL 2330 0 \nL 2330 1107 \nL 2957 1107 \nL 2957 1562 \nL 2330 1562 \nL 2330 4435 \nL 1869 4435 \nL 70 1504 \nL 70 1107 \nL 1805 1107 \nL 1805 0 \nz\nM 608 1562 \nL 608 1581 \nL 1779 3475 \nL 1805 3475 \nL 1805 1562 \nL 608 1562 \nz\n\" id=\"GuardianSansCond-Regular-34\" transform=\"scale(0.015625)\"/>\n     </defs>\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-39\"/>\n     <use x=\"123.999969\" xlink:href=\"#GuardianSansCond-Regular-34\"/>\n    </g>\n   </g>\n   <g id=\"text_34\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(209.654375 143.514964)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_35\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(256.248661 143.514964)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_36\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(304.325134 143.514964)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_37\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(350.91942 143.514964)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_38\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(71.353705 190.10925)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_39\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(116.465804 190.10925)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_40\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(163.060089 190.10925)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_41\">\n    <!-- 0.83 -->\n    <g style=\"fill:#f7fbff;\" transform=\"translate(210.577813 190.10925)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-38\"/>\n     <use x=\"124.499969\" xlink:href=\"#GuardianSansCond-Regular-33\"/>\n    </g>\n   </g>\n   <g id=\"text_42\">\n    <!-- 0.11 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(259.213036 190.10925)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n     <use x=\"103.899963\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_43\">\n    <!-- 0.02 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(303.502946 190.10925)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-32\"/>\n    </g>\n   </g>\n   <g id=\"text_44\">\n    <!-- 0.02 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(350.097232 190.10925)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-32\"/>\n    </g>\n   </g>\n   <g id=\"text_45\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(71.353705 236.703536)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_46\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(117.947991 236.703536)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_47\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(163.060089 236.703536)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_48\">\n    <!-- 0.03 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(210.332188 236.703536)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-33\"/>\n    </g>\n   </g>\n   <g id=\"text_49\">\n    <!-- 0.90 -->\n    <g style=\"fill:#f7fbff;\" transform=\"translate(256.524286 236.703536)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-39\"/>\n     <use x=\"123.999969\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_50\">\n    <!-- 0.03 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(303.520759 236.703536)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-33\"/>\n    </g>\n   </g>\n   <g id=\"text_51\">\n    <!-- 0.03 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(350.115045 236.703536)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-33\"/>\n    </g>\n   </g>\n   <g id=\"text_52\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(69.871518 283.297821)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_53\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(116.465804 283.297821)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_54\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(163.060089 283.297821)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_55\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(211.136563 283.297821)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_56\">\n    <!-- 0.03 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(256.926473 283.297821)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-33\"/>\n    </g>\n   </g>\n   <g id=\"text_57\">\n    <!-- 0.95 -->\n    <g style=\"fill:#f7fbff;\" transform=\"translate(303.802946 283.297821)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-39\"/>\n     <use x=\"123.999969\" xlink:href=\"#GuardianSansCond-Regular-35\"/>\n    </g>\n   </g>\n   <g id=\"text_58\">\n    <!-- 0.01 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(350.91942 283.297821)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n    </g>\n   </g>\n   <g id=\"text_59\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(69.871518 329.892107)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_60\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(116.465804 329.892107)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_61\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(163.060089 329.892107)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_62\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(209.654375 329.892107)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_63\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(256.248661 329.892107)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_64\">\n    <!-- 0.00 -->\n    <g style=\"fill:#08306b;\" transform=\"translate(302.842946 329.892107)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"128.59996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n    </g>\n   </g>\n   <g id=\"text_65\">\n    <!-- 0.99 -->\n    <g style=\"fill:#f7fbff;\" transform=\"translate(349.988482 329.892107)scale(0.12 -0.12)\">\n     <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n     <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n     <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-39\"/>\n     <use x=\"123.999969\" xlink:href=\"#GuardianSansCond-Regular-39\"/>\n    </g>\n   </g>\n   <g id=\"text_66\">\n    <!-- Normalized confusion matrix -->\n    <g transform=\"translate(147.604625 17.7705)scale(0.144 -0.144)\">\n     <defs>\n      <path d=\"M 416 0 \nL 909 0 \nL 909 2912 \nC 909 3424 877 3757 877 3757 \nL 896 3757 \nC 896 3757 1050 3322 1274 2822 \nL 2547 0 \nL 3155 0 \nL 3155 4435 \nL 2662 4435 \nL 2662 1568 \nC 2662 1114 2694 762 2694 762 \nL 2682 762 \nC 2682 762 2502 1254 2330 1626 \nL 1094 4435 \nL 416 4435 \nL 416 0 \nz\n\" id=\"GuardianSansCond-Regular-4e\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 1459 -51 \nC 2272 -51 2733 570 2733 1542 \nL 2733 1747 \nC 2733 2771 2291 3334 1459 3334 \nC 659 3334 186 2714 186 1741 \nL 186 1530 \nC 186 531 614 -51 1459 -51 \nz\nM 1466 378 \nC 992 378 723 762 723 1472 \nL 723 1818 \nC 723 2451 947 2906 1453 2906 \nC 1920 2906 2195 2560 2195 1805 \nL 2195 1466 \nC 2195 826 1958 378 1466 378 \nz\n\" id=\"GuardianSansCond-Regular-6f\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 346 0 \nL 870 0 \nL 870 2688 \nC 1050 2790 1267 2886 1523 2886 \nC 1856 2886 1971 2758 1971 2464 \nL 1971 0 \nL 2490 0 \nL 2490 2688 \nC 2682 2797 2912 2886 3136 2886 \nC 3488 2886 3590 2746 3590 2464 \nL 3590 0 \nL 4115 0 \nL 4115 2528 \nC 4115 3104 3821 3334 3347 3334 \nC 2995 3334 2688 3187 2419 2995 \nL 2394 2995 \nC 2266 3226 2061 3334 1722 3334 \nC 1395 3334 1088 3181 870 3021 \nL 845 3021 \nL 845 3283 \nL 346 3283 \nL 346 0 \nz\n\" id=\"GuardianSansCond-Regular-6d\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 179 0 \nL 2246 0 \nL 2246 435 \nL 794 435 \nL 794 454 \nL 2259 2918 \nL 2259 3283 \nL 256 3283 \nL 256 2854 \nL 1626 2854 \nL 1626 2835 \nL 179 384 \nL 179 0 \nz\n\" id=\"GuardianSansCond-Regular-7a\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 346 0 \nL 870 0 \nL 870 2688 \nC 1037 2778 1306 2886 1587 2886 \nC 1965 2886 2093 2765 2093 2458 \nL 2093 0 \nL 2611 0 \nL 2611 2528 \nC 2611 3085 2342 3334 1811 3334 \nC 1440 3334 1107 3194 870 3021 \nL 845 3021 \nL 845 3283 \nL 346 3283 \nL 346 0 \nz\n\" id=\"GuardianSansCond-Regular-6e\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 493 0 \nL 1018 0 \nL 1018 2848 \nL 1626 2848 \nL 1626 3283 \nL 1018 3283 \nL 1018 3712 \nC 1018 4115 1146 4307 1478 4307 \nC 1632 4307 1754 4282 1837 4262 \nL 1837 4640 \nC 1747 4685 1587 4742 1363 4742 \nC 826 4742 493 4429 493 3757 \nL 493 3283 \nL 96 3283 \nL 96 2848 \nL 493 2848 \nL 493 0 \nz\n\" id=\"GuardianSansCond-Regular-66\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 992 -51 \nC 1658 -51 2054 307 2054 947 \nC 2054 1485 1818 1709 1306 1888 \nL 1050 1978 \nC 768 2074 659 2195 659 2451 \nC 659 2771 864 2893 1222 2893 \nC 1517 2893 1760 2810 1914 2739 \nL 1914 3168 \nC 1773 3251 1568 3334 1197 3334 \nC 531 3334 160 2982 160 2406 \nC 160 1901 422 1658 826 1517 \nL 1082 1427 \nC 1408 1312 1536 1184 1536 896 \nC 1536 582 1370 390 934 390 \nC 582 390 358 480 154 563 \nL 154 141 \nC 358 19 640 -51 992 -51 \nz\n\" id=\"GuardianSansCond-Regular-73\" transform=\"scale(0.015625)\"/>\n      <path d=\"M 83 0 \nL 634 0 \nL 1056 813 \nC 1203 1069 1293 1248 1293 1248 \nL 1299 1248 \nC 1299 1248 1376 1114 1523 819 \nL 1939 0 \nL 2515 0 \nL 1613 1670 \nL 2451 3283 \nL 1920 3283 \nL 1549 2515 \nL 1350 2086 \nL 1338 2086 \nC 1338 2086 1274 2208 1120 2509 \nL 717 3283 \nL 128 3283 \nL 1011 1658 \nL 83 0 \nz\n\" id=\"GuardianSansCond-Regular-78\" transform=\"scale(0.015625)\"/>\n     </defs>\n     <use xlink:href=\"#GuardianSansCond-Regular-4e\"/>\n     <use x=\"55.799988\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n     <use x=\"101.399979\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n     <use x=\"130.199966\" xlink:href=\"#GuardianSansCond-Regular-6d\"/>\n     <use x=\"199.699951\" xlink:href=\"#GuardianSansCond-Regular-61\"/>\n     <use x=\"240.599945\" xlink:href=\"#GuardianSansCond-Regular-6c\"/>\n     <use x=\"259.899933\" xlink:href=\"#GuardianSansCond-Regular-69\"/>\n     <use x=\"279.199921\" xlink:href=\"#GuardianSansCond-Regular-7a\"/>\n     <use x=\"317.399918\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n     <use x=\"360.099915\" xlink:href=\"#GuardianSansCond-Regular-64\"/>\n     <use x=\"406.699905\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n     <use x=\"422.499893\" xlink:href=\"#GuardianSansCond-Regular-63\"/>\n     <use x=\"460.299881\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n     <use x=\"505.899872\" xlink:href=\"#GuardianSansCond-Regular-6e\"/>\n     <use x=\"551.999863\" xlink:href=\"#GuardianSansCond-Regular-66\"/>\n     <use x=\"578.69986\" xlink:href=\"#GuardianSansCond-Regular-75\"/>\n     <use x=\"624.199844\" xlink:href=\"#GuardianSansCond-Regular-73\"/>\n     <use x=\"658.699829\" xlink:href=\"#GuardianSansCond-Regular-69\"/>\n     <use x=\"677.999817\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n     <use x=\"723.599808\" xlink:href=\"#GuardianSansCond-Regular-6e\"/>\n     <use x=\"769.699799\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n     <use x=\"785.499786\" xlink:href=\"#GuardianSansCond-Regular-6d\"/>\n     <use x=\"854.999771\" xlink:href=\"#GuardianSansCond-Regular-61\"/>\n     <use x=\"895.899765\" xlink:href=\"#GuardianSansCond-Regular-74\"/>\n     <use x=\"924.89975\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n     <use x=\"953.699738\" xlink:href=\"#GuardianSansCond-Regular-69\"/>\n     <use x=\"972.999725\" xlink:href=\"#GuardianSansCond-Regular-78\"/>\n    </g>\n   </g>\n  </g>\n </g>\n <defs>\n  <clipPath id=\"pb35a488033\">\n   <rect height=\"326.16\" width=\"326.16\" x=\"57.5\" y=\"23.7705\"/>\n  </clipPath>\n </defs>\n</svg>\n",
      "text/plain": [
       "<Figure size 432x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(df_tokens[\"labels\"], df_tokens[\"predicted_label\"],\n",
    "                      tags.names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>10</th>\n",
       "      <th>11</th>\n",
       "      <th>12</th>\n",
       "      <th>13</th>\n",
       "      <th>14</th>\n",
       "      <th>15</th>\n",
       "      <th>16</th>\n",
       "      <th>17</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>tokens</th>\n",
       "      <td>▁''</td>\n",
       "      <td>8</td>\n",
       "      <td>.</td>\n",
       "      <td>▁Juli</td>\n",
       "      <td>▁''</td>\n",
       "      <td>▁:</td>\n",
       "      <td>▁Protest</td>\n",
       "      <td>camp</td>\n",
       "      <td>▁auf</td>\n",
       "      <td>▁dem</td>\n",
       "      <td>▁Gelände</td>\n",
       "      <td>▁der</td>\n",
       "      <td>▁Republika</td>\n",
       "      <td>n</td>\n",
       "      <td>ischen</td>\n",
       "      <td>▁Gar</td>\n",
       "      <td>de</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>labels</th>\n",
       "      <td>B-ORG</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>preds</th>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>O</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>losses</th>\n",
       "      <td>7.89</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>6.88</td>\n",
       "      <td>8.05</td>\n",
       "      <td>8.68</td>\n",
       "      <td>8.37</td>\n",
       "      <td>0.00</td>\n",
       "      <td>8.65</td>\n",
       "      <td>10.08</td>\n",
       "      <td>8.27</td>\n",
       "      <td>7.90</td>\n",
       "      <td>5.83</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           0     1     2      3      4      5         6     7      8      9   \\\n",
       "tokens    ▁''     8     .  ▁Juli    ▁''     ▁:  ▁Protest  camp   ▁auf   ▁dem   \n",
       "labels  B-ORG   IGN   IGN  I-ORG  I-ORG  I-ORG     I-ORG   IGN  I-ORG  I-ORG   \n",
       "preds       O     O     O      O      O      O         O     O      O      O   \n",
       "losses   7.89  0.00  0.00   6.88   8.05   8.68      8.37  0.00   8.65  10.08   \n",
       "\n",
       "              10     11          12     13      14     15     16    17  \n",
       "tokens  ▁Gelände   ▁der  ▁Republika      n  ischen   ▁Gar     de  </s>  \n",
       "labels     I-ORG  I-ORG       I-ORG    IGN     IGN  I-ORG    IGN   IGN  \n",
       "preds          O      O       B-ORG  I-ORG   I-ORG  I-ORG  I-ORG     O  \n",
       "losses      8.27   7.90        5.83   0.00    0.00   0.01   0.00  0.00  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>10</th>\n",
       "      <th>11</th>\n",
       "      <th>12</th>\n",
       "      <th>13</th>\n",
       "      <th>14</th>\n",
       "      <th>15</th>\n",
       "      <th>16</th>\n",
       "      <th>17</th>\n",
       "      <th>18</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>tokens</th>\n",
       "      <td>▁'</td>\n",
       "      <td>▁''</td>\n",
       "      <td>▁Τ</td>\n",
       "      <td>Κ</td>\n",
       "      <td>▁''</td>\n",
       "      <td>▁'</td>\n",
       "      <td>▁'</td>\n",
       "      <td>▁''</td>\n",
       "      <td>▁T</td>\n",
       "      <td>▁''</td>\n",
       "      <td>▁'</td>\n",
       "      <td>ri</td>\n",
       "      <td>▁''</td>\n",
       "      <td>▁'</td>\n",
       "      <td>k</td>\n",
       "      <td>▁''</td>\n",
       "      <td>▁'</td>\n",
       "      <td>ala</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>labels</th>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>IGN</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>preds</th>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>losses</th>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>3.59</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>10.17</td>\n",
       "      <td>9.22</td>\n",
       "      <td>7.83</td>\n",
       "      <td>7.16</td>\n",
       "      <td>7.23</td>\n",
       "      <td>0.00</td>\n",
       "      <td>7.45</td>\n",
       "      <td>7.75</td>\n",
       "      <td>0.00</td>\n",
       "      <td>7.66</td>\n",
       "      <td>7.78</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          0     1      2     3     4     5      6      7      8      9   \\\n",
       "tokens    ▁'   ▁''     ▁Τ     Κ   ▁''    ▁'     ▁'    ▁''     ▁T    ▁''   \n",
       "labels     O     O      O   IGN     O     O  B-LOC  I-LOC  I-LOC  I-LOC   \n",
       "preds      O     O  B-ORG     O     O     O      O      O  B-ORG      O   \n",
       "losses  0.00  0.00   3.59  0.00  0.00  0.00  10.17   9.22   7.83   7.16   \n",
       "\n",
       "           10    11     12     13    14     15     16    17    18  \n",
       "tokens     ▁'    ri    ▁''     ▁'     k    ▁''     ▁'   ala  </s>  \n",
       "labels  I-LOC   IGN  I-LOC  I-LOC   IGN  I-LOC  I-LOC   IGN   IGN  \n",
       "preds       O     O      O      O     O      O      O     O     O  \n",
       "losses   7.23  0.00   7.45   7.75  0.00   7.66   7.78  0.00  0.00  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>10</th>\n",
       "      <th>11</th>\n",
       "      <th>12</th>\n",
       "      <th>13</th>\n",
       "      <th>14</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>tokens</th>\n",
       "      <td>▁United</td>\n",
       "      <td>▁Nations</td>\n",
       "      <td>▁Multi</td>\n",
       "      <td>dimensional</td>\n",
       "      <td>▁Integra</td>\n",
       "      <td>ted</td>\n",
       "      <td>▁Stabil</td>\n",
       "      <td>ization</td>\n",
       "      <td>▁Mission</td>\n",
       "      <td>▁in</td>\n",
       "      <td>▁the</td>\n",
       "      <td>▁Central</td>\n",
       "      <td>▁African</td>\n",
       "      <td>▁Republic</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>labels</th>\n",
       "      <td>B-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>IGN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>preds</th>\n",
       "      <td>B-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>losses</th>\n",
       "      <td>6.46</td>\n",
       "      <td>5.59</td>\n",
       "      <td>5.51</td>\n",
       "      <td>0.00</td>\n",
       "      <td>5.11</td>\n",
       "      <td>0.00</td>\n",
       "      <td>4.91</td>\n",
       "      <td>0.00</td>\n",
       "      <td>5.18</td>\n",
       "      <td>4.31</td>\n",
       "      <td>4.77</td>\n",
       "      <td>5.32</td>\n",
       "      <td>5.10</td>\n",
       "      <td>4.87</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             0         1       2            3         4      5        6   \\\n",
       "tokens  ▁United  ▁Nations  ▁Multi  dimensional  ▁Integra    ted  ▁Stabil   \n",
       "labels    B-PER     I-PER   I-PER          IGN     I-PER    IGN    I-PER   \n",
       "preds     B-ORG     I-ORG   I-ORG        I-ORG     I-ORG  I-ORG    I-ORG   \n",
       "losses     6.46      5.59    5.51         0.00      5.11   0.00     4.91   \n",
       "\n",
       "             7         8      9      10        11        12         13     14  \n",
       "tokens  ization  ▁Mission    ▁in   ▁the  ▁Central  ▁African  ▁Republic   </s>  \n",
       "labels      IGN     I-PER  I-PER  I-PER     I-PER     I-PER      I-PER    IGN  \n",
       "preds     I-ORG     I-ORG  I-ORG  I-ORG     I-ORG     I-ORG      I-ORG  I-ORG  \n",
       "losses     0.00      5.18   4.31   4.77      5.32      5.10       4.87   0.00  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# hide_output\n",
    "def get_samples(df):\n",
    "    for _, row in df.iterrows():\n",
    "        labels, preds, tokens, losses = [], [], [], []\n",
    "        for i, mask in enumerate(row[\"attention_mask\"]):\n",
    "            if i not in {0, len(row[\"attention_mask\"])}:\n",
    "                labels.append(row[\"labels\"][i])\n",
    "                preds.append(row[\"predicted_label\"][i])\n",
    "                tokens.append(row[\"input_tokens\"][i])\n",
    "                losses.append(f\"{row['loss'][i]:.2f}\")\n",
    "        df_tmp = pd.DataFrame({\"tokens\": tokens, \"labels\": labels, \n",
    "                               \"preds\": preds, \"losses\": losses}).T\n",
    "        yield df_tmp\n",
    "\n",
    "df[\"total_loss\"] = df[\"loss\"].apply(sum)\n",
    "df_tmp = df.sort_values(by=\"total_loss\", ascending=False).head(3)\n",
    "\n",
    "for sample in get_samples(df_tmp):\n",
    "    display(sample)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>tokens</th>\n",
       "      <td>▁Ham</td>\n",
       "      <td>a</td>\n",
       "      <td>▁(</td>\n",
       "      <td>▁Unternehmen</td>\n",
       "      <td>▁)</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>labels</th>\n",
       "      <td>B-ORG</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>IGN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>preds</th>\n",
       "      <td>B-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "      <td>I-ORG</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>losses</th>\n",
       "      <td>0.01</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            0      1      2             3      4      5\n",
       "tokens   ▁Ham      a     ▁(  ▁Unternehmen     ▁)   </s>\n",
       "labels  B-ORG    IGN  I-ORG         I-ORG  I-ORG    IGN\n",
       "preds   B-ORG  I-ORG  I-ORG         I-ORG  I-ORG  I-ORG\n",
       "losses   0.01   0.00   0.01          0.01   0.01   0.00"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>tokens</th>\n",
       "      <td>▁Kesk</td>\n",
       "      <td>kül</td>\n",
       "      <td>a</td>\n",
       "      <td>▁(</td>\n",
       "      <td>▁Mart</td>\n",
       "      <td>na</td>\n",
       "      <td>▁)</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>labels</th>\n",
       "      <td>B-LOC</td>\n",
       "      <td>IGN</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>IGN</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>IGN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>preds</th>\n",
       "      <td>B-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>losses</th>\n",
       "      <td>0.02</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            0      1      2      3      4      5      6      7\n",
       "tokens  ▁Kesk    kül      a     ▁(  ▁Mart     na     ▁)   </s>\n",
       "labels  B-LOC    IGN    IGN  I-LOC  I-LOC    IGN  I-LOC    IGN\n",
       "preds   B-LOC  I-LOC  I-LOC  I-LOC  I-LOC  I-LOC  I-LOC  I-LOC\n",
       "losses   0.02   0.00   0.00   0.01   0.01   0.00   0.01   0.00"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# hide_output\n",
    "df_tmp = df.loc[df[\"input_tokens\"].apply(lambda x: u\"\\u2581(\" in x)].head(2)\n",
    "for sample in get_samples(df_tmp):\n",
    "    display(sample)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Cross-Lingual Transfer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_f1_score(trainer, dataset):\n",
    "    return trainer.predict(dataset).metrics[\"test_f1\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de] model on [de] dataset: 0.868\n"
     ]
    }
   ],
   "source": [
    "f1_scores = defaultdict(dict)\n",
    "f1_scores[\"de\"][\"de\"] = get_f1_score(trainer, panx_de_encoded[\"test\"])\n",
    "print(f\"F1-score of [de] model on [de] dataset: {f1_scores['de']['de']:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>9</th>\n",
       "      <th>10</th>\n",
       "      <th>11</th>\n",
       "      <th>12</th>\n",
       "      <th>13</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Tokens</th>\n",
       "      <td>&lt;s&gt;</td>\n",
       "      <td>▁Jeff</td>\n",
       "      <td>▁De</td>\n",
       "      <td>an</td>\n",
       "      <td>▁est</td>\n",
       "      <td>▁informatic</td>\n",
       "      <td>ien</td>\n",
       "      <td>▁chez</td>\n",
       "      <td>▁Google</td>\n",
       "      <td>▁en</td>\n",
       "      <td>▁Cali</td>\n",
       "      <td>for</td>\n",
       "      <td>nie</td>\n",
       "      <td>&lt;/s&gt;</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Tags</th>\n",
       "      <td>O</td>\n",
       "      <td>B-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>I-PER</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>O</td>\n",
       "      <td>B-ORG</td>\n",
       "      <td>O</td>\n",
       "      <td>B-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>I-LOC</td>\n",
       "      <td>O</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         0      1      2      3     4            5    6      7        8    9   \\\n",
       "Tokens  <s>  ▁Jeff    ▁De     an  ▁est  ▁informatic  ien  ▁chez  ▁Google  ▁en   \n",
       "Tags      O  B-PER  I-PER  I-PER     O            O    O      O    B-ORG    O   \n",
       "\n",
       "           10     11     12    13  \n",
       "Tokens  ▁Cali    for    nie  </s>  \n",
       "Tags    B-LOC  I-LOC  I-LOC     O  "
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "text_fr = \"Jeff Dean est informaticien chez Google en Californie\"\n",
    "tag_text(text_fr, tags, trainer.model, xlmr_tokenizer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def evaluate_lang_performance(lang, trainer):\n",
    "    panx_ds = encode_panx_dataset(panx_ch[lang])\n",
    "    return get_f1_score(trainer, panx_ds[\"test\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2bca3ec35cc44ad7b6bad47bd282206a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/5 [00:00<?, ?ba/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de] model on [fr] dataset: 0.714\n"
     ]
    }
   ],
   "source": [
    "# hide_output\n",
    "f1_scores[\"de\"][\"fr\"] = evaluate_lang_performance(\"fr\", trainer)\n",
    "print(f\"F1-score of [de] model on [fr] dataset: {f1_scores['de']['fr']:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de] model on [fr] dataset: 0.714\n"
     ]
    }
   ],
   "source": [
    "# hide_input\n",
    "print(f\"F1-score of [de] model on [fr] dataset: {f1_scores['de']['fr']:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de] model on [it] dataset: 0.692\n"
     ]
    }
   ],
   "source": [
    "# hide_output\n",
    "f1_scores[\"de\"][\"it\"] = evaluate_lang_performance(\"it\", trainer)\n",
    "print(f\"F1-score of [de] model on [it] dataset: {f1_scores['de']['it']:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de] model on [it] dataset: 0.692\n"
     ]
    }
   ],
   "source": [
    "# hide_input\n",
    "print(f\"F1-score of [de] model on [it] dataset: {f1_scores['de']['it']:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de] model on [en] dataset: 0.589\n"
     ]
    }
   ],
   "source": [
    "#hide_output\n",
    "f1_scores[\"de\"][\"en\"] = evaluate_lang_performance(\"en\", trainer)\n",
    "print(f\"F1-score of [de] model on [en] dataset: {f1_scores['de']['en']:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de] model on [en] dataset: 0.589\n"
     ]
    }
   ],
   "source": [
    "#hide_input\n",
    "print(f\"F1-score of [de] model on [en] dataset: {f1_scores['de']['en']:.3f}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### When Does Zero-Shot Transfer Make Sense?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train_on_subset(dataset, num_samples):\n",
    "    train_ds = dataset[\"train\"].shuffle(seed=42).select(range(num_samples))\n",
    "    valid_ds = dataset[\"validation\"]\n",
    "    test_ds = dataset[\"test\"]\n",
    "    training_args.logging_steps = len(train_ds) // batch_size\n",
    "    \n",
    "    trainer = Trainer(model_init=model_init, args=training_args,\n",
    "        data_collator=data_collator, compute_metrics=compute_metrics,\n",
    "        train_dataset=train_ds, eval_dataset=valid_ds, tokenizer=xlmr_tokenizer)\n",
    "    trainer.train()\n",
    "    if training_args.push_to_hub:\n",
    "        trainer.push_to_hub(commit_message=\"Training completed!\")\n",
    "    \n",
    "    f1_score = get_f1_score(trainer, test_ds)\n",
    "    return pd.DataFrame.from_dict(\n",
    "        {\"num_samples\": [len(train_ds)], \"f1_score\": [f1_score]})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d72903c6f51a464099f892ee122c2745",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/3 [00:00<?, ?ba/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# hide_output\n",
    "panx_fr_encoded = encode_panx_dataset(panx_ch[\"fr\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "training_args.push_to_hub = False\n",
    "metrics_df = train_on_subset(panx_fr_encoded, 250)\n",
    "metrics_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>num_samples</th>\n",
       "      <th>f1_score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>250</td>\n",
       "      <td>0.137329</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   num_samples  f1_score\n",
       "0          250  0.137329"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#hide_input\n",
    "# Hack needed to exclude the progress bars in the above cell\n",
    "metrics_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "for num_samples in [500, 1000, 2000, 4000]:\n",
    "    metrics_df = metrics_df.append(\n",
    "        train_on_subset(panx_fr_encoded, num_samples), ignore_index=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/pdf": "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\n",
      "image/svg+xml": "<?xml version=\"1.0\" encoding=\"utf-8\" standalone=\"no\"?>\n<!DOCTYPE svg PUBLIC \"-//W3C//DTD SVG 1.1//EN\"\n  \"http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd\">\n<svg height=\"269.55125pt\" version=\"1.1\" viewBox=\"0 0 386.73125 269.55125\" width=\"386.73125pt\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n <metadata>\n  <rdf:RDF xmlns:cc=\"http://creativecommons.org/ns#\" xmlns:dc=\"http://purl.org/dc/elements/1.1/\" xmlns:rdf=\"http://www.w3.org/1999/02/22-rdf-syntax-ns#\">\n   <cc:Work>\n    <dc:type rdf:resource=\"http://purl.org/dc/dcmitype/StillImage\"/>\n    <dc:date>2021-10-24T12:35:59.543256</dc:date>\n    <dc:format>image/svg+xml</dc:format>\n    <dc:creator>\n     <cc:Agent>\n      <dc:title>Matplotlib v3.4.3, https://matplotlib.org/</dc:title>\n     </cc:Agent>\n    </dc:creator>\n   </cc:Work>\n  </rdf:RDF>\n </metadata>\n <defs>\n  <style type=\"text/css\">*{stroke-linecap:butt;stroke-linejoin:round;}</style>\n </defs>\n <g id=\"figure_1\">\n  <g id=\"patch_1\">\n   <path d=\"M 0 269.55125 \nL 386.73125 269.55125 \nL 386.73125 0 \nL 0 0 \nz\n\" style=\"fill:none;\"/>\n  </g>\n  <g id=\"axes_1\">\n   <g id=\"patch_2\">\n    <path d=\"M 44.73125 229.044375 \nL 379.53125 229.044375 \nL 379.53125 11.604375 \nL 44.73125 11.604375 \nz\n\" style=\"fill:#ffffff;\"/>\n   </g>\n   <g id=\"matplotlib.axis_1\">\n    <g id=\"xtick_1\">\n     <g id=\"line2d_1\">\n      <defs>\n       <path d=\"M 0 0 \nL 0 3.5 \n\" id=\"m99b8793981\" style=\"stroke:#000000;stroke-width:0.8;\"/>\n      </defs>\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"80.240341\" xlink:href=\"#m99b8793981\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_1\">\n      <!-- 500 -->\n      <g transform=\"translate(71.294716 244.853125)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 1094 -64 \nC 1971 -64 2445 512 2445 1363 \nC 2445 2298 1997 2707 1165 2707 \nC 1011 2707 890 2694 794 2675 \nL 877 3974 \nL 2342 3974 \nL 2342 4435 \nL 429 4435 \nL 294 2170 \nC 467 2208 710 2246 966 2246 \nC 1587 2246 1875 1990 1875 1363 \nC 1875 717 1587 397 954 397 \nC 627 397 365 474 154 582 \nL 154 147 \nC 352 26 672 -64 1094 -64 \nz\n\" id=\"GuardianSansCond-Regular-35\" transform=\"scale(0.015625)\"/>\n        <path d=\"M 1722 -70 \nC 2656 -70 3155 646 3155 2061 \nL 3155 2406 \nC 3155 3891 2598 4499 1715 4499 \nC 819 4499 269 3834 269 2368 \nL 269 2042 \nC 269 499 845 -70 1722 -70 \nz\nM 1728 384 \nC 1069 384 845 979 845 2016 \nL 845 2490 \nC 845 3494 1088 4051 1709 4051 \nC 2336 4051 2592 3514 2592 2458 \nL 2592 1978 \nC 2592 934 2374 384 1728 384 \nz\n\" id=\"GuardianSansCond-Regular-30\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-35\"/>\n       <use x=\"42.099991\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"95.599976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_2\">\n     <g id=\"line2d_2\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"120.822159\" xlink:href=\"#m99b8793981\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_2\">\n      <!-- 1000 -->\n      <g transform=\"translate(109.464347 244.853125)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 832 0 \nL 1363 0 \nL 1363 4474 \nL 1171 4474 \nL 141 4083 \nL 141 3776 \nL 832 3846 \nL 832 0 \nz\n\" id=\"GuardianSansCond-Regular-31\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-31\"/>\n       <use x=\"28.799988\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"82.299973\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"135.799957\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_3\">\n     <g id=\"line2d_3\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"161.403977\" xlink:href=\"#m99b8793981\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_3\">\n      <!-- 1500 -->\n      <g transform=\"translate(150.73054 244.853125)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-31\"/>\n       <use x=\"28.799988\" xlink:href=\"#GuardianSansCond-Regular-35\"/>\n       <use x=\"70.899979\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"124.399963\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_4\">\n     <g id=\"line2d_4\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"201.985795\" xlink:href=\"#m99b8793981\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_4\">\n      <!-- 2000 -->\n      <g transform=\"translate(189.805795 244.853125)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 173 0 \nL 2470 0 \nL 2470 461 \nL 800 461 \nL 800 480 \nL 1574 1421 \nC 2067 2035 2419 2528 2419 3264 \nC 2419 4026 2029 4499 1222 4499 \nC 806 4499 499 4410 269 4282 \nL 269 3853 \nC 467 3949 730 4032 1056 4032 \nC 1619 4032 1850 3750 1850 3213 \nC 1850 2630 1542 2189 1152 1670 \nL 173 397 \nL 173 0 \nz\n\" id=\"GuardianSansCond-Regular-32\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-32\"/>\n       <use x=\"42.499985\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"95.999969\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"149.499954\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_5\">\n     <g id=\"line2d_5\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"242.567614\" xlink:href=\"#m99b8793981\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_5\">\n      <!-- 2500 -->\n      <g transform=\"translate(231.071989 244.853125)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-32\"/>\n       <use x=\"42.499985\" xlink:href=\"#GuardianSansCond-Regular-35\"/>\n       <use x=\"84.599976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"138.09996\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_6\">\n     <g id=\"line2d_6\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"283.149432\" xlink:href=\"#m99b8793981\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_6\">\n      <!-- 3000 -->\n      <g transform=\"translate(270.987244 244.853125)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 1094 -64 \nC 1920 -64 2432 416 2432 1248 \nC 2432 1939 2112 2253 1606 2362 \nL 1606 2381 \nC 2042 2515 2310 2803 2310 3405 \nC 2310 4096 1882 4499 1158 4499 \nC 787 4499 486 4429 275 4320 \nL 275 3891 \nC 461 3968 678 4038 998 4038 \nC 1478 4038 1754 3866 1754 3302 \nC 1754 2797 1478 2547 1018 2547 \nL 717 2547 \nL 717 2086 \nL 1056 2086 \nC 1600 2086 1869 1837 1869 1267 \nC 1869 646 1517 403 947 403 \nC 646 403 358 493 147 608 \nL 147 166 \nC 358 32 640 -64 1094 -64 \nz\n\" id=\"GuardianSansCond-Regular-33\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-33\"/>\n       <use x=\"42.199997\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"95.699982\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"149.199966\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_7\">\n     <g id=\"line2d_7\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"323.73125\" xlink:href=\"#m99b8793981\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_7\">\n      <!-- 3500 -->\n      <g transform=\"translate(312.253438 244.853125)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-33\"/>\n       <use x=\"42.199997\" xlink:href=\"#GuardianSansCond-Regular-35\"/>\n       <use x=\"84.299988\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"137.799973\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"xtick_8\">\n     <g id=\"line2d_8\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"364.313068\" xlink:href=\"#m99b8793981\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_8\">\n      <!-- 4000 -->\n      <g transform=\"translate(351.838693 244.853125)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 1805 0 \nL 2330 0 \nL 2330 1107 \nL 2957 1107 \nL 2957 1562 \nL 2330 1562 \nL 2330 4435 \nL 1869 4435 \nL 70 1504 \nL 70 1107 \nL 1805 1107 \nL 1805 0 \nz\nM 608 1562 \nL 608 1581 \nL 1779 3475 \nL 1805 3475 \nL 1805 1562 \nL 608 1562 \nz\n\" id=\"GuardianSansCond-Regular-34\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-34\"/>\n       <use x=\"47.399994\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"100.899979\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"154.399963\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"text_9\">\n     <!-- Number of Training Samples -->\n     <g transform=\"translate(152.613125 260.035625)scale(0.12 -0.12)\">\n      <defs>\n       <path d=\"M 416 0 \nL 909 0 \nL 909 2912 \nC 909 3424 877 3757 877 3757 \nL 896 3757 \nC 896 3757 1050 3322 1274 2822 \nL 2547 0 \nL 3155 0 \nL 3155 4435 \nL 2662 4435 \nL 2662 1568 \nC 2662 1114 2694 762 2694 762 \nL 2682 762 \nC 2682 762 2502 1254 2330 1626 \nL 1094 4435 \nL 416 4435 \nL 416 0 \nz\n\" id=\"GuardianSansCond-Regular-4e\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1158 -51 \nC 1523 -51 1824 109 2042 262 \nL 2067 262 \nL 2118 0 \nL 2566 0 \nL 2566 3283 \nL 2042 3283 \nL 2042 602 \nC 1888 512 1626 397 1350 397 \nC 947 397 858 557 858 838 \nL 858 3283 \nL 333 3283 \nL 333 826 \nC 333 301 576 -51 1158 -51 \nz\n\" id=\"GuardianSansCond-Regular-75\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 346 0 \nL 870 0 \nL 870 2688 \nC 1050 2790 1267 2886 1523 2886 \nC 1856 2886 1971 2758 1971 2464 \nL 1971 0 \nL 2490 0 \nL 2490 2688 \nC 2682 2797 2912 2886 3136 2886 \nC 3488 2886 3590 2746 3590 2464 \nL 3590 0 \nL 4115 0 \nL 4115 2528 \nC 4115 3104 3821 3334 3347 3334 \nC 2995 3334 2688 3187 2419 2995 \nL 2394 2995 \nC 2266 3226 2061 3334 1722 3334 \nC 1395 3334 1088 3181 870 3021 \nL 845 3021 \nL 845 3283 \nL 346 3283 \nL 346 0 \nz\n\" id=\"GuardianSansCond-Regular-6d\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1382 -64 \nC 2317 -64 2784 531 2784 1619 \nL 2784 1843 \nC 2784 2995 2285 3334 1702 3334 \nC 1338 3334 1075 3219 896 3104 \nL 870 3104 \nL 870 4698 \nL 346 4698 \nL 346 102 \nC 557 19 1005 -64 1382 -64 \nz\nM 1408 378 \nC 1165 378 979 416 870 454 \nL 870 2733 \nC 992 2790 1216 2893 1491 2893 \nC 2003 2893 2246 2650 2246 1818 \nL 2246 1549 \nC 2246 627 1875 378 1408 378 \nz\n\" id=\"GuardianSansCond-Regular-62\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1555 -51 \nC 1984 -51 2317 102 2470 243 \nL 2470 627 \nC 2240 499 1965 397 1613 397 \nC 1037 397 742 698 723 1466 \nL 2554 1466 \nL 2554 1798 \nC 2554 2893 2176 3334 1402 3334 \nC 659 3334 186 2810 186 1722 \nL 186 1549 \nC 186 499 634 -51 1555 -51 \nz\nM 1382 2906 \nC 1830 2906 2010 2611 2010 1894 \nL 723 1894 \nC 736 2630 992 2906 1382 2906 \nz\n\" id=\"GuardianSansCond-Regular-65\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 346 0 \nL 870 0 \nL 870 2605 \nC 1062 2752 1306 2842 1587 2842 \nC 1638 2842 1728 2835 1786 2829 \nL 1786 3302 \nC 1747 3315 1677 3328 1606 3328 \nC 1306 3328 1030 3142 870 2938 \nL 845 2938 \nL 845 3283 \nL 346 3283 \nL 346 0 \nz\n\" id=\"GuardianSansCond-Regular-72\" transform=\"scale(0.015625)\"/>\n       <path id=\"GuardianSansCond-Regular-20\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1459 -51 \nC 2272 -51 2733 570 2733 1542 \nL 2733 1747 \nC 2733 2771 2291 3334 1459 3334 \nC 659 3334 186 2714 186 1741 \nL 186 1530 \nC 186 531 614 -51 1459 -51 \nz\nM 1466 378 \nC 992 378 723 762 723 1472 \nL 723 1818 \nC 723 2451 947 2906 1453 2906 \nC 1920 2906 2195 2560 2195 1805 \nL 2195 1466 \nC 2195 826 1958 378 1466 378 \nz\n\" id=\"GuardianSansCond-Regular-6f\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 493 0 \nL 1018 0 \nL 1018 2848 \nL 1626 2848 \nL 1626 3283 \nL 1018 3283 \nL 1018 3712 \nC 1018 4115 1146 4307 1478 4307 \nC 1632 4307 1754 4282 1837 4262 \nL 1837 4640 \nC 1747 4685 1587 4742 1363 4742 \nC 826 4742 493 4429 493 3757 \nL 493 3283 \nL 96 3283 \nL 96 2848 \nL 493 2848 \nL 493 0 \nz\n\" id=\"GuardianSansCond-Regular-66\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1152 0 \nL 1690 0 \nL 1690 3974 \nL 2746 3974 \nL 2746 4435 \nL 90 4435 \nL 90 3974 \nL 1152 3974 \nL 1152 0 \nz\n\" id=\"GuardianSansCond-Regular-54\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1082 -51 \nC 1395 -51 1638 77 1779 230 \nL 1811 230 \nL 1862 0 \nL 2291 0 \nL 2291 2413 \nC 2291 3027 1958 3334 1293 3334 \nC 915 3334 634 3258 410 3155 \nL 410 2771 \nC 576 2829 806 2886 1094 2886 \nC 1600 2886 1766 2739 1766 2278 \nL 1766 1958 \nL 1088 1856 \nC 480 1766 160 1466 160 883 \nC 160 250 525 -51 1082 -51 \nz\nM 1222 358 \nC 877 358 704 544 704 915 \nC 704 1274 877 1440 1242 1485 \nL 1766 1549 \nL 1766 525 \nC 1658 435 1459 358 1222 358 \nz\n\" id=\"GuardianSansCond-Regular-61\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 358 0 \nL 883 0 \nL 883 3283 \nL 358 3283 \nL 358 0 \nz\nM 352 3910 \nL 883 3910 \nL 883 4474 \nL 352 4474 \nL 352 3910 \nz\n\" id=\"GuardianSansCond-Regular-69\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 346 0 \nL 870 0 \nL 870 2688 \nC 1037 2778 1306 2886 1587 2886 \nC 1965 2886 2093 2765 2093 2458 \nL 2093 0 \nL 2611 0 \nL 2611 2528 \nC 2611 3085 2342 3334 1811 3334 \nC 1440 3334 1107 3194 870 3021 \nL 845 3021 \nL 845 3283 \nL 346 3283 \nL 346 0 \nz\n\" id=\"GuardianSansCond-Regular-6e\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1325 -1235 \nC 2240 -1235 2675 -787 2675 -179 \nC 2675 416 2336 704 1696 704 \nL 1069 704 \nC 858 704 774 806 774 966 \nC 774 1069 826 1178 890 1242 \nC 1018 1197 1171 1178 1331 1178 \nC 1939 1178 2413 1530 2413 2208 \nL 2413 2304 \nC 2413 2560 2336 2739 2234 2874 \nL 2682 2874 \nL 2682 3283 \nL 1747 3283 \nC 1626 3315 1485 3334 1325 3334 \nC 704 3334 250 2944 250 2278 \nL 250 2182 \nC 250 1786 429 1510 634 1376 \nC 422 1210 294 1024 294 787 \nC 294 544 442 403 627 333 \nL 627 307 \nC 320 192 64 -38 64 -422 \nC 64 -915 506 -1235 1325 -1235 \nz\nM 1312 -819 \nC 832 -819 595 -640 595 -326 \nC 595 -51 762 141 966 218 \nL 1632 218 \nC 2022 218 2150 45 2150 -237 \nC 2150 -602 1926 -819 1312 -819 \nz\nM 1331 1581 \nC 998 1581 762 1773 762 2189 \nL 762 2323 \nC 762 2726 973 2931 1331 2931 \nC 1670 2931 1901 2746 1901 2310 \nL 1901 2176 \nC 1901 1779 1683 1581 1331 1581 \nz\n\" id=\"GuardianSansCond-Regular-67\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1184 -70 \nC 2138 -70 2509 499 2509 1248 \nC 2509 2029 2112 2259 1562 2502 \nL 1242 2643 \nC 858 2810 730 2970 730 3360 \nC 730 3808 986 4038 1459 4038 \nC 1843 4038 2144 3936 2336 3840 \nL 2336 4301 \nC 2157 4397 1914 4499 1466 4499 \nC 653 4499 198 4058 198 3322 \nC 198 2637 544 2349 1024 2138 \nL 1344 1997 \nC 1760 1811 1958 1677 1958 1197 \nC 1958 659 1702 397 1146 397 \nC 794 397 442 512 166 646 \nL 166 186 \nC 390 51 723 -70 1184 -70 \nz\n\" id=\"GuardianSansCond-Regular-53\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 346 -1222 \nL 870 -1222 \nL 870 128 \nC 1005 38 1242 -51 1542 -51 \nC 2374 -51 2784 525 2784 1600 \nL 2784 1818 \nC 2784 2912 2342 3334 1670 3334 \nC 1318 3334 1030 3187 870 3034 \nL 845 3034 \nL 845 3283 \nL 346 3283 \nL 346 -1222 \nz\nM 1466 378 \nC 1210 378 973 461 870 538 \nL 870 2688 \nC 992 2784 1229 2893 1523 2893 \nC 2035 2893 2246 2573 2246 1798 \nL 2246 1549 \nC 2246 742 2003 378 1466 378 \nz\n\" id=\"GuardianSansCond-Regular-70\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 358 0 \nL 877 0 \nL 877 4698 \nL 358 4698 \nL 358 0 \nz\n\" id=\"GuardianSansCond-Regular-6c\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 992 -51 \nC 1658 -51 2054 307 2054 947 \nC 2054 1485 1818 1709 1306 1888 \nL 1050 1978 \nC 768 2074 659 2195 659 2451 \nC 659 2771 864 2893 1222 2893 \nC 1517 2893 1760 2810 1914 2739 \nL 1914 3168 \nC 1773 3251 1568 3334 1197 3334 \nC 531 3334 160 2982 160 2406 \nC 160 1901 422 1658 826 1517 \nL 1082 1427 \nC 1408 1312 1536 1184 1536 896 \nC 1536 582 1370 390 934 390 \nC 582 390 358 480 154 563 \nL 154 141 \nC 358 19 640 -51 992 -51 \nz\n\" id=\"GuardianSansCond-Regular-73\" transform=\"scale(0.015625)\"/>\n      </defs>\n      <use xlink:href=\"#GuardianSansCond-Regular-4e\"/>\n      <use x=\"55.799988\" xlink:href=\"#GuardianSansCond-Regular-75\"/>\n      <use x=\"101.299973\" xlink:href=\"#GuardianSansCond-Regular-6d\"/>\n      <use x=\"170.799957\" xlink:href=\"#GuardianSansCond-Regular-62\"/>\n      <use x=\"217.399948\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"260.099945\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n      <use x=\"288.899933\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"304.699921\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n      <use x=\"350.299911\" xlink:href=\"#GuardianSansCond-Regular-66\"/>\n      <use x=\"376.999908\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"392.799896\" xlink:href=\"#GuardianSansCond-Regular-54\"/>\n      <use x=\"437.299881\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n      <use x=\"466.099869\" xlink:href=\"#GuardianSansCond-Regular-61\"/>\n      <use x=\"506.999863\" xlink:href=\"#GuardianSansCond-Regular-69\"/>\n      <use x=\"526.29985\" xlink:href=\"#GuardianSansCond-Regular-6e\"/>\n      <use x=\"572.399841\" xlink:href=\"#GuardianSansCond-Regular-69\"/>\n      <use x=\"591.699829\" xlink:href=\"#GuardianSansCond-Regular-6e\"/>\n      <use x=\"637.79982\" xlink:href=\"#GuardianSansCond-Regular-67\"/>\n      <use x=\"680.499817\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"696.299805\" xlink:href=\"#GuardianSansCond-Regular-53\"/>\n      <use x=\"738.499802\" xlink:href=\"#GuardianSansCond-Regular-61\"/>\n      <use x=\"779.399796\" xlink:href=\"#GuardianSansCond-Regular-6d\"/>\n      <use x=\"848.89978\" xlink:href=\"#GuardianSansCond-Regular-70\"/>\n      <use x=\"895.499771\" xlink:href=\"#GuardianSansCond-Regular-6c\"/>\n      <use x=\"914.799759\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"957.499756\" xlink:href=\"#GuardianSansCond-Regular-73\"/>\n     </g>\n    </g>\n   </g>\n   <g id=\"matplotlib.axis_2\">\n    <g id=\"ytick_1\">\n     <g id=\"line2d_9\">\n      <defs>\n       <path d=\"M 0 0 \nL -3.5 0 \n\" id=\"md8ae89ab47\" style=\"stroke:#000000;stroke-width:0.8;\"/>\n      </defs>\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"44.73125\" xlink:href=\"#md8ae89ab47\" y=\"229.044375\"/>\n      </g>\n     </g>\n     <g id=\"text_10\">\n      <!-- 0.0 -->\n      <g transform=\"translate(22.3 233.44875)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 397 0 \nL 986 0 \nL 986 621 \nL 397 621 \nL 397 0 \nz\n\" id=\"GuardianSansCond-Regular-2e\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n       <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_2\">\n     <g id=\"line2d_10\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"44.73125\" xlink:href=\"#md8ae89ab47\" y=\"185.556375\"/>\n      </g>\n     </g>\n     <g id=\"text_11\">\n      <!-- 0.2 -->\n      <g transform=\"translate(23.62 189.96075)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n       <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-32\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_3\">\n     <g id=\"line2d_11\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"44.73125\" xlink:href=\"#md8ae89ab47\" y=\"142.068375\"/>\n      </g>\n     </g>\n     <g id=\"text_12\">\n      <!-- 0.4 -->\n      <g transform=\"translate(23.03125 146.47275)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n       <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-34\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_4\">\n     <g id=\"line2d_12\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"44.73125\" xlink:href=\"#md8ae89ab47\" y=\"98.580375\"/>\n      </g>\n     </g>\n     <g id=\"text_13\">\n      <!-- 0.6 -->\n      <g transform=\"translate(22.85125 102.98475)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 1632 384 \nC 1082 384 845 915 845 1837 \nL 845 2054 \nC 992 2150 1254 2278 1594 2278 \nC 2086 2278 2323 1984 2323 1357 \nC 2323 742 2086 384 1632 384 \nz\nM 1632 -70 \nC 2438 -70 2886 544 2886 1408 \nC 2886 2234 2490 2746 1754 2746 \nC 1363 2746 1056 2611 845 2445 \nC 877 3853 1440 4038 2029 4038 \nC 2291 4038 2554 3974 2688 3930 \nL 2688 4352 \nC 2547 4442 2272 4499 1958 4499 \nC 1043 4499 275 3981 275 2208 \nL 275 1843 \nC 275 691 717 -70 1632 -70 \nz\n\" id=\"GuardianSansCond-Regular-36\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n       <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-36\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_5\">\n     <g id=\"line2d_13\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"44.73125\" xlink:href=\"#md8ae89ab47\" y=\"55.092375\"/>\n      </g>\n     </g>\n     <g id=\"text_14\">\n      <!-- 0.8 -->\n      <g transform=\"translate(22.79125 59.49675)scale(0.12 -0.12)\">\n       <defs>\n        <path d=\"M 1581 -64 \nC 2445 -64 2931 467 2931 1139 \nC 2931 1760 2592 2074 2106 2330 \nC 2490 2541 2822 2899 2822 3424 \nC 2822 4077 2394 4499 1606 4499 \nC 819 4499 384 4013 384 3398 \nC 384 2848 678 2509 1069 2285 \nC 602 2054 230 1683 230 1082 \nC 230 384 710 -64 1581 -64 \nz\nM 1587 390 \nC 1082 390 768 678 768 1126 \nC 768 1581 1037 1894 1414 2080 \nC 2029 1792 2387 1574 2387 1082 \nC 2387 614 2054 390 1587 390 \nz\nM 1754 2522 \nC 1210 2778 928 3021 928 3430 \nC 928 3821 1171 4051 1606 4051 \nC 2067 4051 2285 3782 2285 3398 \nC 2285 3008 2099 2765 1754 2522 \nz\n\" id=\"GuardianSansCond-Regular-38\" transform=\"scale(0.015625)\"/>\n       </defs>\n       <use xlink:href=\"#GuardianSansCond-Regular-30\"/>\n       <use x=\"53.499985\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n       <use x=\"75.099976\" xlink:href=\"#GuardianSansCond-Regular-38\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"ytick_6\">\n     <g id=\"line2d_14\">\n      <g>\n       <use style=\"stroke:#000000;stroke-width:0.8;\" x=\"44.73125\" xlink:href=\"#md8ae89ab47\" y=\"11.604375\"/>\n      </g>\n     </g>\n     <g id=\"text_15\">\n      <!-- 1.0 -->\n      <g transform=\"translate(25.264375 16.00875)scale(0.12 -0.12)\">\n       <use xlink:href=\"#GuardianSansCond-Regular-31\"/>\n       <use x=\"28.799988\" xlink:href=\"#GuardianSansCond-Regular-2e\"/>\n       <use x=\"50.399979\" xlink:href=\"#GuardianSansCond-Regular-30\"/>\n      </g>\n     </g>\n    </g>\n    <g id=\"text_16\">\n     <!-- F1 Score -->\n     <g transform=\"translate(16.00875 137.375625)rotate(-90)scale(0.12 -0.12)\">\n      <defs>\n       <path d=\"M 416 0 \nL 954 0 \nL 954 1984 \nL 2227 1984 \nL 2227 2445 \nL 954 2445 \nL 954 3974 \nL 2573 3974 \nL 2573 4435 \nL 416 4435 \nL 416 0 \nz\n\" id=\"GuardianSansCond-Regular-46\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1498 -51 \nC 1869 -51 2125 51 2298 173 \nL 2298 595 \nC 2163 518 1907 390 1555 390 \nC 1005 390 723 704 723 1542 \nL 723 1798 \nC 723 2464 941 2893 1510 2893 \nC 1856 2893 2054 2803 2259 2714 \nL 2259 3149 \nC 2093 3258 1875 3334 1542 3334 \nC 768 3334 186 2886 186 1741 \nL 186 1536 \nC 186 416 685 -51 1498 -51 \nz\n\" id=\"GuardianSansCond-Regular-63\" transform=\"scale(0.015625)\"/>\n      </defs>\n      <use xlink:href=\"#GuardianSansCond-Regular-46\"/>\n      <use x=\"42.499985\" xlink:href=\"#GuardianSansCond-Regular-31\"/>\n      <use x=\"71.299973\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"87.09996\" xlink:href=\"#GuardianSansCond-Regular-53\"/>\n      <use x=\"129.299957\" xlink:href=\"#GuardianSansCond-Regular-63\"/>\n      <use x=\"167.099945\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n      <use x=\"212.699936\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n      <use x=\"241.499924\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n     </g>\n    </g>\n   </g>\n   <g id=\"line2d_15\">\n    <path clip-path=\"url(#pcb9ab5d13c)\" d=\"M 44.73125 73.775275 \nL 379.53125 73.775275 \n\" style=\"fill:none;stroke:#ff0000;stroke-dasharray:5.55,2.4;stroke-dashoffset:0;stroke-width:1.5;\"/>\n   </g>\n   <g id=\"line2d_16\">\n    <path clip-path=\"url(#pcb9ab5d13c)\" d=\"M 59.949432 199.183608 \nL 80.240341 81.05556 \nL 120.822159 62.288426 \nL 201.985795 54.803464 \nL 364.313068 46.182096 \n\" style=\"fill:none;stroke:#0071bc;stroke-linecap:square;stroke-width:1.5;\"/>\n   </g>\n   <g id=\"patch_3\">\n    <path d=\"M 44.73125 229.044375 \nL 44.73125 11.604375 \n\" style=\"fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;\"/>\n   </g>\n   <g id=\"patch_4\">\n    <path d=\"M 379.53125 229.044375 \nL 379.53125 11.604375 \n\" style=\"fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;\"/>\n   </g>\n   <g id=\"patch_5\">\n    <path d=\"M 44.73125 229.044375 \nL 379.53125 229.044375 \n\" style=\"fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;\"/>\n   </g>\n   <g id=\"patch_6\">\n    <path d=\"M 44.73125 11.604375 \nL 379.53125 11.604375 \n\" style=\"fill:none;stroke:#000000;stroke-linecap:square;stroke-linejoin:miter;stroke-width:0.8;\"/>\n   </g>\n   <g id=\"legend_1\">\n    <g id=\"patch_7\">\n     <path d=\"M 256.48625 223.044375 \nL 371.13125 223.044375 \nQ 373.53125 223.044375 373.53125 220.644375 \nL 373.53125 187.644375 \nQ 373.53125 185.244375 371.13125 185.244375 \nL 256.48625 185.244375 \nQ 254.08625 185.244375 254.08625 187.644375 \nL 254.08625 220.644375 \nQ 254.08625 223.044375 256.48625 223.044375 \nz\n\" style=\"fill:#ffffff;opacity:0.8;stroke:#cccccc;stroke-linejoin:miter;\"/>\n    </g>\n    <g id=\"line2d_17\">\n     <path d=\"M 258.88625 194.653125 \nL 282.88625 194.653125 \n\" style=\"fill:none;stroke:#ff0000;stroke-dasharray:5.55,2.4;stroke-dashoffset:0;stroke-width:1.5;\"/>\n    </g>\n    <g id=\"line2d_18\"/>\n    <g id=\"text_17\">\n     <!-- Zero-shot from de -->\n     <g transform=\"translate(292.48625 198.853125)scale(0.12 -0.12)\">\n      <defs>\n       <path d=\"M 160 0 \nL 2630 0 \nL 2630 461 \nL 781 461 \nL 781 480 \nL 2688 4032 \nL 2688 4435 \nL 301 4435 \nL 301 3974 \nL 2054 3974 \nL 2054 3955 \nL 160 397 \nL 160 0 \nz\n\" id=\"GuardianSansCond-Regular-5a\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 333 1536 \nL 1408 1536 \nL 1408 1990 \nL 333 1990 \nL 333 1536 \nz\n\" id=\"GuardianSansCond-Regular-2d\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 346 0 \nL 870 0 \nL 870 2688 \nC 1037 2778 1306 2886 1587 2886 \nC 1965 2886 2093 2765 2093 2458 \nL 2093 0 \nL 2611 0 \nL 2611 2528 \nC 2611 3085 2336 3334 1805 3334 \nC 1440 3334 1126 3200 896 3034 \nL 870 3034 \nL 870 4698 \nL 346 4698 \nL 346 0 \nz\n\" id=\"GuardianSansCond-Regular-68\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1222 -38 \nC 1466 -38 1619 26 1722 90 \nL 1722 442 \nC 1632 416 1517 390 1382 390 \nC 1107 390 1005 480 1005 781 \nL 1005 2848 \nL 1683 2848 \nL 1683 3283 \nL 1005 3283 \nL 1005 4051 \nL 486 4051 \nL 486 3283 \nL 96 3283 \nL 96 2848 \nL 486 2848 \nL 486 710 \nC 486 160 774 -38 1222 -38 \nz\n\" id=\"GuardianSansCond-Regular-74\" transform=\"scale(0.015625)\"/>\n       <path d=\"M 1357 -51 \nC 1690 -51 1965 77 2125 243 \nL 2150 243 \nL 2195 0 \nL 2637 0 \nL 2637 4698 \nL 2118 4698 \nL 2118 3174 \nC 1978 3264 1734 3334 1466 3334 \nC 634 3334 198 2739 198 1670 \nL 198 1523 \nC 198 339 723 -51 1357 -51 \nz\nM 1504 378 \nC 992 378 742 749 742 1542 \nL 742 1728 \nC 742 2541 979 2893 1536 2893 \nC 1779 2893 2003 2835 2118 2758 \nL 2118 595 \nC 1984 493 1773 378 1504 378 \nz\n\" id=\"GuardianSansCond-Regular-64\" transform=\"scale(0.015625)\"/>\n      </defs>\n      <use xlink:href=\"#GuardianSansCond-Regular-5a\"/>\n      <use x=\"44.399994\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"87.099991\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n      <use x=\"115.899979\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n      <use x=\"161.499969\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n      <use x=\"188.699966\" xlink:href=\"#GuardianSansCond-Regular-73\"/>\n      <use x=\"223.199951\" xlink:href=\"#GuardianSansCond-Regular-68\"/>\n      <use x=\"269.299942\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n      <use x=\"314.899933\" xlink:href=\"#GuardianSansCond-Regular-74\"/>\n      <use x=\"343.899918\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"359.699905\" xlink:href=\"#GuardianSansCond-Regular-66\"/>\n      <use x=\"386.399902\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n      <use x=\"415.19989\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n      <use x=\"460.799881\" xlink:href=\"#GuardianSansCond-Regular-6d\"/>\n      <use x=\"530.299866\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"546.099854\" xlink:href=\"#GuardianSansCond-Regular-64\"/>\n      <use x=\"592.699844\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n     </g>\n    </g>\n    <g id=\"line2d_19\">\n     <path d=\"M 258.88625 211.753125 \nL 282.88625 211.753125 \n\" style=\"fill:none;stroke:#0071bc;stroke-linecap:square;stroke-width:1.5;\"/>\n    </g>\n    <g id=\"line2d_20\"/>\n    <g id=\"text_18\">\n     <!-- Fine-tuned on fr -->\n     <g transform=\"translate(292.48625 215.953125)scale(0.12 -0.12)\">\n      <use xlink:href=\"#GuardianSansCond-Regular-46\"/>\n      <use x=\"42.499985\" xlink:href=\"#GuardianSansCond-Regular-69\"/>\n      <use x=\"61.799973\" xlink:href=\"#GuardianSansCond-Regular-6e\"/>\n      <use x=\"107.899963\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"150.59996\" xlink:href=\"#GuardianSansCond-Regular-2d\"/>\n      <use x=\"177.799957\" xlink:href=\"#GuardianSansCond-Regular-74\"/>\n      <use x=\"206.799942\" xlink:href=\"#GuardianSansCond-Regular-75\"/>\n      <use x=\"252.299927\" xlink:href=\"#GuardianSansCond-Regular-6e\"/>\n      <use x=\"298.399918\" xlink:href=\"#GuardianSansCond-Regular-65\"/>\n      <use x=\"341.099915\" xlink:href=\"#GuardianSansCond-Regular-64\"/>\n      <use x=\"387.699905\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"403.499893\" xlink:href=\"#GuardianSansCond-Regular-6f\"/>\n      <use x=\"449.099884\" xlink:href=\"#GuardianSansCond-Regular-6e\"/>\n      <use x=\"495.199875\" xlink:href=\"#GuardianSansCond-Regular-20\"/>\n      <use x=\"510.999863\" xlink:href=\"#GuardianSansCond-Regular-66\"/>\n      <use x=\"537.69986\" xlink:href=\"#GuardianSansCond-Regular-72\"/>\n     </g>\n    </g>\n   </g>\n  </g>\n </g>\n <defs>\n  <clipPath id=\"pcb9ab5d13c\">\n   <rect height=\"217.44\" width=\"334.8\" x=\"44.73125\" y=\"11.604375\"/>\n  </clipPath>\n </defs>\n</svg>\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots()\n",
    "ax.axhline(f1_scores[\"de\"][\"fr\"], ls=\"--\", color=\"r\")\n",
    "metrics_df.set_index(\"num_samples\").plot(ax=ax)\n",
    "plt.legend([\"Zero-shot from de\", \"Fine-tuned on fr\"], loc=\"lower right\")\n",
    "plt.ylim((0, 1))\n",
    "plt.xlabel(\"Number of Training Samples\")\n",
    "plt.ylabel(\"F1 Score\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Fine-Tuning on Multiple Languages at Once"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import concatenate_datasets\n",
    "\n",
    "def concatenate_splits(corpora):\n",
    "    multi_corpus = DatasetDict()\n",
    "    for split in corpora[0].keys():\n",
    "        multi_corpus[split] = concatenate_datasets(\n",
    "            [corpus[split] for corpus in corpora]).shuffle(seed=42)\n",
    "    return multi_corpus"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "panx_de_fr_encoded = concatenate_splits([panx_de_encoded, panx_fr_encoded])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "training_args.logging_steps = len(panx_de_fr_encoded[\"train\"]) // batch_size\n",
    "training_args.push_to_hub = True\n",
    "training_args.output_dir = \"xlm-roberta-base-finetuned-panx-de-fr\"\n",
    "\n",
    "trainer = Trainer(model_init=model_init, args=training_args,\n",
    "    data_collator=data_collator, compute_metrics=compute_metrics,\n",
    "    tokenizer=xlmr_tokenizer, train_dataset=panx_de_fr_encoded[\"train\"],\n",
    "    eval_dataset=panx_de_fr_encoded[\"validation\"])\n",
    "\n",
    "trainer.train()\n",
    "trainer.push_to_hub(commit_message=\"Training completed!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de-fr] model on [de] dataset: 0.866\n",
      "F1-score of [de-fr] model on [fr] dataset: 0.868\n",
      "F1-score of [de-fr] model on [it] dataset: 0.815\n",
      "F1-score of [de-fr] model on [en] dataset: 0.677\n"
     ]
    }
   ],
   "source": [
    "#hide_output\n",
    "for lang in langs:\n",
    "    f1 = evaluate_lang_performance(lang, trainer)\n",
    "    print(f\"F1-score of [de-fr] model on [{lang}] dataset: {f1:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score of [de-fr] model on [de] dataset: 0.866\n",
      "F1-score of [de-fr] model on [fr] dataset: 0.868\n",
      "F1-score of [de-fr] model on [it] dataset: 0.815\n",
      "F1-score of [de-fr] model on [en] dataset: 0.677\n"
     ]
    }
   ],
   "source": [
    "#hide_input\n",
    "for lang in langs:\n",
    "    f1 = evaluate_lang_performance(lang, trainer)\n",
    "    print(f\"F1-score of [de-fr] model on [{lang}] dataset: {f1:.3f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "corpora = [panx_de_encoded]\n",
    "\n",
    "# Exclude German from iteration\n",
    "for lang in langs[1:]:\n",
    "    training_args.output_dir = f\"xlm-roberta-base-finetuned-panx-{lang}\"\n",
    "    # Fine-tune on monolingual corpus\n",
    "    ds_encoded = encode_panx_dataset(panx_ch[lang])\n",
    "    metrics = train_on_subset(ds_encoded, ds_encoded[\"train\"].num_rows)\n",
    "    # Collect F1-scores in common dict\n",
    "    f1_scores[lang][lang] = metrics[\"f1_score\"][0]\n",
    "    # Add monolingual corpus to list of corpora to concatenate\n",
    "    corpora.append(ds_encoded)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "corpora_encoded = concatenate_splits(corpora)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# hide_output\n",
    "training_args.logging_steps = len(corpora_encoded[\"train\"]) // batch_size\n",
    "training_args.output_dir = \"xlm-roberta-base-finetuned-panx-all\"\n",
    "\n",
    "trainer = Trainer(model_init=model_init, args=training_args,\n",
    "    data_collator=data_collator, compute_metrics=compute_metrics,\n",
    "    tokenizer=xlmr_tokenizer, train_dataset=corpora_encoded[\"train\"],\n",
    "    eval_dataset=corpora_encoded[\"validation\"])\n",
    "\n",
    "trainer.train()\n",
    "trainer.push_to_hub(commit_message=\"Training completed!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "      \n",
       "      <progress value='419' max='263' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      [263/263 00:10]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# hide_output\n",
    "for idx, lang in enumerate(langs):\n",
    "    f1_scores[\"all\"][lang] = get_f1_score(trainer, corpora[idx][\"test\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>Evaluated on</th>\n",
       "      <th>de</th>\n",
       "      <th>fr</th>\n",
       "      <th>it</th>\n",
       "      <th>en</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fine-tune on</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>de</th>\n",
       "      <td>0.8677</td>\n",
       "      <td>0.7141</td>\n",
       "      <td>0.6923</td>\n",
       "      <td>0.5890</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>each</th>\n",
       "      <td>0.8677</td>\n",
       "      <td>0.8505</td>\n",
       "      <td>0.8192</td>\n",
       "      <td>0.7068</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>all</th>\n",
       "      <td>0.8682</td>\n",
       "      <td>0.8647</td>\n",
       "      <td>0.8575</td>\n",
       "      <td>0.7870</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Evaluated on      de      fr      it      en\n",
       "Fine-tune on                                \n",
       "de            0.8677  0.7141  0.6923  0.5890\n",
       "each          0.8677  0.8505  0.8192  0.7068\n",
       "all           0.8682  0.8647  0.8575  0.7870"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scores_data = {\"de\": f1_scores[\"de\"],\n",
    "               \"each\": {lang: f1_scores[lang][lang] for lang in langs},\n",
    "               \"all\": f1_scores[\"all\"]}\n",
    "f1_scores_df = pd.DataFrame(scores_data).T.round(4)\n",
    "f1_scores_df.rename_axis(index=\"Fine-tune on\", columns=\"Evaluated on\",\n",
    "                         inplace=True)\n",
    "f1_scores_df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Interacting with Model Widgets"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img alt=\"A Hub widget\" caption=\"Example of a widget on the Hugging Face Hub\" src=\"images/chapter04_ner-widget.png\" id=\"ner-widget\"/>  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Conclusion"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.7.3 64-bit",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "name": "python",
   "version": "3.7.3"
  },
  "vscode": {
   "interpreter": {
    "hash": "31f2aee4e71d21fbe5cf8b01ff0e069b9275f58929596ceb00d14d90e3e16cd6"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}