Backports for v0.12.1 (#2626)

* Fix: torch `PoissonOutput` scaling (#2619)

* Remove `dataclasses` requirement (#2623)

* Fix installation docs, fix typos in docstrings (#2625)

---------

Co-authored-by: Kashif Rasul <[email protected]>

docs/getting_started/install.md

@@ -25,7 +25,7 @@ For production usage, we suggest restricting the version when installing
 GluonTS:
 
 ```sh
-pip install gluonts==0.10.*
+pip install gluonts==0.12.*
 ```
 ````
 
@@ -47,7 +47,7 @@ When installing a package, they are passed via ``[...]`` after the package
 name, but before the version specifier:
 
 ```sh
-pip install some-package[extra-1,extra-2]==version
+pip install "some-package[extra-1,extra-2]==version"
 ````
 
 We make extensive use of optional dependencies in GluonTS to keep the amount of
@@ -61,7 +61,7 @@ offer the ``arrow``-extra, which installs the required packages and can be
 simply enabled using:
 
 ```sh
-pip install gluonts[arrow]
+pip install "gluonts[arrow]"
 ````
 
 ### Models
@@ -78,7 +78,7 @@ to be installed as well.
 Both required dependencies are included in the ``torch``-extra:
 
 ```sh
-pip install gluonts[torch]
+pip install "gluonts[torch]"
 ````
 
 
@@ -88,7 +88,7 @@ MXNet based models require a version of ``mxnet`` to be installed.
 
 ```{note}
 
-MXNet provives different package for CPU and GPU usages. Please refer to its
+MXNet provides different package for CPU and GPU usages. Please refer to its
 [documentation](https://mxnet.apache.org/versions/1.9.1/get_started?) to
 select the right version fitting your use-case.
 
@@ -97,7 +97,7 @@ select the right version fitting your use-case.
 The ``mxnet``-extra will install a CPU-only version:
 
 ```sh
-pip install gluonts[mxnet]
+pip install "gluonts[mxnet]"
 
 ````
 
@@ -115,22 +115,20 @@ install the `forecast` package:
 R -e 'install.packages(c("forecast", "nnfor"), repos="https://cloud.r-project.org")'
 ```
 
-In addition, we require rpy2 to be installed:
+You will also need to install ``R``-extra dependencies with:
 
 ```sh
-pip install 'rpy2>=2.9.*,<3.*'
+pip install "gluonts[R]"
 ````
 
 ##### Prophet
 
 The [Prophet](https://facebook.github.io/prophet/) forecasting library is
 available via `gluonts.model.prophet` and requires the ``prophet`` package to
-be installed.
-
-The ``prophet``-extra also depends on it:
+be installed. You can get it by installing the ``prophet``-extra:
 
 ```sh
-pip install gluonts[prophet]
+pip install "gluonts[prophet]"
 ```
 
 
@@ -166,7 +164,7 @@ To utilise these, either install the ``pyarrow`` package or use the
 ``arrow``-extra:
 
 ```sh
-pip install gluonts[arrow]
+pip install "gluonts[arrow]"
 ```
 
 ### Other
@@ -177,5 +175,5 @@ The ``shell`` module offers integration with Amazon SageMaker and is available
 through:
 
 ```sh
-pip install gluonts[shell]
+pip install "gluonts[shell]"
 ```

requirements/requirements.txt

@@ -6,5 +6,3 @@ toolz~=0.10
 
 # used for Protocol, can be removed once Python 3.8 is required
 typing-extensions~=4.0
-
-dataclasses

src/gluonts/core/component.py

@@ -74,7 +74,7 @@ def equals(this: Any, that: Any) -> bool:
 
     In addition, the function dispatches to specialized implementations based
     on the type of the first argument, so the above conditions might be
-    sticter for certain types.
+    stricter for certain types.
 
     Parameters
     ----------
@@ -112,7 +112,7 @@ def equals_default_impl(this: Any, that: Any) -> bool:
 
     1. Their types match.
     2. If their initializer are :func:`validated`, their initializer arguments
-       are pairlise structurally equal.
+       are pairwise structurally equal.
     3. If their initializer are not :func:`validated`, they are referentially
        equal (i.e. ``this == that``).
 

src/gluonts/core/settings.py

@@ -291,7 +291,7 @@ def __getitem__(self, key):
         raise KeyError(key)
 
     def __getattribute__(self, key):
-        # We check the key, to check whether we want to acces our chainmap
+        # We check the key, to check whether we want to access our chainmap
         # or handle it as a normal attribute.
         if key.startswith("_"):
             return super().__getattribute__(key)
@@ -367,7 +367,7 @@ def _inject(self, *keys, **kwargs):
         """
         Dependency injection.
 
-        This will inject values from settings if avaiable and not passed
+        This will inject values from settings if available and not passed
         directly::
 
             @settings._inject("foo")

src/gluonts/dataset/arrow/enc.py

@@ -98,7 +98,7 @@ def _encode_py_to_arrow(val):
 
 
 @_encode_py_to_arrow.register
-def _encode_py_pd_preiod(val: pd.Period):
+def _encode_py_pd_period(val: pd.Period):
     return val.to_timestamp()
 
 

src/gluonts/dataset/artificial/_base.py

@@ -250,7 +250,7 @@ def compute_data_from_recipe(
             max_train_length=max_train_length,
             prediction_length=self.prediction_length,
             # Add 1 time series at a time in the loop for different constant
-            # valus per time series
+            # values per time series
             num_timeseries=1,
         )
         generated = data.generate()

src/gluonts/dataset/hierarchical.py

@@ -42,7 +42,7 @@ class HierarchicalTimeSeries:
         corresponding to the leaves of the hierarchy.
     S
         Summation or aggregation matrix whose ordering should be consistent
-        with the orderig of the columns of `ts_at_all_levels`.
+        with the ordering of the columns of `ts_at_all_levels`.
         In particular, the bottom `k x k` sub-matrix should be identity matrix,
         where `k` is the number of leaves of the hierarchy.
     """
@@ -109,7 +109,7 @@ def aggregate_ts(
             corresponding to the leaves of the hierarchy.
         S
             Summation or aggregation matrix whose ordering should be consistent
-            with the orderig of the columns of `ts_at_all_levels`.
+            with the ordering of the columns of `ts_at_all_levels`.
             In particular, the bottom `k x k` sub-matrix should be an identity
             matrix, where `k` is the number of leaves of the hierarchy.
 
@@ -128,7 +128,7 @@ def aggregate_ts(
             f"{ts_at_bottom_level.shape}."
         )
 
-        # Last `num_bottom_ts` rows contain the identity marix.
+        # Last `num_bottom_ts` rows contain the identity matrix.
         assert (S[num_agg_ts:, :] == np.eye(num_bottom_ts)).all(), (
             f"The last {num_bottom_ts} rows of aggregation matrix `S`"
             f" should contain Identity matrix."
@@ -168,7 +168,7 @@ def to_dataset(
         ----------
         feat_dynamic_real
             A pandas dataframe containing dynamic features as columns.
-            Note that features of any (or all) time series in the hierachy
+            Note that features of any (or all) time series in the hierarchy
             can be passed here, since all time series are considered together
             as a single multivariate time series.
 
@@ -184,7 +184,7 @@ def to_dataset(
                 self.ts_at_all_levels.index[0] == feat_dynamic_real.index[0]
             ), (
                 "The staring time point of dynamic features should match "
-                "with that of the hierarhical time series. "
+                "with that of the hierarchical time series. "
                 f"Start of `feat_dynamic_real`: "
                 f"{feat_dynamic_real.index[0]} and "
                 f"the start of hierarchical time series: "
@@ -194,7 +194,7 @@ def to_dataset(
             assert feat_dynamic_real.index.intersection(
                 self.ts_at_all_levels.index
             ).equals(self.ts_at_all_levels.index), (
-                "Dynamic features should be provied for all time "
+                "Dynamic features should be provided for all time "
                 "points where the target is defined. "
                 f"Index of `feat_dynamic_real`: {feat_dynamic_real.index}, \n"
                 f"Index of `ts_at_all_levels` of `hts`: "

src/gluonts/dataset/pandas.py

@@ -257,7 +257,7 @@ def from_long_dataframe(
         Returns
         -------
         PandasDataset
-            Gluonts dataset based on ``pandas.DataFrame``s.
+            Dataset containing series data from the given long dataframe.
         """
         if not isinstance(dataframe.index, DatetimeIndexOpsMixin):
             dataframe.index = pd.to_datetime(dataframe.index)

src/gluonts/ev/aggregations.py

@@ -68,7 +68,7 @@ class Mean(Aggregation):
 
     `partial_result` represents one of two things, depending on the axis:
     Case 1 - axis 0 is aggregated (axis is None or 0):
-        First sum values acoording to axis and keep track of number of entries
+        First sum values according to axis and keep track of number of entries
         summed over (`n`) to divide by in the end.
 
     Case 2 - axis 0 is not being aggregated:

src/gluonts/evaluation/_base.py

@@ -432,7 +432,7 @@ def get_metrics_per_ts(
                         )
                     }
                 except Exception:
-                    logging.warning(f"Error occured when evaluating {k}.")
+                    logging.warning(f"Error occurred when evaluating {k}.")
                     val = {k: np.nan}
 
                 metrics.update(val)

src/gluonts/ext/rotbaum/_preprocess.py

@@ -348,15 +348,15 @@ def _pre_transform(
         time_series_window: list
         subtract_mean: bool
             Whether or not to subtract the mean of the context window in the
-            feaurization process.
+            featurization process.
         count_nans: bool
             Whether or not to add as a feature the number of nan values in the
             context window.
 
         Returns
         -------------
         tuple
-            trasnformed time series, dictionary with transformation data (
+            Transformed time series, dictionary with transformation data (
             std, mean, number of lag features, and number of nans of
             count_nans)
         """

src/gluonts/mx/trainer/learning_rate_scheduler.py

@@ -92,7 +92,7 @@ def reset(self) -> None:
 
     def step(self, metric_value: float) -> bool:
         if self.exceeded:
-            raise RuntimeError("Patience alread exceeded.")
+            raise RuntimeError("Patience already exceeded.")
 
         has_improved = self.objective.update(metric_value)
 

src/gluonts/mx/trainer/model_iteration_averaging.py

@@ -186,7 +186,7 @@ def __init__(
         epochs
             The total number of epochs.
         n
-            The non-montone interval.
+            The non-monotone interval.
         maximize
             Whether to maximize or minimize the validation metric.
         eta

src/gluonts/mx/util.py

@@ -34,7 +34,7 @@ class HybridContext:
         The network whose hybrid mode has to be modified within the enclosing
         context.
     hybridize
-        A boolean flag inidicating whether the hybrid mode should be set or
+        A boolean flag indicating whether the hybrid mode should be set or
         not.
     kwargs
         A dictionary of optional arguments to pass to the `hybridize()` call

src/gluonts/torch/distributions/distribution_output.py

@@ -240,6 +240,22 @@ def domain_map(cls, rate: torch.Tensor):
         rate_pos = F.softplus(rate).clone()
         return (rate_pos.squeeze(-1),)
 
+    # Overwrites the parent class method. We cannot scale using the affine
+    # transformation since Poisson should return integers. Instead we scale
+    # the parameters.
+    def distribution(
+        self,
+        distr_args,
+        loc: Optional[torch.Tensor] = None,
+        scale: Optional[torch.Tensor] = None,
+    ) -> Distribution:
+        (rate,) = distr_args
+
+        if scale is not None:
+            rate *= scale
+
+        return Poisson(rate=rate)
+
     @property
     def event_shape(self) -> Tuple:
         return ()

src/gluonts/torch/model/deep_npts/_network.py

@@ -100,7 +100,7 @@ def forward(self, features: torch.Tensor):
 class DeepNPTSNetwork(nn.Module):
     """Base class implementing a simple feed-forward neural network that takes
     in static and dynamic features and produces `num_hidden_nodes` independent
-    outptus. These outputs are then used by derived classes to construct the
+    outputs. These outputs are then used by derived classes to construct the
     forecast distribution for a single time step.
 
     Note that the dynamic features are just treated as independent features
@@ -211,7 +211,7 @@ def forward(
 class DeepNPTSNetworkDiscrete(DeepNPTSNetwork):
     """
     Extends `DeepNTPSNetwork` by implementing the output layer which
-    converts the ouptuts from the base network into probabilities of length
+    converts the outputs from the base network into probabilities of length
     `context_length`. These probabilities together with the past values in the
     context window constitute the one-step-ahead forecast distribution.
     Specifically, the forecast is always one of the values observed in the
@@ -270,7 +270,7 @@ def forward(
 class DeepNPTSNetworkSmooth(DeepNPTSNetwork):
     """
     Extends `DeepNTPSNetwork` by implementing the output layer which
-    converts the ouptuts from the base network into a smoothed mixture
+    converts the outputs from the base network into a smoothed mixture
     distribution. The components of the mixture are Gaussians centered around
     the observations in the context window. The mixing probabilities as well as
     the width of the Gaussians are predicted by the network.
@@ -320,7 +320,7 @@ def forward(
 
 class DeepNPTSMultiStepNetwork(nn.Module):
     """
-    Implements multi-step prediction given a trained `DeepNPTSNewtork` model
+    Implements multi-step prediction given a trained `DeepNPTSNetwork` model
     that outputs one-step-ahead forecast distribution.
     """
 

src/gluonts/torch/model/deepar/module.py

@@ -350,7 +350,7 @@ def forward(
             (Optional) tensor of dynamic real features in the past,
             shape: ``(batch_size, prediction_length, num_feat_dynamic_real)``.
         num_parallel_samples
-            How many future sampels to produce.
+            How many future samples to produce.
             By default, self.num_parallel_samples is used.
         """
         if num_parallel_samples is None:

src/gluonts/torch/model/mqf2/module.py

@@ -167,7 +167,7 @@ def forward(
         Parameters
         ----------
         feat_static_cat
-            Static categorial features (batch_size, num_feat_static_cat)
+            Static categorical features (batch_size, num_feat_static_cat)
         feat_static_real
             Static real-valued features (batch_size, num_feat_static_real)
         past_time_feat

src/gluonts/torch/model/simple_feedforward/estimator.py

@@ -58,7 +58,7 @@
 
 class SimpleFeedForwardEstimator(PyTorchLightningEstimator):
     """
-    An estimator training a feedforward model for forecasting.
+    An estimator training a feed-forward model for forecasting.
 
     This class is uses the model defined in ``SimpleFeedForwardModel``,
     and wraps it into a ``SimpleFeedForwardLightningModule`` for training
@@ -73,7 +73,7 @@ class SimpleFeedForwardEstimator(PyTorchLightningEstimator):
         Number of time steps prior to prediction time that the model
         takes as inputs (default: ``10 * prediction_length``).
     hidden_dimensions
-        Size of hidden layers in the feedforward network
+        Size of hidden layers in the feed-forward network
         (default: ``[20, 20]``).
     lr
         Learning rate (default: ``1e-3``).

src/gluonts/torch/model/simple_feedforward/module.py

@@ -42,7 +42,7 @@ class SimpleFeedForwardModel(nn.Module):
     context_length
         Number of time steps prior to prediction time that the model.
     hidden_dimensions
-        Size of hidden layers in the feedforward network.
+        Size of hidden layers in the feed-forward network.
     distr_output
         Distribution to use to evaluate observations and sample predictions.
         Default: ``StudentTOutput()``.

src/gluonts/transform/convert.py

@@ -591,7 +591,7 @@ def _empirical_cdf_forward_transform(
         Returns
         -------
         quantiles
-            Empirical CDF quantiles in [0, 1] interval with winzorized cutoff.
+            Empirical CDF quantiles in [0, 1] interval with winsorized cutoff.
         """
         m = sorted_values.shape[0]
         quantiles = self._forward_transform(