Merge pull request #1 from GilesStrong/fixes_and_checks

Fixes and checks
GilesStrong · Jan 28, 2022 · 1989a58 · 1989a58
2 parents 7343372 + 7a7bf46
commit 1989a58
Show file tree

Hide file tree

Showing 14 changed files with 71 additions and 37 deletions.
diff --git a/README.md b/README.md
@@ -2,9 +2,7 @@
 
 # High-Energy Muon Calorimeter Regression Study
 
-Public version of code used in "Calorimetric Measurement of Multi-TeV Muons via Deep Regression" by Jan Kieseler, Giles C. Strong, Filippo Chiandotto, Tommaso Dorigo, & Lukas Layer, (2021), [arXiv:2107.02119 [physics.ins-det]](https://arxiv.org/abs/2107.02119)
-
-**Warning: This repo is currently being populated with a public version of the research code-base. It is currently incomplete, and has not yet been fully tested. Please check back soon**
+Public version of code used in "Calorimetric Measurement of Multi-TeV Muons via Deep Regression" by Jan Kieseler, Giles C. Strong, Filippo Chiandotto, Tommaso Dorigo, & Lukas Layer, [The European Physical Journal C volume 82, Article number: 79 (2022)](https://link.springer.com/article/10.1140%2Fepjc%2Fs10052-022-09993-5)
 
 ## Installation
 
@@ -21,7 +19,7 @@ conda env create -f environment.yml
 
 This will create a new Conda environment called `muon-regression`, which can be activated using:
 ```
-conda activate muon-regression
+conda activate calo_muon_regression
 ```
 
 Alternatively:
@@ -32,6 +30,8 @@ may be used to install the dependencies.
 
 Self-hosted documentation, installation guide, and user guide are available by opening `./docs/build/html/index.html` with a web-browser.
 
+For GPU use (which is highly recommended), please separately install a suitable version of PyTorch based on your drivers.
+
 ## Data
 
 Preprocessed datasets are available from https://doi.org/10.5281/zenodo.5163817. The example commands assume the data files are stored in `./data`

diff --git a/docs/requirements.txt b/docs/requirements.txt
@@ -2,8 +2,9 @@ sphinx==2.2.0
 sphinx_rtd_theme
 sphinx_autodoc_typehints==1.7.0
 sphinx_autodoc_annotation==1.0.post1
-lumin==0.7.2
+lumin==0.8.0
 typer
 uproot
 m2r
-sparse
+sparse
+mistune==0.8.4
diff --git a/docs/source/installation.md b/docs/source/installation.md
@@ -19,7 +19,7 @@ conda env create -f environment.yml
 
 This will create a new Conda environment called `muon-regression`, which can be activated using:
 ```
-conda activate muon-regression
+conda activate calo_muon_regression
 ```
 
 Alternatively:

diff --git a/docs/source/modules.rst b/docs/source/modules.rst
@@ -2,6 +2,6 @@ muon_regression
 ===============
 
 .. toctree::
-   :maxdepth: 1
+   :maxdepth: 4
 
    muon_regression
diff --git a/docs/source/muon_regression.data_proc.rst b/docs/source/muon_regression.data_proc.rst
@@ -10,8 +10,7 @@ Submodules
 ----------
 
 .. toctree::
-   :maxdepth: 1
-
+
    muon_regression.data_proc.data_import
    muon_regression.data_proc.detector
    muon_regression.data_proc.pre_proc
diff --git a/docs/source/muon_regression.experiment.rst b/docs/source/muon_regression.experiment.rst
@@ -10,6 +10,5 @@ Submodules
 ----------
 
 .. toctree::
-   :maxdepth: 1
-
+
    muon_regression.experiment.exp_journal
diff --git a/docs/source/muon_regression.fqs.rst b/docs/source/muon_regression.fqs.rst
@@ -0,0 +1,7 @@
+muon\_regression.fqs module
+===========================
+
+.. automodule:: muon_regression.fqs
+   :members:
+   :undoc-members:
+   :show-inheritance:
diff --git a/docs/source/muon_regression.nn.rst b/docs/source/muon_regression.nn.rst
@@ -10,14 +10,10 @@ Submodules
 ----------
 
 .. toctree::
-   :maxdepth: 1
-
+
    muon_regression.nn.callbacks
    muon_regression.nn.conv3d
-   muon_regression.nn.experimental
    muon_regression.nn.hooks
    muon_regression.nn.loss
    muon_regression.nn.metrics
    muon_regression.nn.models
-   muon_regression.nn.running_batchnorm
-   muon_regression.nn.tails
diff --git a/docs/source/muon_regression.plotting.rst b/docs/source/muon_regression.plotting.rst
@@ -10,13 +10,7 @@ Submodules
 ----------
 
 .. toctree::
-<<<<<<< Updated upstream
-   :maxdepth: 4
 
-=======
-   :maxdepth: 1
-
->>>>>>> Stashed changes
    muon_regression.plotting.event_display
    muon_regression.plotting.fitting
    muon_regression.plotting.plot_regression
diff --git a/docs/source/muon_regression.result_proc.rst b/docs/source/muon_regression.result_proc.rst
@@ -10,7 +10,6 @@ Submodules
 ----------
 
 .. toctree::
-   :maxdepth: 1
-
+
    muon_regression.result_proc.binning
    muon_regression.result_proc.correction
diff --git a/docs/source/muon_regression.rst b/docs/source/muon_regression.rst
@@ -10,10 +10,16 @@ Subpackages
 -----------
 
 .. toctree::
-   :maxdepth: 1
-
+
    muon_regression.data_proc
    muon_regression.experiment
    muon_regression.nn
    muon_regression.plotting
    muon_regression.result_proc
+
+Submodules
+----------
+
+.. toctree::
+
+   muon_regression.fqs
diff --git a/environment.yml b/environment.yml
@@ -1,19 +1,21 @@
-name: muon-regression
+name: calo_muon_regression
 channels:
   - conda-forge
   - defaults
 dependencies:
-  - python>=3.6.5
+  - python>=3.6.5,<3.8
+  - ipython
   - pip
+  - pip:
+    - torch==1.9.0
     - seaborn==0.9.0
     - matplotlib==3.2.2
     - pandas==0.25.0
-    - torch==1.9.0
     - fastprogress==0.1.21
-    - scipy==1.5.4
+    - scipy==1.2.0
     - numpy==1.21
-    - h5py==2.7.1
-    - scikit-learn==0.22.0
+    - h5py
+    - scikit-learn==0.22.2
     - statsmodels==0.9.0
     - sympy==1.1.1
     - pdpbox~=0.2.0

diff --git a/muon_regression/nn/conv3d.py b/muon_regression/nn/conv3d.py
@@ -9,12 +9,42 @@
 from lumin.nn.models.layers.activations import lookup_act
 from lumin.nn.models.layers.batchnorms import RunningBatchNorm3d
 from lumin.utils.misc import to_device, to_tensor
-from lumin.nn.models.blocks.conv_blocks import SEBlock3d
+from lumin.nn.models.blocks.conv_blocks import SEBlock1d
 
 
 __all__ = ['PreActMuonConv3dBlock']
 
 
+class SEBlock3d(SEBlock1d):
+    r'''
+    Copyright 2018 onwards Giles Strong
+
+    Licensed under the Apache License, Version 2.0 (the "License");
+    you may not use this file except in compliance with the License.
+    You may obtain a copy of the License at
+
+    Include fix from lumin=0.8.1
+
+    Squeeze-excitation block [Hu, Shen, Albanie, Sun, & Wu, 2017](https://arxiv.org/abs/1709.01507).
+    Incoming data is averaged per channel, fed through a single layer of width `n_in//r` and the chose activation, then a second layer of width `n_in` and a sigmoid activation.
+    Channels in the original data are then multiplied by the learned channe weights.
+    Arguments:
+        n_in: number of incoming channels
+        r: the reduction ratio for the channel compression
+        act: string representation of argument to pass to lookup_act
+        lookup_init: function taking choice of activation function, number of inputs, and number of outputs an returning a function to initialise layer weights.
+        lookup_act: function taking choice of activation function and returning an activation function layer
+    '''
+
+    def __init__(self, n_in:int, r:int, act:str='relu', lookup_init:Callable[[str,Optional[int],Optional[int]],Callable[[Tensor],None]]=lookup_normal_init,
+                 lookup_act:Callable[[str],Any]=lookup_act):
+        super().__init__(n_in=n_in, r=r)
+        self.n_in,self.r,self.act,self.lookup_init,self.lookup_act = n_in,r,act,lookup_init,lookup_act
+        self.layers = self._get_layers()
+        self.sz = [1,1,1]
+        self.pool = nn.AdaptiveAvgPool3d(self.sz)
+
+
 class PreActMuonConv3dBlock(nn.Module):
     r'''
     Pre-activation Conv3D block with better placement of BN layers and slightly modified shortcut layers. Energy is expected to be channel 0.

diff --git a/requirements.txt b/requirements.txt
@@ -5,8 +5,8 @@ torch==1.9.0
 fastprogress==0.1.21
 scipy==1.5.4
 numpy==1.21
-h5py==2.7.1
-scikit-learn==0.22.0
+h5py
+scikit-learn==0.22.2
 statsmodels==0.9.0
 sympy==1.1.1
 pdpbox~=0.2.0
@@ -16,4 +16,5 @@ sparse==0.11.2
 fastcore==1.3.5
 lumin==0.8.0
 typer==0.3.0
-uproot==3.11
+uproot==3.11
+ipython