first commit

README.md

@@ -0,0 +1,87 @@
+# Graph ConvNets in PyTorch
+September 30, 2017
+<br>
+<br>
+
+
+
+<img align="right" src="pic/graph_convnet.jpg" style="width: 200px;"/>
+
+
+### Xavier Bresson
+<img src="pic/home100.jpg" style="width: 15px;"/> http://www.ntu.edu.sg/home/xbresson<br>
+<img src="pic/github100.jpg" style="width: 15px;"/> https://github.com/xbresson<br>
+<img src="pic/twitter100.jpg" style="width: 15px;"/> https://twitter.com/xbresson <br>
+<br>
+
+
+### Description
+Prototype implementation in PyTorch of the NIPS'16 paper:<br>
+Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering<br>
+M Defferrard, X Bresson, P Vandergheynst<br>
+Advances in Neural Information Processing Systems, 3844-3852, 2016<br>
+ArXiv preprint: [arXiv:1606.09375](https://arxiv.org/pdf/1606.09375.pdf) <br>
+<br>
+
+### Code objective
+The code provides a simple example of graph ConvNets for the MNIST classification task.<br>
+The graph is a 8-nearest neighbor graph of a 2D grid.<br>
+The signals on graph are the MNIST images vectorized as $28^2 \times 1$ vectors.<br>
+<br>
+
+
+### Installation
+   ```sh
+   git clone https://github.com/xbresson/graph_cnn_pytorch
+   cd graph_cnn_pytorch
+   pip install -r requirements.txt # installation for python 3.6.2
+   python check_install.py
+   jupyter notebook
+   ```
+
+<br>
+
+
+
+### Results
+GPU Quadro M4000<br>
+* Standard ConvNets: **01_standard_convnet_lenet5_mnist_pytorch.ipynb**, accuracy= 99.31, speed= 6.9 sec/epoch. <br>
+* Graph ConvNets: **02_graph_convnet_lenet5_mnist_pytorch.ipynb**, accuracy= 99.21, speed= 100.8 sec/epoch <br>
+<br>
+
+
+### Note
+PyTorch has not yet implemented function torch.mm(sparse, dense) for variables: https://github.com/pytorch/pytorch/issues/2389. It will be certainly implemented in the near future but in the meantime, I implemented a new autograd function for sparse variables, called "my_sparse_mm", by subclassing torch.autograd.function and implementing the forward and backward passes.
+
+
+```python
+class my_sparse_mm(torch.autograd.Function):
+    """
+    Implementation of a new autograd function for sparse variables, 
+    called "my_sparse_mm", by subclassing torch.autograd.Function 
+    and implementing the forward and backward passes.
+    """
+
+    def forward(self, W, x):  # W is SPARSE
+        self.save_for_backward(W, x)
+        y = torch.mm(W, x)
+        return y
+
+    def backward(self, grad_output):
+        W, x = self.saved_tensors 
+        grad_input = grad_output.clone()
+        grad_input_dL_dW = torch.mm(grad_input, x.t()) 
+        grad_input_dL_dx = torch.mm(W, grad_input )
+        return grad_input_dL_dW, grad_input_dL_dx
+```
+<br>
+
+
+### When to use this algorithm?
+Any problem that can be cast as analyzing a set of signals on a fixed graph, and you want to use ConvNets for this analysis.<br>
+
+<br>
+
+<br>
+<br>
+

check_install.py

@@ -0,0 +1,33 @@
+#!/bin/env python3
+
+print('\nRun Python installation test for graph ConvNets')
+
+import os
+import sys
+major, minor = sys.version_info.major, sys.version_info.minor
+
+if ( (major is not 3) or (minor is not 6) ):
+    raise Exception('please use Python 3.6 (for PyTorch), you have Python {}.{}'.format(major,minor))
+
+try:
+    import numpy 
+    print('Recommended version of numpy is {}. You have {}.'.format('1.13.1',numpy.__version__))
+    import jupyter
+    print('Recommended version of jupyter is {}. You have {}.'.format('1.0.0',jupyter.__version__))
+    import torch
+    print('Recommended version of pytorch is {}. You have {}.'.format('0.2.0_1',torch.__version__))
+    import tensorflow
+    print('Recommended version of tensorflow is {}. You have {}.'.format('0.11.0rc0',tensorflow.__version__))
+    import scipy
+    print('Recommended version of scipy is {}. You have {}.'.format('0.19.1',scipy.__version__))
+    import sklearn
+    print('Recommended version of sklearn is {}. You have {}.'.format('0.19.0',sklearn.__version__))
+
+
+except:
+    print('A package is missing or the version of the package.')
+    print('Install the package below.\n')
+    raise
+
+print('Successful installation of Python {}.{} and '
+      'most of the packages required to run the code.\n'.format(major, minor))