Graph Convolution Network, molecule activity

Question

I am trying to translate an example code for graph convolution network from Deep Learning for Physical Scientists: Accelerating Research with Machine Learning. Edward O. Pyzer-Knapp et. al. They have an example (Chapter 6, Section 6.3, Page 104-106) where they build a graph convolution network using Keras.

The model takes a 132*132 adjacency matrix built out of molecule connectivity, with a flag of 1 or 0 for active or inactive molecule.

 from tensorflow.keras import datasets, layers, models 
 from sklearn.model_selection import train_test_split 
 model = models.Sequential() 
 model.add(layers.Conv2D(64, (3, 3), activation='relu', input_shape=(None, None, 1), padding='SAME')) 
 model.add(layers.MaxPooling2D((2, 2))) 
 model.add(layers.Conv2D(32, (3, 3), activation='relu')) 
 model.add(layers.GlobalMaxPooling2D()) 
 model.add(layers.Dense(2, activation='softmax')) 
 model.compile(optimizer='adam', 
               loss='sparse_categorical_crossentropy', 
               metrics=['accuracy']) 

I would like to build this model in WL. My attempt is

net = NetChain[{ConvolutionLayer[64, {3, 3}], 
   ElementwiseLayer["ReLU"], PoolingLayer[{2, 2}, "Function" -> Max], 
   ConvolutionLayer[32, {2, 2}], 
   PoolingLayer[{2, 2}, "Function" -> Max], LinearLayer[2], 
   SoftmaxLayer[]}]

I can't find the GlobalMaxPooling option, and I am uncertain if my convolution layers are set up correctly. I appreciate if anyone could show how this could be done.

Thank you,

Answer 1

GlobalMaxPooling = AggregationLayer[Max, 2]

net = NetChain[
  {
   ConvolutionLayer[64, {3, 3}],
   ElementwiseLayer[Ramp],
   PoolingLayer[{2, 2}],
   ConvolutionLayer[64, {3, 3}],
   ElementwiseLayer[Ramp],
   AggregationLayer[Max, 2],
   LinearLayer[2],
   SoftmaxLayer[]
   },
  "Input" -> {1, 132, 132},
  "Output" -> NetDecoder[{"Class", {0, 1}}]
  ]

SeedRandom[0];
n = 1024;
X = RandomInteger[{0, 1}, {n, 1, 132, 132}];
Y = RandomInteger[{0, 1}, n];

netT = NetTrain[net, X -> Y, All, MaxTrainingRounds -> 10]