PyTorch 0.4 Hint Sheet Cheat Sheet (DRAFT) by jwhitt

<font size="3­"­>
import torch.nn as nn

class Classifier(nn.Module):
    def __init__(self, num_features, num_classes):
        super(Classifier, self).__init__()
        
        self.fully_connected_1 = nn.Linear(num_features, 4)
        self.fully_connected_2 = nn.Linear(4, num_classes)
        
    def forward(self, x):
        x = F.relu(self.fully_connected_1(x))
        x = self.fully_connected_2(x)

        return x

# Instantiate the model
classifier = Classifier(4, 2)

print(f"classifer.train: {classifier.training}") # Should be true until test time.
print(classifier)  # print the layers of the model
</­fon­t>

import torch.optim as optim

LEARNING_RATE = 0.01

minimizer = optim.SGD(classifier, lr=LEARNING_RATE)

# Each epoch is a full run through all examples.
for epoch in range(NUM_EPOCHS):

    # Iterate through the data, one batch at a time.
    for i, (X_data, y_labels) in enumerate(train_loader):
        # Zero out the gradients from previous loop.
        minimizer.zero_grad()   
        
        # Apply the classifier(model) to a batch of examples, 
        # get scores.
        y_scores = classifier(X_data) # Predict.

        # Calculate the classifier's error from labels
        loss = cross_entropy_loss(y_scores, y_labels)

        # Calculate the gradients with respect to the
        # model's weights.
        loss.backward()

        # Update the model's weights.
        minimizer.step()