def eval_model(epoch):
correct = 0
total = 0
loss = 0
for images, labels in test_loader:
images = Variable(images)
labels_var = Variable(labels)
outputs = model(images)
predicted = np.argmax(outputs.data, 1)
loss += F.cross_entropy(outputs, labels_var, size_average=False).data
total += labels.shape[0]
correct += (predicted == labels).sum()
print('\rEpoch [{}/{}], Test Accuracy: {}% Loss: {:.4f}'.format(
epoch + 1, num_epochs, 100 * correct / total, loss / total))
images = Variable(images)
labels_var = Variable(labels)
outputs = model(images)
predicted = np.argmax(outputs.data, 1)
loss += F.cross_entropy(outputs, labels_var, size_average=False).data
total += labels.shape[0]
correct += (predicted == labels).sum()
print('\rEpoch [{}/{}], Test Accuracy: {}% Loss: {:.4f}'.format(
epoch + 1, num_epochs, 100 * correct / total, loss / total))
for epoch in range(num_epochs):
model.train()
for i, (images, labels) in enumerate(train_loader):
images = Variable(images, requires_grad=True)
labels = Variable(labels)
# Forward pass
outputs = model(images)
loss = F.cross_entropy(outputs, labels)
# Backward and optimize
optimizer.zero_grad()
loss.backward()
optimizer.step()
print('\rEpoch [{}/{}], Step [{}/{}], Loss: {:.4f}'.format(
epoch + 1, num_epochs, i + 1, len(train_loader), loss.data),
end=' ')
eval_model(epoch)
import bintorch from bintorch.autograd import Variable import bintorch.nn.functional as F import autograd.numpy as np target = Variable(np.array((1, 3, 4, 3, 3)), requires_grad=False) y = Variable(np.zeros((5, 5)), requires_grad=False) l = Variable(np.ones((5, 5)), requires_grad=True) m = Variable(np.ones((5, 5)), requires_grad=True) x = l + m + y x = F.cross_entropy(x, target) x.backward() print(x.data) print(l.grad)