def test(batch_size=200):
model.eval()
test_loss = []
idx = np.random.permutation(x_test.shape[0])
count = 0
while count < x_test.shape[0] / batch_size:
batch_index = idx[count * batch_size:(count + 1) * batch_size]
loss = dsn_loss(model(x_test[batch_index, :]), label_test[batch_index],
event_test[batch_index])
test_loss += [loss.data]
count += 1
return test_loss[:-1]
def train(batch_size=100):
model.train()
train_loss = []
idx = np.random.permutation(x_train.shape[0])
count = 0
while count < len(idx) / batch_size:
batch_index = idx[count * batch_size:(count + 1) * batch_size]
optimizer.zero_grad()
loss = dsn_loss(model(x_train[batch_index, :]),
label_train[batch_index], event_train[batch_index])
loss.backward()
train_loss += [loss.data]
optimizer.step()
count += 1
return train_loss[:-1]