def get_loss(self, y_pred, y_true, X=None, training=False):
y_true = to_var(y_true, use_cuda=False)
loss_a = torch.abs(y_true.float() - y_pred[:, 1]).mean()
loss_b = ((y_true.float() - y_pred[:, 1])**2).mean()
if training:
self.history.record_batch('loss_a', to_numpy(loss_a)[0])
self.history.record_batch('loss_b', to_numpy(loss_b)[0])
return loss_a + loss_b
def test_dropout(self, net_fit, data):
# Note: does not test that dropout is really active during
# training.
X = data[0]
# check that dropout not active by default
y_proba = to_numpy(net_fit.forward(X))
y_proba2 = to_numpy(net_fit.forward(X))
assert np.allclose(y_proba, y_proba2, rtol=1e-7)
# check that dropout can be activated
y_proba = to_numpy(net_fit.forward(X, training=True))
y_proba2 = to_numpy(net_fit.forward(X, training=True))
assert not np.allclose(y_proba, y_proba2, rtol=1e-7)