for i, batch in enumerate(test_batches):
(id_slice, padded_y_slice, y_slice_lengths, x_slice) = batch
test_loss, test_predictions = model.forward(padded_y_slice,
y_slice_lengths, x_slice)
test_predictions = torch.cat(test_predictions, 0)
test_predictions = test_predictions.transpose(0, 1)
references = padded_y_slice.transpose(0, 1)
color_vectors = x_slice.data.cpu().numpy()[0] # a bit ugly
for test_prediction, reference, color in zip(test_predictions,
references,
color_vectors):
comparison.append(([
v.string for v in vocab.indices2tokens(test_prediction.data)
], [v.string for v in vocab.indices2tokens(reference.data)],
[str(ck) for ck in color]))
total_batches_test += 1
total_loss_test += test_loss.data[0]
if i % log_interval == 0 and i > 0:
writer.add_scalar('test_loss_per_batch',
total_loss_test / total_batches_test,
i + delta_test)
total_loss_test = 0
total_batches_test = 0
delta_test = delta_test + len(test_batches)
with open(output_path + str(epoch) + ".txt", 'w') as f:
