t_d = t_d.cuda()
t_x = t_x.view(-1, 10, 8)
test_output = rnn(t_x) # (samples, time_step, input_size)
if gpu_avaliable:
pred_y = torch.max(test_output, 1)[1].cuda().data
else:
pred_y = torch.max(test_output, 1)[1].data.numpy()
# t_y = t_y.data.numpy()
all_pred_y.extend(pred_y)
all_test_y.extend(list(t_y.data.cpu().numpy()))
all_test_d.extend(list(t_d.data.cpu().numpy()))
# accuracy = torch.sum(torch.LongTensor(all_pred_y) == torch.LongTensor(all_test_y)).type(torch.FloatTensor) / len(all_test_y)
# print_out = 'Epoch: ' + str(epoch) + '| train loss: %.4f' % loss.data.cpu().numpy() + '| test accuracy: %.4f' % accuracy
print_out = 'Epoch: ' + str(epoch) + '| train loss: %.4f' % loss.data.cpu().numpy() + \
'| test accuracy: %.4f' % Evaluate.accuracy(all_pred_y, all_test_y) + \
'| test MAE: %.4f' % Evaluate.region_MAE(all_pred_y, all_test_d) + \
'| test RMSE: %.4f' % Evaluate.region_RMSE(all_pred_y, all_test_d)
print(print_out)
elogger.log(str(print_out))
torch.save(rnn.state_dict(), 'cnn_lstm_prediction_label_is_region.pkl')
# print 10 predictions from test data
# test_output = rnn(test_data[:10].view(-1, 10, 5))
# if gpu_avaliable:
# pred_y = torch.max(test_output, 1)[1].cuda().data
# else:
# pred_y = torch.max(test_output, 1)[1].data.numpy()
# print(pred_y, 'prediction number')
# print(test_labels[:10], 'real number')
all_test_y = []
for t_step, (t_x, t_y) in enumerate(test_loader):
if gpu_avaliable:
t_x = t_x.cuda()
t_y = t_y.cuda()
t_x = t_x.view(-1, 10, 5)
test_output = rnn(t_x) # (samples, time_step, input_size)
if gpu_avaliable:
pred_y = torch.max(test_output, 1)[1].cuda().data
else:
pred_y = torch.max(test_output, 1)[1].data.numpy()
all_pred_y.extend(pred_y)
all_test_y.extend(list(t_y.data.cpu().numpy()))
# accuracy = torch.sum(torch.LongTensor(all_pred_y) == torch.LongTensor(all_test_y)).type(torch.FloatTensor) / len(all_test_y)
# print_out = 'Epoch: ' + str(epoch) + '| train loss: %.4f' % loss.data.cpu().numpy() + '| test accuracy: %.4f' % accuracy
print_out = 'Epoch: ' + str(epoch) + '| train loss: %.4f' % loss.data.cpu().numpy() + \
'| test accuracy: %.4f' % Evaluate.accuracy(all_pred_y, all_test_y) + \
'| test MAE: %.4f' % Evaluate.MAE(all_pred_y, all_test_y) + \
'| test RMSE: %.4f' % Evaluate.RMSE(all_pred_y, all_test_y)
print(print_out)
elogger.log(str(print_out))
# print 10 predictions from test data
# test_output = rnn(test_data[:10].view(-1, 10, 5))
# if gpu_avaliable:
# pred_y = torch.max(test_output, 1)[1].cuda().data
# else:
# pred_y = torch.max(test_output, 1)[1].data.numpy()
# print(pred_y, 'prediction number')
# print(test_labels[:10], 'real number')