cross_entropy = cross_entropy.data.cpu().numpy()
kld = kld.data.cpu().numpy()
print('\n')
print('------------VALID-------------')
print('--------CROSS-ENTROPY---------')
print(cross_entropy)
print('-------------KLD--------------')
print(kld)
print('------------------------------')
ce_result += [cross_entropy]
kld_result += [kld]
if iteration % 20 == 0:
seed = np.random.normal(size=[1, parameters.latent_variable_size])
sample = rvae.sample(batch_loader, 50, seed, args.use_cuda)
print('\n')
print('------------SAMPLE------------')
print('------------------------------')
print(sample)
print('------------------------------')
t.save(rvae.state_dict(), 'trained_RVAE')
np.save('ce_result_{}.npy'.format(args.ce_result), np.array(ce_result))
np.save('kld_result_npy_{}'.format(args.kld_result), np.array(kld_result))
print('dropout = ', args.dropout)
for i in range(3):
target_sentence, predicted_sentence = validation_sample(
args.use_cuda)
print(' target : ', target_sentence)
print('sample : ', predicted_sentence)
print('------------------------------')
ce_result += [cross_entropy]
kld_result += [kld]
# generate sample
if iteration % 300 == 0:
source = 'she should control the speed of her car'
result = rvae.conditioned_sample(source, batch_loader, args)
print('\n')
print('------------SAMPLE------------')
print('------------------------------')
print('source : ', source)
print('sample : ', result)
print('------------------------------')
# save model
if iteration % 1000 == 0 or iteration == (args.num_iterations - 1):
t.save(rvae.state_dict(),
'saved_models/trained_RVAE_' + args.model_name)
np.save('saved_models/ce_result_{}.npy'.format(args.model_name),
np.array(ce_result))
np.save('saved_models/kld_result_npy_{}'.format(args.model_name),
np.array(kld_result))
print('-------------KLD--------------')
print(kld)
print('------------------------------')
ce_result += [cross_entropy]
kld_result += [kld]
if iteration % 20 == 0:
seed = np.random.normal(size=[1, parameters.latent_variable_size])
sample = rvae.sample(batch_loader, 50, seed, args.use_cuda)
print('\n')
print('------------SAMPLE------------')
print('------------------------------')
print(sample)
print('------------------------------')
if iteration % 25000 == 0:
t.save(
rvae.state_dict(),
'./trained_model/{}_trained_{}'.format(args.train_data,
iteration))
t.save(
optimizer.state_dict(),
'./trained_model/{}_trained_optimizer_{}'.format(
args.train_data, iteration))
# np.save('ce_result_{}.npy'.format(args.ce_result), np.array(ce_result))
# np.save('kld_result_npy_{}'.format(args.kld_result), np.array(kld_result))
print('------------SAMPLE------------')
print('------------------------------')
print(f'original: {sentence.encode("utf-8")}')
print(f'reference: {reference.encode("utf-8")}')
print(f'generated: {result.encode("utf-8")}')
print('------------------------------')
bar.finish()
if iteration % 10 == 0:
print('\n')
print('------------TRAIN-------------')
print('----------ITERATION-----------')
print(iteration)
print('--------CROSS-ENTROPY---------')
print(cross_entropy.data.cpu().numpy())
print('-------------KLD--------------')
print(kld.data.cpu().numpy())
print('-----------KLD-coef-----------')
print(coef)
print('------------------------------')
print("--------------------saving checkpoint-----------------------")
t.save(rvae.state_dict(), 'trained_RVAE_checkpoint_para_out')
print("--------------------saved checkpoint-----------------------")
print("\n\n\n")
t.save(rvae.state_dict(), 'trained_RVAE')
np.save('ce_result_{}.npy'.format(args.ce_result), np.array(ce_result))
np.save('kld_result_npy_{}'.format(args.kld_result), np.array(kld_result))