target_variable = training_pair[1]
loss = train(input_variable, target_variable, encoder, decoder,
encoder_optimizer, decoder_optimizer, criterion, device)
print_loss_total += loss
plot_loss_total += loss
if epoch % print_every == 0:
print_loss_avg = print_loss_total / print_every
print_loss_total = 0
time_since = helpers.time_since(start, epoch / n_epochs)
print('%s (%d %d%%) %.4f' %
(time_since, epoch, epoch / n_epochs * 100, print_loss_avg))
if epoch % 100 == 0:
model_out_path = "checkpoint/" + "params_epoch_{}.tar".format(epoch)
if not os.path.exists("checkpoint/"):
os.makedirs("checkpoint/")
print("세이브 시작")
torch.save(
{
'epoch': epoch,
'encoder': encoder.state_dict(),
'decoder': decoder.state_dict(),
'encoder_optim': encoder_optimizer.state_dict(),
'decoder_optim': decoder_optimizer.state_dict(),
'decoder.attention': decoder.attention.state_dict()
}, model_out_path)
print("세이브 끝")
helpers.show_plot(plot_losses)
# Keep track of loss
print_loss_total += loss
plot_loss_total += loss
if epoch == 0:
continue
if epoch % print_every == 0:
print_loss_avg = print_loss_total / print_every
print_loss_total = 0
time_since = helpers.time_since(start, epoch / n_epochs)
print('%s (%d %d%%) %.4f' %
(time_since, epoch, epoch / n_epochs * 100, print_loss_avg))
if epoch % plot_every == 0:
plot_loss_avg = plot_loss_total / plot_every
plot_losses.append(plot_loss_avg)
plot_loss_total = 0
# Save our models
torch.save(encoder.state_dict(),
'../data/encoder_params_{}'.format(args.language))
torch.save(decoder.state_dict(),
'../data/decoder_params_{}'.format(args.language))
torch.save(decoder.attention.state_dict(),
'../data/attention_params_{}'.format(args.language))
# Plot loss
helpers.show_plot(plot_losses)
print_loss_total += loss
plot_loss_total += loss
if epoch == 0:
continue
if epoch % print_every == 0:
print_loss_avg = print_loss_total / print_every
print_loss_total = 0
time_since = helpers.time_since(start, epoch / n_epochs)
print('%s (%d %d%%) %.4f' % (time_since, epoch, epoch / n_epochs * 100, print_loss_avg))
if (epoch / n_epochs * 100) % 5 == 0 and epoch > 100 :
# if epoch == 30:
torch.save(encoder.state_dict(), './data/encoder_params_{}'.format(language))
torch.save(decoder.state_dict(), './data/decoder_params_{}'.format(language))
torch.save(decoder.attention.state_dict(), './data/attention_params_{}'.format(language))
exec(open("eval.py").read())
if epoch % plot_every == 0:
plot_loss_avg = plot_loss_total / plot_every
plot_losses.append(plot_loss_avg)
plot_loss_total = 0
# Save models
torch.save(encoder.state_dict(), './data/encoder_params_{}'.format(language))
torch.save(decoder.state_dict(), './data/decoder_params_{}'.format(language))
torch.save(decoder.attention.state_dict(), './data/attention_params_{}'.format(language))