if epoch < 20 or epoch % 200 == 0:
print(
"Epoch {}: Loss: lc {:.3f}, train_B {:.3f}, val_B {:.3f}, train_I {:.3f}, val_I {:.3f}, \n\t\t Acc: train_B {:.3f}, val_B {:.3f}, train_I {:.3f}, val_I {:.3f}"
.format(epoch, loss_C, train_loss_B, val_loss_B, train_loss_I,
val_loss_I, train_accuracy_B, val_accuracy_B,
train_accuracy_I, val_accuracy_I))
print("-" * 20)
# choose model
# TODO: not save at the same time, may have bad common representation
if max_val_acc_B <= val_accuracy_B:
model_dir = logDir + model_name_B + '.pt'
print("Saving model at {} epoch to {}".format(epoch, model_dir))
max_val_acc_B = val_accuracy_B
torch.save(model_B.state_dict(), model_dir)
if max_val_acc_I <= val_accuracy_I:
model_dir = logDir + model_name_I + '.pt'
print("Saving model at {} epoch to {}".format(epoch, model_dir))
max_val_acc_I = val_accuracy_I
torch.save(model_I.state_dict(), model_dir)
training_end = datetime.now()
training_time = training_end - training_start
print("RAE training takes time {}".format(training_time))
model_B.is_fitted = True
model_I.is_fitted = True
model_B.eval()
val_accuracy = correct / val_num# / len(val_loader.dataset)
val_loss /= val_num #len(val_loader.dataset)
epoch_val_loss.append(val_loss) # only save the last batch
epoch_val_acc.append(val_accuracy)
if epoch < 20 or epoch%200 == 0:
print("train_num {}, val_num {}".format(train_num, val_num))
print('Epoch: {} Loss: train {:.3f}, valid {:.3f}. Accuracy: train: {:.3f}, valid {:.3f}'.format(epoch, train_loss, val_loss, train_accuracy, val_accuracy))
# choose model
if max_val_acc <= val_accuracy:
model_dir = logDir + model_name + '.pt'
print('Saving model at {} epoch to{}'.format(epoch, model_dir))
max_val_acc = val_accuracy
torch.save(model.state_dict(), model_dir)
training_end = datetime.now()
training_time = training_end -training_start
print("training takes time {}".format(training_time))
# In[9]:
model.is_fitted = True
model.eval()
# In[10]:
epoch_val_loss.append(val_loss) # only save the last batch
epoch_val_acc.append(val_accuracy)
if epoch < 20 or epoch % 200 == 0:
print("train_num {}, val_num {}".format(train_num, val_num))
print(
'Epoch: {} Loss: train {}, valid {}. Accuracy: train: {}, valid {}'
.format(epoch, train_loss, val_loss, train_accuracy, val_accuracy))
print("-" * 20)
# choose model
if max_val_acc <= val_accuracy:
print('Saving model at ', epoch, ' epoch')
max_val_acc = val_accuracy
torch.save(model.state_dict(), logDir + model_name + '.pt')
training_end = datetime.now()
training_time = training_end - training_start
print("training takes time {}".format(training_time))
model.is_fitted = True
# testing set check
net_trained = VRAEC(num_class=num_class,
sequence_length=sequence_length,
number_of_features=number_of_features,
hidden_size=hidden_size,
hidden_layer_depth=hidden_layer_depth,
latent_length=latent_length,
batch_size=batch_size,
if epoch < 20 or epoch % 200 == 0:
# print("train_num {}, val_num {}".format(train_num, val_num))
print(
"Epoch {}: Loss: lc {:.3f}, train_B {:.3f}, val_B {:.3f}, train_I{:.3f}, val_I{:.3f}, \n\t\t Acc: train_B {:.3f}, val_B {:.3f}, train_I {:.3f}, val_I {:.3f}"
.format(epoch, loss_C, train_loss_B, val_loss_B, train_loss_I,
val_loss_I, train_accuracy_B, val_accuracy_B,
train_accuracy_I, val_accuracy_I))
print("-" * 20)
# choose model
# TODO: not save at the same time, may have bad common representation
if max_val_acc_B <= val_accuracy_B:
model_dir = logDir + model_name_B + '.pt'
print("Saving model at {} epoch to {}".format(epoch, model_dir))
max_val_acc_B = val_accuracy_B
torch.save(model_B.state_dict(), model_dir)
if max_val_acc_I <= val_accuracy_I:
model_dir = logDir + model_name_I + '.pt'
print("Saving model at {} epoch to {}".format(epoch, model_dir))
max_val_acc_I = val_accuracy_I
torch.save(model_I.state_dict(), model_dir)
training_end = datetime.now()
training_time = training_end - training_start
print("RAE training takes time {}".format(training_time))
# In[ ]:
model_B.is_fitted = True
model_I.is_fitted = True
epoch_val_loss_C.append(val_loss_C)
epoch_val_loss_tot.append(val_loss_tot)
if epoch < 20 or epoch % 200 == 0:
print(
"Epoch {}: Loss: lc {:.3f}, train_B{:.3f}, val_B {:.3f}, \n\t Acc: train_B {:.3f}, val_B {:.3f}, train_I {:.3f}, val_I {:.3f}"
.format(epoch, loss_C, train_loss_B, val_loss_B, train_accuracy_B,
val_accuracy_B, train_accuracy_I, val_accuracy_I))
print("-" * 20)
# choose model
if max_val_acc_B <= val_accuracy_B:
model_dir = logDir + model_name_B + '.pt'
print("Saving model at {} epoch to {}".format(epoch, model_dir))
max_val_acc_B = val_accuracy_B
torch.save(model_B.state_dict(), model_dir)
training_end = datetime.now()
training_time = training_end - training_start
print("RAE training takes time {}".format(training_time))
model_B.is_fitted = True
model_I_pretrained.is_fitted = True
model_B.eval()
model_I_pretrained.eval()
# copy the classifier from B to I to examine the classification result
classifier_keys = ['classifier.0.weight', 'classifier.0.bias']
classifier_dict_B = {
k: v
epoch_val_loss_I.append(val_loss_I)
epoch_val_acc_I.append(val_accuracy_I)
epoch_val_loss_C.append(val_loss_C)
epoch_val_loss_tot.append(val_loss_tot)
if epoch < 20 or epoch%200 == 0:
print("Epoch {}: Loss: lc {:.3f}, train_I {:.3f}, val_I {:.3f}, \n\t Acc: train_B {:.3f}, val_B {:.3f}, train_I {:.3f}, val_I {:.3f}" .format(epoch, loss_C, train_loss_I, val_loss_I, train_accuracy_B, val_accuracy_B, train_accuracy_I, val_accuracy_I))
print("-"*20)
# choose model
if max_val_acc_I <= val_accuracy_I:
model_dir = logDir + model_name_I + '.pt'
print("Saving model at {} epoch to {}".format(epoch, model_dir))
max_val_acc_I = val_accuracy_I
torch.save(model_I.state_dict(), model_dir)
training_end = datetime.now()
training_time = training_end -training_start
print("RAE training takes time {}".format(training_time))
model_B_pretrained.is_fitted = True
model_I.is_fitted = True
model_B_pretrained.eval()
model_I.eval()
# copy the classifier from B to I to examine the classification result
classifier_keys = ['classifier.0.weight', 'classifier.0.bias']
classifier_dict_B = {k: v for k, v in model_B_pretrained.state_dict().items() if k in classifier_keys}