optimizer = optim.Adam(model.parameters())
start_time = time.time()
for epoch in range(start_epoch, epochs + 1):
train_loss, train_acc = train(epoch, model, optimizer, train_criterion,
train_loader, device)
history["train_loss"].append(train_loss)
history["train_acc"].append(train_acc)
print('epoch: ', epoch)
print('{}: loss: {:.4f} acc: {:.4f}'.format('training', train_loss,
train_acc))
valid_loss, valid_acc, valid_preds_list, valid_truelabels_list, valid_probas_list, valid_auc_score = validation(
epoch, model, optimizer, valid_criterion, valid_loader, device)
history["valid_loss"].append(valid_loss)
history["valid_acc"].append(valid_acc)
history["valid_preds_list"].append(valid_preds_list)
history["valid_truelabels_list"].append(valid_truelabels_list)
history["valid_probas_list"].append(valid_probas_list)
history["valid_auc_score"].append(valid_auc_score)
print('{}: loss: {:.4f} acc: {:.4f} auc: {:.4f}'.format(
'validation', valid_loss, valid_acc, valid_auc_score))
print()
# save models
is_best = valid_loss < best_val_loss
best_val_loss = min(valid_loss, best_val_loss)
history = { # "train_loss":[], "train_acc":[],
"test_loss": [],
"test_acc": [],
"test_preds_list": [],
"test_truelabels_list": [],
"test_probas_list": [],
"test_auc_score": []
}
optimizer = optim.Adam(model.parameters())
start_time = time.time()
for epoch in range(start_epoch, epochs + 1):
test_loss, test_acc, test_preds_list, test_truelabels_list, test_probas_list, test_auc_score = validation(
epoch, model, optimizer, test_criterion, test_loader, device)
history["test_loss"].append(test_loss)
history["test_acc"].append(test_acc)
history["test_preds_list"].append(test_preds_list)
history["test_truelabels_list"].append(test_truelabels_list)
history["test_probas_list"].append(test_probas_list)
history["test_auc_score"].append(test_auc_score)
print('{}: loss: {:.4f} acc: {:.4f} auc: {:.4f}'.format(
'test', test_loss, test_acc, test_auc_score))
print()
print('time elapsed:', time.time() - start_time)
# save results
with open("history_googlenet_test.pkl", "wb") as fout:
def train_valid(optimizer = optim.Adam(model.parameters()), epochs = 20, model = model,
train_criterion = nn.CrossEntropyLoss(), train_loader = train_loader,
valid_criterion = nn.CrossEntropyLoss(), valid_loader = valid_loader,
device = device):
start_epoch = 1
#or: best_val_acc = 0
best_val_loss = np.inf
history = {"train_loss":[], "train_acc":[],
"valid_loss":[], "valid_acc":[], "valid_preds_list":[],
"valid_truelabels_list":[], "valid_probas_list":[], "valid_auc_score":[]}
start_time = time.time()
for epoch in range(start_epoch, epochs + 1):
train_loss, train_acc = train(epoch, model, optimizer, train_criterion, train_loader, device)
history["train_loss"].append(train_loss)
history["train_acc"].append(train_acc)
print('epoch: ', epoch)
print('{}: loss: {:.4f} acc: {:.4f}'.format('training', train_loss, train_acc))
valid_loss, valid_acc, valid_preds_list, valid_truelabels_list, valid_probas_list, valid_auc_score = validation(epoch, model, optimizer, valid_criterion, valid_loader, device)
history["valid_loss"].append(valid_loss)
history["valid_acc"].append(valid_acc)
history["valid_preds_list"].append(valid_preds_list)
history["valid_truelabels_list"].append(valid_truelabels_list)
history["valid_probas_list"].append(valid_probas_list)
history["valid_auc_score"].append(valid_auc_score)
print('{}: loss: {:.4f} acc: {:.4f} auc: {:.4f}'.format('validation', valid_loss, valid_acc, valid_auc_score))
print()
# save models(use valid loss as best model criterion, please change
# criterion here if needed(eg. valid acc)
is_best = valid_loss < best_val_loss
best_val_loss = min(valid_loss, best_val_loss)
if is_best:
# please change model file path here
best_model_file = "best_models/best_model.pth"
torch.save(model.state_dict(), best_model_file)
# save model from every training epoch
# can be deleted if do not need this one, or adapt it to save 5th, 10th, 15th ...models
model_file = "models/resnet152_model_" + str(epoch) + ".pth"
torch.save(model.state_dict(), model_file)
# save training/validation results
with open("history.pkl", "wb") as fout:
pickle.dump(history, fout)
print('time elapsed:', time.time() - start_time)
return history
for epoch in range(start_epoch, epochs + 1):
train_loss, train_acc = train(epoch, model, optimizer, train_criterion,
train_loader, device)
history["train_loss"].append(train_loss)
history["train_acc"].append(train_acc)
print('epoch: ', epoch)
print('{}: loss: {:.4f} acc: {:.4f}'.format('training', train_loss,
train_acc))
valid_loss, valid_acc, valid_preds_list, valid_truelabels_list, valid_probas_list = validation(
epoch,
model,
optimizer,
valid_criterion,
valid_loader,
device,
multiclass=True)
history["valid_loss"].append(valid_loss)
history["valid_acc"].append(valid_acc)
history["valid_preds_list"].append(valid_preds_list)
history["valid_truelabels_list"].append(valid_truelabels_list)
history["valid_probas_list"].append(valid_probas_list)
#history["valid_auc_score"].append(valid_auc_score)
print('{}: loss: {:.4f} acc: {:.4f} auc: {:.4f}'.format(
'validation', valid_loss, valid_acc, valid_auc_score))
print()
# save models