crlapi/sl/clmodels/finetune_grow.py [86:125]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            model.train()

            #Training loop
            for i, (raw_x, y) in enumerate(training_loader):
                raw_x, y = raw_x.to(device), y.to(device)

                # apply transformations
                x = train_aug(raw_x)

                predicted=model(x)
                loss=F.cross_entropy(predicted,y)
                nb_ok=predicted.max(1)[1].eq(y).float().sum().item()
                accuracy=nb_ok/x.size()[0]
                logger.add_scalar("train/loss",loss.item(),iteration)
                logger.add_scalar("train/accuracy",accuracy,iteration)

                optimizer.zero_grad()
                loss.backward()
                optimizer.step()

                iteration += 1
                n_fwd_samples += x.size(0)

            #Validation
            out=self._validation_loop(model,device,validation_loader)
            validation_loss,validation_accuracy=out["loss"],out["accuracy"]

            logger.add_scalar("validation/loss",validation_loss,epoch)
            logger.add_scalar("validation/accuracy",validation_accuracy,epoch)

            # Right now CV against accuracy
            # if best_loss is None or validation_loss < (best_loss - patience_delta):
            if best_acc is None or validation_accuracy > (best_acc + patience_delta):
                print("\tFound best model at epoch ",epoch)
                best_model.load_state_dict(_state_dict(model,"cpu"))
                best_loss = validation_loss
                best_acc  = validation_accuracy
                patience_count = 0
            else:
                patience_count += 1
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



crlapi/sl/clmodels/firefly.py [111:150]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            model.train()

            #Training loop
            for i, (raw_x, y) in enumerate(training_loader):
                raw_x, y = raw_x.to(device), y.to(device)

                # apply transformations
                x = train_aug(raw_x)

                predicted=model(x)
                loss=F.cross_entropy(predicted,y)
                nb_ok=predicted.max(1)[1].eq(y).float().sum().item()
                accuracy=nb_ok/x.size()[0]
                logger.add_scalar("train/loss",loss.item(),iteration)
                logger.add_scalar("train/accuracy",accuracy,iteration)

                optimizer.zero_grad()
                loss.backward()
                optimizer.step()

                iteration += 1
                n_fwd_samples += x.size(0)

            #Validation
            out=self._validation_loop(model,device,validation_loader)
            validation_loss,validation_accuracy=out["loss"],out["accuracy"]

            logger.add_scalar("validation/loss",validation_loss,epoch)
            logger.add_scalar("validation/accuracy",validation_accuracy,epoch)

            # Right now CV against accuracy
            # if best_loss is None or validation_loss < (best_loss - patience_delta):
            if best_acc is None or validation_accuracy > (best_acc + patience_delta):
                print("\tFound best model at epoch ",epoch)
                best_model.load_state_dict(_state_dict(model,"cpu"))
                best_loss = validation_loss
                best_acc  = validation_accuracy
                patience_count = 0
            else:
                patience_count += 1
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -