When no val dataloader is present and user implements validation_step need to throw useful error

[Ne0Ment]

I have been using pytorch before, but for performance I decided to use lightning. I rewrote my pytorch into a pl.LightningModule class. While training for the first epoch, everything seems fine, but when it starts validation there is a TypeError: 'NoneType' object is not iterable.
Epoch 1: 100%|████████████████████████| 6514/6514 [01:27<00:00, 71.17batch/s, batch_nb=6513, gpu=0, loss=1.099, v_nb=7]
Validating: 0batch [00:00, ?batch/s]

train_dataset = t.utils.data.TensorDataset(X,y)
trainloader = t.utils.data.DataLoader(train_dataset, batch_size=64)

class FastNN(pl.LightningModule):

    def __init__(self, dataload):
        super(FastNN, self).__init__()
        self.fc1 = nn.Linear(110, 1024)
        self.fc2 = nn.Linear(1024, 512)
        self.fc3 = nn.Linear(512, 256)
        self.fc4 = nn.Linear(256, 128)
        self.fc5 = nn.Linear(128, 64)
        self.fc6 = nn.Linear(64, 32)
        self.fc7 = nn.Linear(32, 3)
        self.dataloader = dataload

    def forward(self, x):
        x = f.relu(self.fc1(x))
        x = f.relu(self.fc2(x))
        x = f.relu(self.fc3(x))
        x = f.relu(self.fc4(x))
        x = f.relu(self.fc5(x))
        x = f.relu(self.fc6(x))
        x = f.relu(self.fc7(x))
        return x

    def training_step(self, batch, batch_nb):
        # REQUIRED
        x, y = batch
        y_hat = self.forward(x)
        loss = t.nn.functional.cross_entropy(y_hat, y)
        tensorboard_logs = {'train_loss': loss}
        return {'loss': loss, 'log': tensorboard_logs}

    def validation_step(self, batch, batch_nb):
        # OPTIONAL
        x, y = batch
        y_hat = self.forward(x)
        return {'val_loss': t.nn.functional.cross_entropy(y_hat, y)}

    def validation_end(self, outputs):
        # OPTIONAL
        avg_loss = t.stack([x['val_loss'] for x in outputs]).mean()
        tensorboard_logs = {'val_loss': avg_loss}
        return {'avg_val_loss': avg_loss, 'log': tensorboard_logs}

    def configure_optimizers(self):
        # REQUIRED
        # can return multiple optimizers and learning_rate schedulers
        # (LBFGS it is automatically supported, no need for closure function)
        return t.optim.Adam(self.parameters(), lr=0.02)

    @pl.data_loader
    def train_dataloader(self):
        # REQUIRED
        return self.dataloader

net = FastNN(trainloader)
trainer = pl.Trainer(gpus=1)    
trainer.fit(net)

[williamFalcon]

You have to add the val_dataloader method:

    @pl.data_loader
    def val_dataloader(self):
        # REQUIRED
        return self.other_dataloader

[williamFalcon]

@ILYEEV thanks for pointing out though. We should have a nicer user experience for this error.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[Borda]

@ILYEEV are you interested in sending PR for a better warning? maybe even do this kind of check before starting the training... something like self-diagnostic?

[awaelchli]

This is fixed on master.
If the user implements validation_step but no validation dataloader, it throws an exception:

pytorch_lightning.utilities.exceptions.MisconfigurationException: 
You have defined `validation_step()`, but have not passed in a val_dataloader().

This issue can be closed.