hi, how to plot different(e.g. training acc and val acc) in tensorboard in same window?

[JimXu1989]

Hi, thanks for your great work!

I want plot training accuracy and validation accuracy in same tensorboard, so

I can realize overfiting very convient

Thanks a lot!!

[SkafteNicki]

you can achieve this by calling the add_scalars method of the underlying summarywriter:

    def training_step(self, batch, batch_idx):
        x, y = batch
        y_hat = self(x)
        loss = F.cross_entropy(y_hat, y)
        # dummy example
        self.logger.experiment.add_scalars('losses', {'train': loss, 'val': loss+torch.randn_like(loss)},
                                           global_step = self.global_step)
        return loss

this will create two curves in the same window:

however, do note that it will mess with your runs (left side of the image) because it is essentially adding a new run/tensorboard file per scalar.