Number of steps per epoch

[celsofranssa]

Some learning rate schedulers as OneCycleLR requires the number of steps per epoch.
Then, how to get the number of steps in configure_optimizers(self) scope?

[teddykoker]

Normally I pass the total number of samples in the dataset as well as the batch size into my model's __init__(). Then the number of steps per epoch can be computed with self.samples // self.batch_size.

[celsofranssa]

Currently, I am using the following (for single device):

    def num_training_steps(self) -> int:
        """Total training steps inferred from datamodule and number of epochs."""
        steps_per_epochs = len(self.train_dataloader()) / self.trainer.accumulate_grad_batches
        max_epochs = self.trainer.max_epochs
        return steps_per_epochs * max_epochs

However, maybe exist a better solution as pointed out in #5449.