how to call train_one_epoch explicitly?

[marsggbo]

Pytorch-lightning seems to provide only fit, validate, test, and predict. How can we explicitly call a function of training only one epoch? Below is a pseudo-code

for epoch in range(10):
    if epoch <= 5:
        # do something
    else:
        # do something
    train_results = trainer.train_one_epoch(...)
    valid_results = trainer.validate(....)

[akihironitta]

@marsggbo Have you tried overriding hooks? PyTorch Lightning provides quite a number of hooks so that you can customise the default behaviour:
https://pytorch-lightning.readthedocs.io/en/stable/common/lightning_module.html#hooks

You can also utilise current_epoch property of the trainer if you want to write some logic that depends on epoch index:
https://pytorch-lightning.readthedocs.io/en/stable/common/trainer.html#current-epoch