Test set

Lightning forces the user to run the test set separately to make sure it isn’t evaluated by mistake

Test after fit

To run the test set after training completes, use this method

# run full training
trainer.fit(model)

# run test set
trainer.test()

Test pre-trained model

To run the test set on a pre-trained model, use this method.

model = MyLightningModule.load_from_checkpoint(
    checkpoint_path='/path/to/pytorch_checkpoint.ckpt',
    hparams_file='/path/to/test_tube/experiment/version/hparams.yaml',
    map_location=None
)

# init trainer with whatever options
trainer = Trainer(...)

# test (pass in the model)
trainer.test(model)

In this case, the options you pass to trainer will be used when running the test set (ie: 16-bit, dp, ddp, etc…)