39 lines
931 B
ReStructuredText
39 lines
931 B
ReStructuredText
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
|
|
|
|
.. code-block:: python
|
|
|
|
# run full training
|
|
trainer.fit(model)
|
|
|
|
# run test set
|
|
trainer.test()
|
|
|
|
|
|
Test pre-trained model
|
|
----------------------
|
|
To run the test set on a pretrained model, use this method.
|
|
|
|
.. code-block:: python
|
|
|
|
model = MyLightningModule.load_from_metrics(
|
|
weights_path='/path/to/pytorch_checkpoint.ckpt',
|
|
tags_csv='/path/to/test_tube/experiment/version/meta_tags.csv',
|
|
on_gpu=True,
|
|
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... |