32 lines
868 B
Markdown
32 lines
868 B
Markdown
|
To ensure you don't accidentally use test data to guide training decisions Lightning makes running the test set deliberate.
|
||
|
|
|
||
|
|
---
|
||
|
|
#### test
|
||
|
|
You have two options to run the test set.
|
||
|
|
First case is where you test right after a full training routine.
|
||
|
|
``` {.python}
|
||
|
|
# run full training
|
||
|
|
trainer.fit(model)
|
||
|
|
|
||
|
|
# run test set
|
||
|
|
trainer.test()
|
||
|
|
```
|
||
|
|
|
||
|
|
Second case is where you load a model and run the test set
|
||
|
|
```{.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 second case, the options you pass to trainer will be used when running the test set (ie: 16-bit, dp, ddp, etc...)
|
||
|
|
|