Write predictions in LightningModule instead of EvalResult #3882
Conversation
| def write_prediction(self, name, value, filename='predictions.pt'): | ||
| self.trainer.evaluation_loop.predictions._add_prediction(name, value, filename) | ||
|
|
||
| def write_prediction_dict(self, predictions_dict, filename='predictions.pt'): | ||
| for k, v in predictions_dict.items(): | ||
| self.write_prediction(k, v, filename) | ||
|
|
There was a problem hiding this comment.
does this write out as a CSV file? do you think this makes more sense as a callback?
There was a problem hiding this comment.
writes to list of dictionaries via torch.save. making predictions is different for every use case, so this way lets user define how they want to do it.
Could be achieved with callback (in fact, that's how I originally was doing this), but that would only work for 1 use case.
Codecov Report
@@ Coverage Diff @@
## master #3882 +/- ##
=======================================
+ Coverage 84% 87% +4%
=======================================
Files 119 117 -2
Lines 9764 9155 -609
=======================================
- Hits 8169 8002 -167
+ Misses 1595 1153 -442 |
|
@nateraw please use fsspec here: https://github.com/PyTorchLightning/pytorch-lightning/blob/cea5f1f53876399dfaa0d37accdc527af7ca39af/pytorch_lightning/trainer/supporters.py#L138-L165 so we can save files beyond local disk |
What does this PR do?
Fixes #3870
Before submitting
PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
Did you have fun?
Make sure you had fun coding 🙃