feat(doctest): precision metric doctest #2306
Conversation
|
@ydcjeff could you please review this PR |
|
Code looks good. @halsawadi Can you change the description of the PR to something like Thanks. |
| def thresholded_output_transform(engine, output): | ||
| y_pred, y = output | ||
| y_pred = torch.round(y_pred) | ||
| return y_pred, y | ||
| engine = Engine(thresholded_output_transform) | ||
| metric = Precision() |
There was a problem hiding this comment.
I'm not sure if this is what we want to expose to our users. thresholded_output_transform should remain the argument to Precision and not for Engine.
There was a problem hiding this comment.
Actually that should be process_function.
There was a problem hiding this comment.
The example shows how to binarize predictions for Precision metric only. Returning binarized preds from validator may be a limitation...
There was a problem hiding this comment.
In general we can improve the docstring for Precision (same for Accuracy and this comment: #2287 (comment)) and show all possible cases:
binary case
multiclass case
multilabel case
|
|
||
| def thresholded_output_transform(output): | ||
| def thresholded_output_transform(engine, output): |
There was a problem hiding this comment.
| def thresholded_output_transform(engine, output): | |
| def process_fn(engine, output): |
| y_pred, y = output | ||
| y_pred = torch.round(y_pred) | ||
| return y_pred, y | ||
| engine = Engine(thresholded_output_transform) |
There was a problem hiding this comment.
| engine = Engine(thresholded_output_transform) | |
| engine = Engine(process_fn) |
|
@halsawadi Thank you for your work. Are you still working on this ? |
|
@halsawadi Thank you for your help. I close this PR since the doctest for Precision has been merged. |
Fixes part of #2265
Description:
Add a doctest for Precision in the metrics module.
Check list: