ENH: ReduceLROnPlateau records the learning rate and works on batches #1075
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Previously, when using ReduceLROnPlateau, we would not record the learning rates in history. The comment says that's because this class does not expose the get_last_lr method. I checked it again and it's now present, so let's use it. Furthermore, I made a change to enable ReduceLROnPlateau to step on each batch instead of each epoch. This is consistent with other learning rate schedulers.
|
The CI is failing because of a change in sklearn 1.6 and requires PR #1076. |
githubnemo
approved these changes
Dec 19, 2024
BenjaminBossan
deleted the
enh-reduce-lr-on-plateau-recording-and-per-batch
branch
githubnemo
added a commit
that referenced
this pull request
Jan 9, 2025
Please welcome skorch 1.1.0 - a smaller release with a few fixes, a new notebook showcasing learning rate schedulers and mainly support for scikit-learn 1.6.0. Full list of changes: ### Added - Added a [notebook](https://github.com/skorch-dev/skorch/blob/master/notebooks/Learning_Rate_Scheduler.ipynb) that shows how to use Learning Rate Scheduler in skorch.(#1074) ### Changed - All neural net classes now inherit from sklearn's [`BaseEstimator`](https://scikit-learn.org/stable/modules/generated/sklearn.base.BaseEstimator.html). This is to support compatibility with sklearn 1.6.0 and above. Classification models additionally inherit from [`ClassifierMixin`](https://scikit-learn.org/stable/modules/generated/sklearn.base.ClassifierMixin.html) and regressors from [`RegressorMixin`](https://scikit-learn.org/stable/modules/generated/sklearn.base.RegressorMixin.html). (#1078) - When using the `ReduceLROnPlateau` learning rate scheduler, we now record the learning rate in the net history (`net.history[:, 'event_lr']` by default). It is now also possible to to step per batch, not only by epoch (#1075) - The learning rate scheduler `.simulate()` method now supports adding step args which is useful when simulation policies such as `ReduceLROnPlateau` which expect metrics to base their schedule on. (#1077) - Removed deprecated `skorch.callbacks.scoring.cache_net_infer` (#1088) ### Fixed - Fix an issue with using `NeuralNetBinaryClassifier` with `torch.compile` (#1058)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previously, when using
ReduceLROnPlateau, we would not record the learning rates in history. The code comment says that's because this class does not expose theget_last_lrmethod. I checked it again and it's now present, so let's use it.Furthermore, I made a change to enable
ReduceLROnPlateauto step on each batch (so far, only each epoch was supported). This is consistent with other learning rate schedulers.