Explainer base #38
Merged
Explainer base #38
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
|
One thing to decide: Currently, the user is supposed to pass a functional explainer to the Randomization metric. I think metrics should accept both functional and stateful explainers. (we also have an interface that makes the stateful into a functional explainer) what do you think about this issue? |
This reverts commit 888aefd.
# Conflicts: # src/explainers/aggregators.py # src/explainers/base.py # src/explainers/captum/base.py # src/explainers/captum/captum_influence.py # src/explainers/captum/similarity.py # src/explainers/functional.py # tests/explainers/test_base_explainer.py # tests/explainers/test_explainers.py # tests/explainers/test_self_influence.py
|
Dilya the PR saver!! 🙏🏼 |
dilyabareeva
approved these changes
Jun 21, 2024
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Outdated suggestions cannot be applied.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.
Hello @dilyabareeva
I have implemented the base explainer interface, and an interface for the captum base explanation class DataInfluence. It turns out that their subclasses have different names for the same parameters (module instead of model, influence_src_dataset instead of dataset etc.). CaptumSimilarityExplainer implements the Similarity explainer. Other explainers from captum should be similarly easily defined.
The base explainer's self_influence computes the full explanations for the training dataset and gets the diagonals.
IMPORTANT CHANGE: Our explainer_wrapper had a bug (torch.gather statement was not doing the correct thing). So I recreated the mnist_explanations_1 asset in the test assets folder.
I got rid of old implementations of explain_wrapper and self_influence.
I added functional versions for explainers and self_influence. These are bbuilt on a general interface funtion which takes an explainer class and makes it functional.
I also fixed the test according to all changes that I made. Randomization metric test uses the functional interface to use the stateful explainer. I seperately check that the functional, stateful and ground-truth self influences are correct. So I think all the code that I add here is covered in the tests.
Best ☮️
I think you will like most of it.