Permutation Feature Importance DOC #553
Open
Conversation
This file contains hidden or 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
…50-doc-permutation-feature-importance
jpaillard
reviewed
Dec 8, 2025
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #553 +/- ##
=======================================
Coverage 98.37% 98.37%
=======================================
Files 23 23
Lines 1602 1602
=======================================
Hits 1576 1576
Misses 26 26 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
jpaillard
reviewed
Dec 8, 2025
Comment on lines
29
to
37
| Note that this method was initially introduced as the mean decrease accuracy (MDA) | ||
| by :footcite:t:`breimanRandomForests2001` for Random Forests. It was initially proposed | ||
| as an heuristic Variable Importance Measure and not as a formal estimator of a | ||
| interesting theoretical quantity. Moreover, it was shown in | ||
| :footcite:t:`benard2022SobolMDA` that PFI estimates a quantity that can be decomposed | ||
| as the sum of the Total Sobol Index (TSI) :ref:`total_sobol_index` and two extra terms | ||
| that are not significant due to correlations. Thus, the theoretical quantity estimated by PFI is | ||
| not a relevant quantity contrarily to :ref:`leave_one_covariate_out` or | ||
| :ref:`conditional_feature_importance`. |
Collaborator
There was a problem hiding this comment.
- Here, I would be a bit more direct, saying that it does not estimate any meaningful / previously studied quantity.
- Maybe we can make it a Note like for "extrapolation issues" below?
Comment on lines
45
to
48
| estimating a conditional sampler as in :ref:`conditional_feature_importance`. Since | ||
| the distribution from which we are sampling is the marginal distribution of the feature | ||
| breaking the relationship with the others, a simple permutation of the feature values | ||
| across the individuals is sufficient. Also, note that the same estimated model is |
Collaborator
There was a problem hiding this comment.
Suggested change
| estimating a conditional sampler as in :ref:`conditional_feature_importance`. Since | |
| the distribution from which we are sampling is the marginal distribution of the feature | |
| breaking the relationship with the others, a simple permutation of the feature values | |
| across the individuals is sufficient. Also, note that the same estimated model is | |
| estimating a conditional sampler as in :ref:`conditional_feature_importance`. A simple permutation | |
| of the feature values across the individuals is sufficient since the distribution from which we are sampling is the | |
| marginal distribution of the feature, thus breaking the relationship with the others. Also, note that the | |
| same estimated model is |
docs/tools/references.bib
Outdated
Comment on lines
234
to
235
| url = {https://doi.org/10.1007/s11222-021-10057-z}, | ||
| abstract = {This paper reviews and advocates against the use of permute-and-predict (PaP) methods for interpreting black box functions. Methods such as the variable importance measures proposed for random forests, partial dependence plots, and individual conditional expectation plots remain popular because they are both model-agnostic and depend only on the pre-trained model output, making them computationally efficient and widely available in software. However, numerous studies have found that these tools can produce diagnostics that are highly misleading, particularly when there is strong dependence among features. The purpose of this work is to review the growing body of literature, demonstrate these drawbacks, explain why they occur, and advocate for alternative measures involving additional modeling. In particular, breaking dependencies between features forces extrapolation into sparse regions of the feature space, over-emphasizing correlated features in both variable importance measures and partial dependence plots.} |
Collaborator
There was a problem hiding this comment.
Suggested change
| url = {https://doi.org/10.1007/s11222-021-10057-z}, | |
| abstract = {This paper reviews and advocates against the use of permute-and-predict (PaP) methods for interpreting black box functions. Methods such as the variable importance measures proposed for random forests, partial dependence plots, and individual conditional expectation plots remain popular because they are both model-agnostic and depend only on the pre-trained model output, making them computationally efficient and widely available in software. However, numerous studies have found that these tools can produce diagnostics that are highly misleading, particularly when there is strong dependence among features. The purpose of this work is to review the growing body of literature, demonstrate these drawbacks, explain why they occur, and advocate for alternative measures involving additional modeling. In particular, breaking dependencies between features forces extrapolation into sparse regions of the feature space, over-emphasizing correlated features in both variable importance measures and partial dependence plots.} |
9 tasks
…ub.com/mind-inria/hidimstat into 550-doc-permutation-feature-importance
bthirion
reviewed
Dec 10, 2025
docs/src/model_agnostic_methods/permutation_feature_importance.rst
Outdated
Show resolved
Hide resolved
hidimpy/bin/Activate.ps1
Outdated
Collaborator
There was a problem hiding this comment.
OOps, what are all these files ?
Collaborator
|
It looks almost good to me. I still have comments regarding the I see that the test are not passing but merging #558 shoud improve |
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.
No description provided.