Skip to content

331 adding faith banzhaf approximator#333

Merged
mmschlk merged 23 commits intomainfrom
331-adding-faith-banzhaf-approximator
Mar 20, 2025
Merged

331 adding faith banzhaf approximator#333
mmschlk merged 23 commits intomainfrom
331-adding-faith-banzhaf-approximator

Conversation

@Advueu963
Copy link
Copy Markdown
Collaborator

This PR closes #331, adding the Faithful Banzhaf Interaction Index (FBII) as supported variant for Least Squares Approximation.

For this, the RegressionFBII(..) was introduced to approximate FBII.
Additionally, FBII support was introduced in TabularExplainer and MonteCarlo-Approximator.

Also, the BV exact computation was updated to set the empty players value equal to zero and game normalization for FBII.

@Advueu963 Advueu963 requested a review from FFmgll March 14, 2025 16:09
@Advueu963 Advueu963 self-assigned this Mar 14, 2025
@Advueu963 Advueu963 linked an issue Mar 14, 2025 that may be closed by this pull request
@mmschlk
Copy link
Copy Markdown
Owner

mmschlk commented Mar 15, 2025

@Advueu963 is a 🚀! Will also take a look at it coming days. Is FBII also added to ExactComputer and the MoebiusConverter? I forgot to list this as possible places in the library. Plus, could we also add it to the TreeExplainer?

@Advueu963
Copy link
Copy Markdown
Collaborator Author

@mmschlk FBII was already supported by ExactComputer and MoebiusConverter.
I will look into the suppport for TreeExplainer as this depends on TreeShapIQ.

mmschlk and others added 4 commits March 17, 2025 13:02
# Conflicts:
#	shapiq/approximator/regression/__init__.py
#	shapiq/approximator/regression/_base.py
…baseline_empty_player

# Conflicts:
#	shapiq/approximator/_base.py
#	shapiq/game_theory/indices.py
@Advueu963
Copy link
Copy Markdown
Collaborator Author

The updates consist of equalizing the baseline value and empty player value in the Interactionvalues object.

In particular:

  • Finalizing the Approximation values is done in a central function in InteractionValues
  • The Interactionvalues baseline and empty player value will be equal, unless the index is the SII, as SII is often just a step inbetween.

@Advueu963
Copy link
Copy Markdown
Collaborator Author

The sampling weights of FBII were updated to work appropriate.
They will be approximated if the number of players is to big

@mmschlk mmschlk merged commit f892935 into main Mar 20, 2025
8 checks passed
@mmschlk mmschlk deleted the 331-adding-faith-banzhaf-approximator branch March 20, 2025 10:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Adding Faith-Banzhaf Approximator

2 participants