Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add composite LRP example #261

Draft
wants to merge 3 commits into
base: master
Choose a base branch
from

Conversation

adrhill
Copy link
Collaborator

@adrhill adrhill commented Jun 30, 2021

An example on how to use composite LRP is missing from the docs and frequently requested.

Closes #252
Addresses #162, #190, #249, #255


This PR adds a Jupyter notebook explaining how to use the keyword arguments of the LRP analyzer class.
I'm opening this as a draft PR as we should

  • add references
  • check whether the example in the notebook can be improved

The references I would add are

  • Lapuschkin et al.: Understanding and Comparing Deep Neural Networks for Age and Gender Classification
  • Montavon et al.: Layer-Wise Relevance Propagation: An Overview

Maybe @sebastian-lapuschkin you can help me out here?

@adrhill
Copy link
Collaborator Author

adrhill commented Jun 30, 2021

Here's a screenshot of the example in the notebook for quick reference:
image

@codecov-commenter
Copy link

codecov-commenter commented Jun 30, 2021

Codecov Report

Merging #261 (d5bbee1) into master (c6ce893) will not change coverage.
The diff coverage is n/a.

Impacted file tree graph

@@           Coverage Diff           @@
##           master     #261   +/-   ##
=======================================
  Coverage   71.36%   71.36%           
=======================================
  Files          41       41           
  Lines        4173     4173           
  Branches      637      637           
=======================================
  Hits         2978     2978           
  Misses       1007     1007           
  Partials      188      188           

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c6ce893...d5bbee1. Read the comment docs.

@sebastian-lapuschkin
Copy link
Contributor

another reference to add would be

Kohlbrenner et al.: Towards Best Practice in Explaining Neural Network Decisions with LRP

which properly introduces the composite LRP pretty much at the same time as Montavon et al (preprints existed a bit earlier) and evaluates the composite approach against other methods and approaches.

is there anything else I can help with?

@adrhill
Copy link
Collaborator Author

adrhill commented Jul 1, 2021

is there anything else I can help with?

Is there anything you would change about the example in the screenshot?

@sebastian-lapuschkin
Copy link
Contributor

maybe add some line(s) actually using this custom built analyzer, and some example heatmaps of this composite rule with non-composite counter parts for comparison?

a hint at the existing LRP presets (if they will be carried over to version 2.0) would also be interesting to the users I guess.

@adrhill
Copy link
Collaborator Author

adrhill commented Jul 1, 2021

A usage example is already in the notebook.
I will add a comparison with the existing presets.

Thanks!

@adrhill adrhill marked this pull request as ready for review July 1, 2021 10:50
@adrhill adrhill marked this pull request as draft July 1, 2021 13:14
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Add Composite LRP example to docs
3 participants