Skip to content

Conversation

@Nimanui
Copy link

@Nimanui Nimanui commented Oct 16, 2025

Goal of this pull request is to add basic gradient saliency mapping in the appropriate location in pyhealth as well as providing a clean python notebook example with a basic CNN since a CNN is required to display the saliency map. Assuming this looks ok, I have a basic LRP saliency mapping and GradCAM to implement next since I already have basic pytorch implementations of those already.

Nimanui and others added 24 commits May 7, 2025 21:45
Applying and visualizing gradient saliency on a basic pyhealth model claissifying image data
…h dataloader and models

(cherry picked from commit 858c4f7)
Trying to find a version of this that github can render without errors
Still trying to get something that displays correctly in github
@Nimanui Nimanui closed this Oct 16, 2025
@Nimanui Nimanui reopened this Oct 16, 2025
Remove temporary branch references from python notebook example import
Copy link
Collaborator

@jhnwu3 jhnwu3 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Awesome! Can you write a super minimal test case in tests/core?

@Nimanui
Copy link
Author

Nimanui commented Oct 16, 2025

Reasonable, there's probably something small I can test without a full CNN, I'll look at the other tests for options.

Adding minimal testing of the gradient saliency function
@Nimanui
Copy link
Author

Nimanui commented Oct 19, 2025

I wasn't able to find something without a CNN, but it was simple to create a test with a dummy CNN. Would you like me to make it use a CNN with other pyhealth modules or is this sufficient? Keeping it with something small will keep it running pretty quickly, I noticed some of the unit tests in the broader suite are very slow.

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.

2 participants