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

Number of UMAP components #165

Closed
ziw-liu opened this issue Sep 16, 2024 · 3 comments · Fixed by #153
Closed

Number of UMAP components #165

ziw-liu opened this issue Sep 16, 2024 · 3 comments · Fixed by #153
Labels
bug Something isn't working representation Representation learning (SSL)
Milestone

Comments

@ziw-liu
Copy link
Collaborator

ziw-liu commented Sep 16, 2024

          @ziw-liu , I have been using 2 UMAP components for plotting and 4 components when I compute correlation with computed features. Does this mean it will change the interpretation of results?

Originally posted by @Soorya19Pradeep in #153 (comment)

@ziw-liu
Copy link
Collaborator Author

ziw-liu commented Sep 16, 2024

2, 3, and 4 components UMAPs are being used in different places. My intuition is that we should only use 2 components for plotting, and do all the correlations in PC space for better interpretability.

@ziw-liu
Copy link
Collaborator Author

ziw-liu commented Sep 16, 2024

cc @alishbaimran @mattersoflight

@ziw-liu ziw-liu added the representation Representation learning (SSL) label Sep 16, 2024
@mattersoflight
Copy link
Member

@ziw-liu @Soorya19Pradeep @alishbaimran I agree with Ziwen - use UMAP for plotting/visualization and PCA for computing correlations with engineered features. We should do 3D UMAP renderings in napari, like in zebrahub paper: https://github.com/royerlab/zebrahub-paper-umap-3d. So # of UMAP components can be 2 or 3.
When computing PCA, use the number of components that explain 95% variance of embeddings.

@ziw-liu ziw-liu added this to the v0.3.0 milestone Oct 16, 2024
@ziw-liu ziw-liu added the bug Something isn't working label Oct 16, 2024
@ziw-liu ziw-liu linked a pull request Oct 16, 2024 that will close this issue
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working representation Representation learning (SSL)
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants