added options for embeddings dimension reduction

[nadyadevani3112]

provided two options of embeddings dimension reduction: UMAP or T-SNE

[SebieF]

Thank you so much for your contribution to biotrainer!! :) I provided some feedback on your changes, in general they look very good.

I also rebased your branch to be consistent with the latest version of the develop branch, such that you do not have to do this somewhat tedious work of resolving conflicts: https://github.com/biocentral/biotrainer/commits/feature/emb-dim-redu-rebased/ I will provide you with detailed instructions how to use that before applying the suggestions in the next comment. Thanks alot again for your contribution!

[SebieF]

You could use Protocol.using_per_sequence_embeddings() here instead.

[nadyadevani3112]

I copied this part from the Interaction class in general_options.py, should I change it there too?

[SebieF]

Here's how you can change your branch such that it resolves the conflicts:

Assuming that you are on your current PR branch:

git remote add upstream https://github.com/biocentral/biotrainer.git
git fetch upstream
git reset --hard upstream/feature/emb-dim-redu-rebased
git push --force-with-lease

[SebieF]

P.S.: Do not worry about the broken Windows tests, they are failing because of onnx at the moment: #111

[nadyadevani3112]

i couldn't "git push --force-with-lease" successfully, had to pull and merge some conflicts before pushing it again successfully

[nadyadevani3112]

thank you for all your feedbacks, really appreciate them! I have made the changes, are you able to see the commits and review the PR again?

[SebieF]

Only one minor additional suggestion :)

[SebieF]

I think it would be better to raise an exception here, because if the user provides both the dimension_reduction_method and n_reduced_components than he or she will probably expect this to work and might not be aware, that it was not done after training. Admittedly, this is an edge case because if the number of samples is less than 3, then the train/validation/test splitting will also not work. But better be safe than sorry :)

[SebieF]

Thank you for applying the changes so quickly! I will look into rebasing again this week. Do you also want to update the documentation on the new config file options and add a new example on how to use it? If you do not have time for that currently, I would just add it after the PR is merged, so just let me know if you want to work on that. Thanks again for the contribution :)

[nadyadevani3112]

hi sorry it took me a while to implement the changes. I have also updated the config file options documentation and added a new example on how to apply the dimension reduction.

[SebieF]

Hi! Thanks for applying the changes and all your work! The final state looked good to me. In order to fix the conflicts with the current develop branch, I cherry picked your commits and made some tiny additions to them with you as the co-author. Then I had to force push the updated branch in order to be able to merge the PR. I hope that was okay for you! You can still find the original branch with the merges here: https://github.com/biocentral/biotrainer/tree/feature/embedding-dimension-reduction-keep

Thanks again for your contribution. If you plan to contribute to biotrainer in the future, feel free to reach out to me via my university mail address (see my profile). I would also be curious to hear how you are using biotrainer for your application or research :)