Refactor dashboard

[stefsmeets]

This PR refactors the dashboard and moves it to streamlit.

Closes #542

In addition:

Resolves #418 (caching is handled by st.cache)
Resolves #428 (stop button top right)
Resolves #511 (no longer necessary)
Resolves #521 (no longer relevant, dashboard auto-updates)
Resolves #525 (methods are in their own containers)

Todo

• Sort kernelshap issue
• Organize model prepare step

[stefsmeets]

Hi @laurasootes Would you be willing to have a look at this and review it?

pip install streamlit
git checkout dashboard --
cd dashboard
streamlit run dashboard.py

[laurasootes]

I get an error message when trying to start the dashboard :( (using python 3.9.10, no problems installing streamlit)

[stefsmeets]

Are you in the right branch? app.run_server(debug=True) is the way the old dashboard got started. There is no such code in this branch 😅

[laurasootes]

never mind, I did not pull all changes yet.

[laurasootes]

Looks really good! The general outline already looks much nicer than the dash version. Also a lot less code, which makes it way more readable.

Some suggestions:

• I do get a whole lot of prints while running the dashboard for kernelshap, but can't find where it originates from (np.sum(w_aug), np.sum(self.kernelWeights) =, phi =). If you do know, can you remove it?
• Could you move all data to the data folder and remove the old app_data? (or the other way around)

I do have some aesthetic suggestions, but I think we can start from here and have another session with some people to decide how want the exact layout to look like.
One thing is that the colours for resnet don't look right yet and you can't see the original figure anymore, but I don't think that has to be fixed in this PR, since it is coming from the Dianna visualisation).

[stefsmeets]

Thanks for the review. I checked the data directory and I think that should all be OK. You may have existing data in the app_data directory, so that git did not remove it when you pulled the branch.

Let's create a new issue to follow up on the remaining points.