Can't make dashboard(runs forever) on remote private cloud

[GISH123]

As the picture shows, the ExplanationDashboard function runs forever even with only 10 test data (data is tabular with 91 columns).
Because the jupyter notebook is on the remote private cloud, and i don't have full privilege on this cloud,
I suspect that it is the problem with IP address or something server-related problems.

I can successfully execute other functions(local or global feature importance) and all original InterpretML functions(although
I have to do the visualization on the notebook inline, and if opened on new tab, the access will be denied).

Is it possible to make the visualization without posting it to local server?

Sorry for my weak english.

[GISH123]

Add a picture for more helpful messages:

[imatiach-msft]

@GISH123 the posting is only for what if analysis and ICE plots. It seems that starting the flask server is giving you trouble in the second screenshot:

https://github.com/interpretml/interpret-community/blob/master/python/interpret_community/widget/explanation_dashboard.py#L174

This only happens when with_credentials=True, but I'm not sure you need this set - this is only if cookies are needed for your requests to the VM.

In the first screenshot you don't have this parameter set, but I'm not sure why it is failing. There should be a different error message somewhere. Without a repro and access to the environment it is really difficult to debug these types of issues.

Without with_credentials parameter set, the widget should still show, and WSGIServer would just run on localhost. So I'm not sure what is happening in the first screenshot.

If you like, we can try to have a call and debug together. In particular, I'm interested why in the first case the code is getting stuck, and where it is getting stuck. Cloning the repository and adding a bunch of print debug statements everywhere in the explanation_dashboard.py should reveal at what line the widget is getting stuck.

[GISH123]

It seems like my global_explanation was not functioning properly, it was producing no outputs, and now currently it is producing a tqdm with shap outputs.
(I pip installed interpret-community and cloned repo on my home directory, now i deleted the cloned repo and running the codes it functioned properly, not sure the cause though)

Now the explanationdashboard is happening, but it seems like i need extra certification from my server. I'm trying to figuring it out on my own, and if I need help i will ask again.
Thank you for your help!

[GISH123]

By the way,if I do something like this:

global_explanation = explainer.explain_global(x_train[:100])
then
ExplanationDashboard(global_explanation, model, datasetX=x_test[:50])

would it be reasonable to use different data volume?

[GISH123]

Now the dashboard becomes something like this:

It says I have to use an account to get access to the website.
Well I just want to print the dashboard on the jupyter cell.

[imatiach-msft]

@GISH123 interesting, I've never seen this before! It seems the javascript can't be served to the jupyter notebook. There seems to be some middleware on your VM causing this error to appear. I'm not quite sure how to fix it though 😕

[imatiach-msft]

@GISH123 ah, I just thought of something, can you try setting the use_cdn=False parameter? Maybe it is having issues retrieving the javascript from the CDN - this way you can retrieve it locally.