Richard Edgar

The two tests which appear to be devouring time on Windows are: test_text_classification_unified_information test_explain_model_BERT_seq_classification see, e.g.: https://dev.azure.com/responsibleai/interpret-text/_build/results?buildId=7329&view=ms.vss-test-web.build-test-results-tab&runId=78160&resultId=100015&paneView=debug

Initial pass on this is done.

Thanks for calling attention to this. It's been on our radar for a while, but we've not had time to think about the best way to show it in the...

In principle, for sure - the dashboard calls back into Python for all the metrics, and there is a 'mapping' object to route the calls. I'm not sure how much...

Weren't we planning to get rid of the `compute()` functions?

Per comment in Teams thread, we may want to pause on this, since I believe that SciKit-Learn doesn't tend to keep pickle-compatibility between versions.

Not quite what you're asking, but you can start up a dashboard with `y_pred` as a probability, rather than a class. You'll get a different set of metrics, which might...

Ahhh, I thought you might have access to `predict_proba()` or something like that. That would let you have probabilities. It wouldn't get you everything, but might help with alternative plots.

I don't have any objection to bumping the TF requirement @hildeweerts . Those tests are only there to show that we can work with other packages, and are not meant...

I wouldn't object to rethinking how that works (and I agree that we don't check that build enough). My main concern would be to have separate environments for the 'other'...