Emmanuel confusion matrix

[tomasoignons]

This pull request add a confusion matrix to the testing page

The confusion matrix has been made so that it works with multiple class classfication, and not only binary classification.

This pull request modify the Validator class in the discoJS library, in order to retrieve

• If the element was correctly predicted
• The prediction
• the original label of the element

[JulienVig]

Hi @tomasoignons! Thanks for the PR! I'm super busy these days so I can maybe review it next week.
Could you add a screenshot of how it looks like for MNIST? (when there are more categories)

[tomasoignons]

I had to train my own model, because we don't have the MNIST on the models available as a simple demo, and it looks really good in my opinion :

And in light mode :

[tharvik]

thanks for the nicer test visualisation, love a table filling up of data!
I've got a few comments on how it can be tidier but the overall feature is great!

I also got a bunch of errors toaster at the start of the test:TypeError: matrix[output.predicted] is undefined

[tomasoignons]

I have no clue why I don't pass the timeout here, maybe my code is suboptimal, or the postprocess takes longer and I am just above, if @JulienVig or @tharvik could check this point (not review the whole merge request, just check why the test fails, because I try random things, but the test seems to work fine until the last bit) if would be perfect.

[tharvik]

I have no clue why I don't pass the timeout here

ho, yeah, some integration tests are a bit flaky, fell free to retry the CI/CD workflow if it doesn't look like it's coming from your code.
in this specific case, I think that it comes from the postprocess pass which wasn't needed in the previous implementation, so it's not the test flakiness. you can double the timeout of the validator/"can read and predict randomly on simple_face" test (we only care that it finishes, not that it does it fast, we would need benchmark for that)