Implement evaluation API for multiclass classification problem

[przemekwitek]

This PR adds a new evaluation type ("classification") and a new metric ("MulticlassConfusionMatrix").

Relates #46735

[elasticmachine]

Pinging @elastic/ml-core

[dimitris-athanasiou]

Looks good! Left a few comments.

[dimitris-athanasiou]

We generally prefer that an else if clause is operating on the same data as the if clause to keep reasoning simple. In this case I think we don't really need the else at all as we are returning from each branch. So we could have:

if (topActualClassNames == null) {
    // This is step 1
    ...
    return {};
}
if (result == null) {
    // This is step 2
    ...
    return {};
}
return List.of();

[dimitris-athanasiou]

I wonder if it would also make it more readable to extract the building of the aggregations of each step into its own method.

[przemekwitek]

In this case I think we don't really need the else at all as we are returning from each branch.

Done.

I wonder if it would also make it more readable to extract the building of the aggregations of each step into its own method.

I'm not sure this would help much. The method is not (yet) very long so I'd keep it as it is if you don't have a strong opinion about refactor.

[dimitris-athanasiou]

I wonder if this should also be called just _other_.

[przemekwitek]

Done on the API level.

LMK if you think I should also rename the variables. In the code I would slightly prefer more descriptive "other_classes_count".

[przemekwitek]

run elasticsearch-ci/default-distro

[przemekwitek]

run elasticsearch-ci/bwc

[przemekwitek]

run elasticsearch-ci/bwc

[przemekwitek]

run elasticsearch-ci/1
run elasticsearch-ci/bwc

[dimitris-athanasiou]

LGTM