Enforce positive sample_weight

As discussed in https://github.com/scikit-learn/scikit-learn/issues/12464#issuecomment-433815773 and https://github.com/scikit-learn/scikit-learn/issues/15358#issuecomment-549048650 in some specific use-cases negative sample_weight may be meaningful, however in most cases they should never happen.

So we may want to

• add force_positive=None parameter to _check_sample_weights.
• add a assume_positive_sample_weights=True config parameter to sklearn.set_config / get_config.

By default, force_positive=None would error on negative sample weight, but this check could be disabled globally with sklearn.set_config(assume_positive_sample_weights=False).

With _check_sample_weights(.., force_positive=True) the check would always be done irrespective of the config parameter.

If there are no objections to this, please tag this issue as “Help Wanted”.

[NicolasHug]

If an there are estimators which support negative sample weight, then it should be an estimator tag, and not a global config

I think the config still makes sense since negative sample weights don’t make sense in the vast majority of applications. You really want to know what you’re doing when using negative SW.

Whether estimators can support them is another thing, independent of the application.

[lorentzenchr]

I think, zero should always be allowed as a valid value for sample weights, but sum(sw) > 0 should hold.

For our dear hep friends, who want negative sw, often in linear models, they are implemented via scaling X and y by sqrt(sw). This makes supporting negative values difficult.