Feat/support weights passed from dataframe

[L-M-Sherlock]

No description provided.

[Expertium]

I'll copy what I said in Discord

I think we shouldn't use weights, since that would make it easy to game the metrics just by choosing some oddly specific weighting function.
And the value of logloss would become essentially arbitrary since it depends on the choice of the weighting function
That being said, I do see a good usecase for this - finally making a proper "Ignore reviews before"
By setting weights for all reviews before a selected date to 0
In that case, the weights would have to be passed to RMSE as well, not just to logloss

[L-M-Sherlock]

I think we shouldn't use weights, since that would make it easy to game the metrics just by choosing some oddly specific weighting function.

The PR doesn't game the metrics. The weights are used for the optimization. In the evaluation, all reviews' weights are set to one.

[Expertium]

Good to know. I'll be eagerly waiting for you to update the benchmark code in other.py