I think this was discussed before but I don't remember where and what the outcome was :-/
Maybe @arjoly or @glouppe remember.

Currently sample weights or class weights are not taken into account when doing bootstrap sampling, they are applied after the sampling. I think it would be good to have an option to apply them before the sampling. That is particularly useful for datasets with strong class imbalance, where we can rebalance using class-weights. Wdyt?