Sample_weight is a misnomer

[caugusta]

There seems to be a lot of confusion around sample_weight and sample_weight_mode (e.g 2594, 7120, 7497). As explained elsewhere, these apply a weight on the loss, which means if sample_weight_model=temporal then you should have 3D outputs. However, the name 'sample_weight' implies a weight on the samples, which is not what's happening. Should the name be changed? Is there a way to weight the actual samples instead of the loss based on the samples?

[FitMachineLearning]

I second that.