enh(skore): Revisit the default metrics in metrics.summary and adapt depending on the target #2110
Description
What would you like to say?
Currently, we implemented a couple of metrics that comes from scikit-learn.
#1378 proposes to add many more metrics to the accessor.
Once done, we should revisit which metrics we put as default metrics when scoring (or in the future metrics) is None. We should consider the following:
- Depending of the type of target, then we should select a sensible choice of metrics. Right now we are almost there.
- However, depending on some specificity of the target feature (e.g. class imbalanced, zero-value in target for regression), some metrics would not be the right choice out of the box (e.g. accuracy, MAPE). In this case we need to adapt the default metrics as well.