Guillaume Lemaitre

> Hmm, perhaps. I understand it's not the greatest way of doing things, but this wasn't my code - I used the following example (https://github.com/serengil/tensorflow-101/blob/master/python/Ensemble-Face-Recognition.ipynb), where the initial decision (target)...

> Yup, that is certainly a very useful feature! I'm not sure if LightGBM can also handle string labels though - which is what is being used in the notebook....

After reviewing #24448, I see that I wrongly suggest a fix. `type_of_target` should not raise an error. Instead, the callee of `type_of_target` should catch `"unknown"` and then raise the proper...

Actually, this is not the right fix. I made a mistake in the original issue. `type_of_target` does not raise the error. So we should not modify `type_of_target`. Instead, we need...

Plotting uncertainty or confidence intervals is not something straightforward and I think that there is a benefit to doing so for our users. I think that our displays should be...

@Micky774 Could you ping me once you addressed @jjerphan comments such that I make a last review before the merge?

I think that a discrete decision boundary might be enough. To be honest, what I miss in this example is a bit of narration. I would prefer to have a...

I agree that usually, we should monitor the loss instead of the final metric. This is indeed the default in other estimators such as gradient boosting. We could think of...

I can reproduce. I don't think that this check should fail. It looks like a real bug.

I took a closer look. It seems that the rounding has an effect here. Indeed, the rounding will make that we will have ties in probability while this is not...