✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 17ace67. Link to the linter CI: here}

Maybe we could explain that the motivation is to be consistent with the convention of accuracy_score that also outputs a value between 0 and 1 instead of 0 and 100.

🤔 MAPE is not in the range [0, 1], it can be arbitrarily large, right? I have to say I don't really know how to word this to make it clearer that a percentage is not returned despite the name.

Also I guess that whatever we improve should be done similarly in the docstring because I would say the reference API doc is likely where people are going to go for information first.

In [1]: from sklearn.metrics import mean_absolute_percentage_error
mes
In [2]: mean_absolute_percentage_error([1e-6, 1e-6, 1e-6], [1, 2, 3])
Out[2]: 1999999.0

Indeed you can make an error of 200%. We need to reformulate and drop the notion of upper bound.

Maybe give an example where the error is 1 (or 100%) in the user guide to clarify that the returned value is 1 and not 100?

In [5]: mean_absolute_percentage_error([1, 2, 3], [2, 4, 6])
Out[5]: 1.0

Indeed sorry for the misleading comment. Still I think the fact that it's consistent with our convention that 1.0 returned by accuracy_score means 100%.

Any volunteer for a follow-up PR to fix this?