Describe the issue linked to the documentation

The example used in the stable TargetEncoder documentation is confusing as the order of the label (that is: dog, cat, snake) is not coherent with the expected order of enc_low_smooth.encodings_ (the 80 corresponds to 'dog' but is is in second order not first).

Printing TargetEncoder.categories_ reveal that the order is indeed coherent with TargetEncoder.encodings_. However, as I was trying to understand where this difference of order came from, I wasn't able to find in TargetEncoder class definition where self.categories_ was set.

Suggest a potential alternative/fix

• make it more explicit in documentation, such as adding a print of enc_auto.categories_
• make TargetEncoder() preserve the columns order found in the dataset