I think is a good idea to create bins from np.array values but we need to wait an answer of a core developer for this issue. I can work on it if it is accepted.

Also, it's better that edge's bins take strictly less than or less and equal values for the discretization?

One issue is that you will need to pass a list of arrays of length n_features. I don't know if it will be easy to deal with a high number of features.

ping @ogrisel

One issue is that you will need to pass a list of arrays of length n_features. I don't know if it will be easy to deal with a high number of features.

Doesn't the same issue pop up with OrdinalEncoder when we pass in lists of ordered categories for each feature?

Secondly, the user was any way going to use pd.cut( ) and pass in list of arrays one feature at a time, and manually do it for the train and test sets separately.

It will simplify things for them as they can now automate the transform step for all the test set features which were binned.

scikit-learn will remember the bin edges from the fit and appropriately do the work on the test set when transformed

I'm in favour to offer a manual setup for bins. In my experience, this is important for use cases of just one or a few features that you want to bin yourself, often together with a ColumnTransformer.

The SplineTransformer of PR #18368 supports this, i.e. manually specifying bins (there it is knot positions). Note that SplineTransformer(degree=0, n_knots=n_bins+1) is equivalent to KBinsDiscretizer(n_bins=n_bins, encode='onehot-dense').

It would just be an additional argument. If we decide against, it would be good to give an example with the FunctionTransformer.

I would be in favour of using the FunctionTransformer. An example seems the best way. I think that it should go to the user guide.

Hello @glemaitre! I am pretty new to this and would like to contribute. Is this issue still open or someone is already working on it? Thanks.

@bhargavasomya sorry for the delay to answer. The issue is still open and you can submit a pull request

Sir I also want to work on this issue. If no one is working on it, can I ?

@bhargavasomya are you working on the issue?