Error: Fixed user_splits are removed because produce pure prebins

[ben-herbst]

I need to fix the bins for certain categorical values. Your example in the tutorial is the following:

user_splits = np.array([
               ['Businessman'],
               ['Working'],
               ['Commercial associate'],
               ['Pensioner', 'Maternity leave'],
               ['State servant'],
               ['Unemployed', 'Student']], dtype=object)
optb = OptimalBinning(name=variable_cat, dtype="categorical", solver="cp",
                      user_splits=user_splits,
                      user_splits_fixed=[False, True, True, True, True, True])

optb.fit(x_cat, y_cat)

Then

opt.splits

gives

[['Businessman', 'Pensioner', 'Maternity leave'],
 ['State servant'],
 ['Commercial associate'],
 ['Working'],
 ['Unemployed', 'Student']]

This is not what I need. I need the bins to remain exactly the same as specified. If I change the following

optb = OptimalBinning(name=variable_cat, dtype="categorical", solver="cp",
                      user_splits=user_splits,
                      user_splits_fixed=[True, True, True, True, True, True])

optb.fit(x_cat, y_cat)

I get the error:

ValueError: Fixed user_splits [list(['Businessman'])] are removed because produce pure prebins. Provide different splits to be fixed.

I get the same error if I try

user_splits = np.array([
               ['Businessman'],
               ['Working'],
               ['Commercial associate'],
               ['Pensioner', 'Maternity leave'],
               ['State servant'],
               ['Unemployed'], 
               ['Student']], dtype=object)
optb = OptimalBinning(name=variable_cat, dtype="categorical", solver="cp",
                      user_splits=user_splits,
                      user_splits_fixed=[False, True, True, True, True, True, True])

optb.fit(x_cat, y_cat)

but the following seemingly similar change does not give the error

user_splits = np.array([
               ['Businessman'],
               ['Working'],
               ['Commercial associate'],
               ['Pensioner'], 
               ['Maternity leave'],
               ['State servant'],
               ['Unemployed', 'Student']], dtype=object)
optb = OptimalBinning(name=variable_cat, dtype="categorical", solver="cp",
                      user_splits=user_splits,
                      user_splits_fixed=[False, True, True, True, True, True, True])

optb.fit(x_cat, y_cat)

[guillermo-navas-palencia]

Hi @ben-herbst.

The reason is simple, both categories "Businessman" and "Student" are pure bins, in this case, the average target is zero. In the latter example, the first bin ["Businessman"] is not fixed, and "Student" is merged with "Unemployed".

[ben-herbst]

Thanks, much appreciated! [image: Praelexis] <https://praelexis.com/> Ben Herbst Data Scientist | Machine Learning Engineer 7 Neutron Avenue ⋅ Techno Park ⋅ Stellenbosch ⋅ 7600 PO Box 3396 ⋅ Matieland ⋅ Stellenbosch ⋅ 7602 mobile: +27 83 566 4466 ⋅ office: +27 21 200 5817 website <http://www.praelexis.com/>| map ***@***.***,18.8270702,17z/data=!3m1!4b1!4m5!3m4!1s0x1dcdb3226a13a605:0x12022a8f60a2a6bb!8m2!3d-33.9651846!4d18.8292589> | email ***@***.***> [image: Twitter] <https://twitter.com/praelexis>[image: Facebook] <http://www.facebook.com/praelexis>[image: LinkedIn] <https://www.linkedin.com/company/praelexis/> *Confidentiality Note: *This email may contain confidential and/or private information. If you received this email in error please delete and notify sender.

…

On Fri, Dec 1, 2023 at 10:26 PM Guillermo ***@***.***> wrote: Closed #285 <#285> as completed. — Reply to this email directly, view it on GitHub <#285 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AZO3U53JIAXUFXDADDAQNWLYHI4QLAVCNFSM6AAAAAA7X7TLCSVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJRGEZDMMRSGY3DINI> . You are receiving this because you were mentioned.Message ID: <guillermo-navas-palencia/optbinning/issue/285/issue_event/11126226645@ github.com>