allow_const_label=True does not seem to work with CatBoostClassifier

[tristancinquin]

Hi,

I am trying to get a CatBoostClassifier to work in a case where I only have one label. I saw that CatBoostClassifiers have the allow_const_label parameter for this. However, I still run into the error : catboost.CatBoostError: catboost/private/libs/target/target_converter.cpp:374: Target contains only one unique value.

The error is reproducible with the following code :

import catboost
model = CatBoostClassifier(allow_const_label=True)
model.fit([[1,1,1],[1,0,1],[0,1,1]], [1,1,1])

catboost version: 1.0.3
Operating System: MacOS 11.6.1
CPU: Yes
GPU: No

[andrey-khropov]

Fixed in 79f1702