Catboost crashes when attempting to train a multilabel classifier with embedding features #2249
Description
Problem: Cannot use embedding_features with CatBoostClassifier when the loss_function is MultiLogLoss. This crashes when attempting to fit a model for a multilabel classification task.
Example code to reproduce:
import pandas as pd
import catboost
features_df = pd.DataFrame({
'f0': [0, 1, 2],
'embedding1': [[0.0, 1.1, 0.0], [1.0, 2.1, 0.0], [0.11, 0.2, 0.3]]
})
label = [[0, 0], [0, 1], [1, 0]]
pool = catboost.Pool(features_df, label=label, embedding_features=['embedding1'])
model = catboost.CatBoostClassifier(iterations=10, loss_function='MultiLogloss')
model.fit(pool)
Catboost will throw an exception with this error message:
CatBoostError: catboost/libs/data/target.h:315: Attempt to use multi-dimensional target as one-dimensional
catboost version: 1.1.1
Operating System: Linux
CPU: Yes
GPU: Yes