Problem: for some reason, the API set_feature_names does not set the feature names properly afterward. I have a model trained with 430 features in RAW values (e.g. pandas.DataFrame.values). Later, I tried to reset its feature names on the trained model, with the set_feature_names call. However, the last 3 features in the model were not reset.

More specifically, the missing features are text features.

model.set_feature_names(feature_list)

print(model.feature_names_) 
...
"a_proper_name"
"428"
"429"
"430",

catboost version: 1.0.0