Without this change, attempting to pickle the model will result in an error due to the FM_fast Class lacking a proper implementation of the pickle interface. This PR implements it and removes a few unnecessary spaces. Also attributes of the class are made public because if not pickle has no access to them.
The models can be potentially quite large, over 4GB depending on how many features you use. So I recommend using joblib instead of pickle.
Without this change, attempting to pickle the model will result in an error due to the FM_fast Class lacking a proper implementation of the pickle interface. This PR implements it and removes a few unnecessary spaces. Also attributes of the class are made public because if not pickle has no access to them.
The models can be potentially quite large, over 4GB depending on how many features you use. So I recommend using joblib instead of pickle.