Converting numpy arrays to pandas dataframes for sharding implicitly adds coluimn names which are stored in the reuslting xgboost booster objects. However, calling predict on such boosters with numpy arrays leads to errors. Instead we should unset feature names for numpy arrays if they haven't been deliberately passed to RayDMatrix.
Converting numpy arrays to pandas dataframes for sharding implicitly adds coluimn names which are stored in the reuslting xgboost booster objects. However, calling
predicton such boosters with numpy arrays leads to errors. Instead we should unset feature names for numpy arrays if they haven't been deliberately passed toRayDMatrix.