I am trying to minimize memory usage while using LightGBM with python. Reading the source code at https://testlightgbm.readthedocs.io/en/latest/_modules/lightgbm/basic.html , I can see that when the input data is provided as a pandas dataframe, then the data is automatically converted to float type.
My question is, if I provide data to the LGBMRegressor.fit() method that is already formatted in a numpy array with data type np.float16, will the data automatically get converted to float? Or will it stay as np.float16? In the source code, I only see special treatment for input that is a pandas dataframe.