Eliminate extra dataset copy in Python.
trivialfis opened this issue · comments
Jiaming Yuan commented
During construction of DMatrix, Python wrapper might duplicate the dataset because it's not continuous or not of the right data type. But we can handle these situations inside c++ code to avoid constructing an extra copy of dataset.
My goal is not to share the underlying buffer with Python data structure, but to eliminate the extra copies constructed during conversion inside Python wrapper.
Jiaming Yuan commented
Wrong list, sorry~~