I would like to know if I can use sklearn
precision_recall_fscore_support()
with CUDA tensors.
Normally, this function gets two arrays (specifically vectors), one with labels and another with predictions.
Can I give it multiple labels and prediction vectors in parallel, utilizing the GPUs?
Thank you in advance.
You won’t be able to directly pass CUDATensors from PyTorch to this method, since sklearn
uses numpy as the backend, which cannot use the GPU.
However, you could check the implementation and rewrite this method using torch
functions, which could then use CUDATensors.
This is unfortunate, however thank you @ptrblck for your fast reply.