Creating a tensor from a list of numpy.ndarrays is extremely slow

[zakaria-narjis]

/usr/local/lib/python3.10/dist-packages/pfrl/replay_buffer.py:180: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at ../torch/csrc/utils/tensor_new.cpp:245.) "action": torch.as_tensor(

pfrl/pfrl/replay_buffer.py

Line 180 in b29533b

"action": torch.as_tensor(

I found something related to this issue pytorch/pytorch#13918

[Pranav-Malpure]

I was facing the same issue, I think this is because of some lazy development of PFRL. This is how I solved it:
replace the below

"action": torch.as_tensor(
            [elem[0]["action"] for elem in experiences], device=device
        )

by

"action": torch.from_numpy(
            np.asarray([elem[0]["action"] for elem in experiences])).to(device)

This should solve it