[BUG FIX] fixed load method for dqn in pytorch

[Brunozml]

Problem

the past implementation of this function did not follow the syntax indicated by torch.load() docs, but rather seemed to imitate what is required by torch.save() docs.

This resulted in the following error:

  File "../train/dqn_pytorch/block_dominoes_deep_cfr.py", line 77, in main
    agents[i].load(FLAGS.checkpoint_dir + f'{i}.pt')
  File "../.venv/lib/python3.10/site-packages/open_spiel/python/pytorch/dqn.py", line 431, in load
    torch.load(self._q_network, data_path)
  File "../.venv/lib/python3.10/site-packages/torch/serialization.py", line 997, in load
    with _open_file_like(f, 'rb') as opened_file:
  File "../.venv/lib/python3.10/site-packages/torch/serialization.py", line 449, in _open_file_like
    return _open_buffer_reader(name_or_buffer)
  File "../.venv/lib/python3.10/site-packages/torch/serialization.py", line 434, in __init__
    _check_seekable(buffer)
  File "../.venv/lib/python3.10/site-packages/torch/serialization.py", line 543, in _check_seekable
    raise_err_msg(["seek", "tell"], e)
  File "../.venv/lib/python3.10/site-packages/torch/serialization.py", line 536, in raise_err_msg
    raise type(e)(msg)
AttributeError: 'MLP' object has no attribute 'seek'. You can only torch.load from a file that is seekable. Please pre-load the data into a buffer like io.BytesIO and try to load from it instead.

Solution

use correct function calls when loading .pt objects, e. g.

self._q_network = torch.load(data_path)

instead of

torch.load(self._q_network, data_path)

[lanctot]

Thanks @Brunozml !