diff --git a/open_spiel/python/pytorch/dqn.py b/open_spiel/python/pytorch/dqn.py index 7b5bc775e9..f027b115e6 100644 --- a/open_spiel/python/pytorch/dqn.py +++ b/open_spiel/python/pytorch/dqn.py @@ -30,7 +30,7 @@ "Transition", "info_state action reward next_info_state is_final_step legal_actions_mask") -ILLEGAL_ACTION_LOGITS_PENALTY = sys.float_info.min +ILLEGAL_ACTION_LOGITS_PENALTY = torch.finfo(torch.float).min class SonnetLinear(nn.Module):