b3972d8247961127b4e71d0af0cff5f6bf19ab5f,slm_lab/agent/algorithm/algorithm_util.py,,act_with_epsilon_greedy,#Any#Any#Any#Any#,37

Before Change

    if epsilon > np.random.rand():
        action = np.random.randint(body.action_dim)
    else:
        torch_state = Variable(torch.from_numpy(state).float())
        out = net.wrap_eval(torch_state)
        action = int(torch.max(out, dim=0)[1][0])
    return action

After Change

    if epsilon > np.random.rand():
        action = np.random.randint(body.action_dim)
    else:
        recurrent = body.agent.len_state_buffer > 0
        logger.debug2(f"Length state buffer: {body.agent.len_state_buffer}")
        torch_state = create_torch_state(state, body.state_buffer, recurrent, body.agent.len_state_buffer)
        out = net.wrap_eval(torch_state).squeeze_(dim=0)
        action = int(torch.max(out, dim=0)[1][0])
        logger.debug2(f"Outs {out} Action {action}")
    return action

Instances