9e80c11073af48db2876fc943df9264a7ab0488e,ch11/lib/common.py,,unpack_batch,#Any#Any#Any#Any#,82

Before Change

        if exp.last_state is not None:
            not_done_idx.append(idx)
            last_states.append(np.array(exp.last_state, copy=False))
    states_v = Variable(torch.from_numpy(np.array(states, copy=False)))
    actions_t = torch.LongTensor(actions)
    if cuda:
        states_v = states_v.cuda()
        actions_t = actions_t.cuda()

    // handle rewards
    rewards_np = np.array(rewards, dtype=np.float32)
    if not_done_idx:
        last_states_v = Variable(torch.from_numpy(np.array(last_states, copy=False)), volatile=True)
        if cuda:
            last_states_v = last_states_v.cuda()
        last_vals_v = net(last_states_v)[1]
        last_vals_np = last_vals_v.data.cpu().numpy()[:, 0]
        rewards_np[not_done_idx] += last_val_gamma * last_vals_np

    ref_vals_v = Variable(torch.from_numpy(rewards_np))
    if cuda:
        ref_vals_v = ref_vals_v.cuda()

    return states_v, actions_t, ref_vals_v

After Change

        if exp.last_state is not None:
            not_done_idx.append(idx)
            last_states.append(np.array(exp.last_state, copy=False))
    states_v = torch.FloatTensor(states).to(device)
    actions_t = torch.LongTensor(actions).to(device)

    // handle rewards
    rewards_np = np.array(rewards, dtype=np.float32)
    if not_done_idx:
        last_states_v = torch.FloatTensor(last_states).to(device)
        last_vals_v = net(last_states_v)[1]
        last_vals_np = last_vals_v.data.cpu().numpy()[:, 0]
        rewards_np[not_done_idx] += last_val_gamma * last_vals_np

    ref_vals_v = torch.FloatTensor(rewards_np).to(device)
    return states_v, actions_t, ref_vals_v

Instances