std::tuple step()

in atari/atari_env.h [113:155]
33 lines of code
7 McCabe index (conditional complexity)

  std::tuple<rela::TensorDict, float, bool> step(
      const rela::TensorDict& action) final {
    // take an ale step
    torch::Tensor a = action.at("a");
    float reward = aleStep(a.item<int>());

    // update state
    state_->addReward(reward);
    state_->setLives(ale_->lives());
    if (ale_->game_over() || numSteps_ * frameSkip_ > maxNumFrame_) {
      state_->setTerminal();
    }

    // compute reward and terminal signal
    float clippedReward = clipRewards(reward);
    bool terminalSignal =
        state_->terminal() || (terminalSignalOnLifeLoss_ && state_->lostLife());
    if (state_->terminal()) {
      // state should not matter, but we still need to send it
      torch::Tensor obs = state_->computeFeature();
      rela::TensorDict input = {
        {"s", obs},
        {"eps", exploreEps_},
        {"legal_move", legalActionMask_}
      };
      return std::make_tuple(input, clippedReward, true);
    }

    // if lost life (but game is not over) need to press start key again
    if (state_->lostLife()) {
      pressStartKey();
    }

    // compute obs
    ale_->getScreenRGB(state_->getObservationBuffer());
    torch::Tensor obs = state_->computeFeature();
    rela::TensorDict input = {
      {"s", obs},
      {"eps", exploreEps_},
      {"legal_move", legalActionMask_}
    };
    return std::make_tuple(input, clippedReward, terminalSignal);
  }