Method actionUtility

src/solver/PCfrSolver.cpp:406–588

}

vector<float>
PCfrSolver::actionUtility(int player, shared_ptr<ActionNode> node, const vector<float> &reach_probs, int iter,
                         uint64_t current_board,int deal) {
    int oppo = 1 - player;
    const vector<PrivateCards>& node_player_private_cards = this->ranges[node->getPlayer()];

    vector<float> payoffs = vector<float>(this->ranges[player].size());
    fill(payoffs.begin(),payoffs.end(),0);
    vector<shared_ptr<GameTreeNode>>& children =  node->getChildrens();
    vector<GameActions>& actions =  node->getActions();

    shared_ptr<Trainable> trainable;

    /*
    if(iter <= this->warmup){
        vector<int> deals = this->getAllAbstractionDeal(deal);
        trainable = node->getTrainable(deals[0]);
    }else{
        trainable = node->getTrainable(deal);
    }
     */
    trainable = node->getTrainable(deal);

#ifdef DEBUG
    if(trainable == nullptr){
        throw runtime_error("null trainable");
    }
#endif

    const vector<float> current_strategy = trainable->getcurrentStrategy();
#ifdef DEBUG
    if (current_strategy.size() != actions.size() * node_player_private_cards.size()) {
        node->printHistory();
        throw runtime_error(tfm::format(
                "length not match %s - %s \n action size %s private_card size %s"
                ,current_strategy.size()
                ,actions.size() * node_player_private_cards.size()
                ,actions.size()
                ,node_player_private_cards.size()
        ));
    }
#endif

    //为了节省计算成本将action regret 存在一位数组而不是二维数组中，两个纬度分别是（该infoset有多少动作,该palyer有多少holecard）
    vector<float> regrets(actions.size() * node_player_private_cards.size());

    vector<vector<float>> all_action_utility(actions.size());
    int node_player = node->getPlayer();

    vector<vector<float>> results(actions.size());
    for (int action_id = 0; action_id < actions.size(); action_id++) {

        if (node_player != player) {
            vector<float> new_reach_prob = vector<float>(reach_probs.size());
            for (int hand_id = 0; hand_id < new_reach_prob.size(); hand_id++) {
                float strategy_prob = current_strategy[hand_id + action_id * node_player_private_cards.size()];
                new_reach_prob[hand_id] = reach_probs[hand_id] * strategy_prob;
            }