randomforest.cpp

#pragma GCC optimize ("O3")
#pragma GCC target ("avx")
#include "bits/stdc++.h" // define macro "/D__MAI"

using namespace std;
typedef long long int ll;

#define debug(v) {printf("L%d %s > ",__LINE__,#v);cout<<(v)<<endl;}
#define debugv(v) {printf("L%d %s > ",__LINE__,#v);for(auto e:(v)){cout<<e<<" ";}cout<<endl;}
#define debuga(m,w) {printf("L%d %s > ",__LINE__,#m);for(int x=0;x<(w);x++){cout<<(m)[x]<<" ";}cout<<endl;}
#define debugaa(m,h,w) {printf("L%d %s >\n",__LINE__,#m);for(int y=0;y<(h);y++){for(int x=0;x<(w);x++){cout<<(m)[y][x]<<" ";}cout<<endl;}}
#define ALL(v) (v).begin(),(v).end()
#define repeat(cnt,l) for(auto cnt=0ll;(cnt)<(l);++(cnt))
#define rrepeat(cnt,l) for(auto cnt=(l)-1;0<=(cnt);--(cnt))
#define iterate(cnt,b,e) for(auto cnt=(b);(cnt)!=(e);++(cnt))
#define diterate(cnt,b,e) for(auto cnt=(b);(cnt)!=(e);--(cnt))
#define MD 1000000007ll
#define PI 3.1415926535897932384626433832795
template<typename T1, typename T2> ostream& operator <<(ostream &o, const pair<T1, T2> p) { o << "(" << p.first << ":" << p.second << ")"; return o; }
template<typename T> T& maxset(T& to, const T& val) { return to = max(to, val); }
template<typename T> T& minset(T& to, const T& val) { return to = min(to, val); }
void bye(string s, int code = 0) { cout << s << endl; exit(code); }
mt19937_64 randdev(8901016);
inline ll rand_range(ll l, ll h) {
    return uniform_int_distribution<ll>(l, h)(randdev);
}

#if defined(_WIN32) || defined(_WIN64)
#define getchar_unlocked _getchar_nolock
#define putchar_unlocked _putchar_nolock
#elif defined(__GNUC__)
#else
#define getchar_unlocked getchar
#define putchar_unlocked putchar
#endif
namespace {
#define isvisiblechar(c) (0x21<=(c)&&(c)<=0x7E)
    class MaiScanner {
    public:
        template<typename T> void input_integer(T& var) {
            var = 0; T sign = 1;
            int cc = getchar_unlocked();
            for (; cc<'0' || '9'<cc; cc = getchar_unlocked())
                if (cc == '-') sign = -1;
            for (; '0' <= cc && cc <= '9'; cc = getchar_unlocked())
                var = (var << 3) + (var << 1) + cc - '0';
            var = var * sign;
        }
        inline int c() { return getchar_unlocked(); }
        inline MaiScanner& operator>>(int& var) { input_integer<int>(var); return *this; }
        inline MaiScanner& operator>>(long long& var) { input_integer<long long>(var); return *this; }
        inline MaiScanner& operator>>(string& var) {
            int cc = getchar_unlocked();
            for (; !isvisiblechar(cc); cc = getchar_unlocked());
            for (; isvisiblechar(cc); cc = getchar_unlocked())
                var.push_back(cc);
            return *this;
        }
        template<typename IT> void in(IT begin, IT end) { for (auto it = begin; it != end; ++it) *this >> *it; }
    };
    class MaiPrinter {
    public:
        template<typename T>
        void output_integer(T var) {
            if (var == 0) { putchar_unlocked('0'); return; }
            if (var < 0)
                putchar_unlocked('-'),
                var = -var;
            char stack[32]; int stack_p = 0;
            while (var)
                stack[stack_p++] = '0' + (var % 10),
                var /= 10;
            while (stack_p)
                putchar_unlocked(stack[--stack_p]);
        }
        inline MaiPrinter& operator<<(char c) { putchar_unlocked(c); return *this; }
        inline MaiPrinter& operator<<(int var) { output_integer<int>(var); return *this; }
        inline MaiPrinter& operator<<(long long var) { output_integer<long long>(var); return *this; }
        inline MaiPrinter& operator<<(char* str_p) { while (*str_p) putchar_unlocked(*(str_p++)); return *this; }
        inline MaiPrinter& operator<<(const string& str) {
            const char* p = str.c_str();
            const char* l = p + str.size();
            while (p < l) putchar_unlocked(*p++);
            return *this;
        }
        template<typename IT> void join(IT begin, IT end, char sep = '\n') { for (auto it = begin; it != end; ++it) *this << *it << sep; }
    };
}
MaiScanner scanner;
MaiPrinter printer;


template<typename input_t, typename output_t>
class DecisionTree {
    using comparator_t = function<bool(input_t)>;

    bool leaf_;
    comparator_t comparator_;
    output_t out_;
    unique_ptr<DecisionTree> childlen_[2];
public:

    DecisionTree() :leaf_(true), comparator_(nullptr){}

    inline DecisionTree& operator[](int i) { return *childlen_[i]; }
    inline bool leaf() { return leaf_; }

    inline void generate_leaf(output_t out) {
        leaf_ = true;
        childlen_[0].release();
        childlen_[1].release();
        out_ = out;
    }

    inline void generate_branch(comparator_t comparator) {
        leaf_ = false;
        childlen_[0].reset(new DecisionTree());
        childlen_[1].reset(new DecisionTree());
        comparator_ = comparator;
    }

    output_t eval(input_t val) {
        return leaf() ? out_ : childlen_[comparator_(val)]->eval(val);
    }
};


namespace Program {

    // 乱数生成用
    mt19937_64 _randdev(8901016);
    inline double rand_real(double low, double high){ return uniform_real_distribution<double>(low, high)(_randdev); }
    inline int rand_int(int low, int high) { return uniform_int_distribution<int>(low, high)(_randdev); }


    // データ の次元
    const int dimension = 2;

    // データの制約
    const double value_upper = 1.0;
    const double value_lower = -1.0;

    // 教師データ 
    using data_t = array<double, dimension>;
    vector<pair<int, data_t>> data;

    // ランダムフォレスト
    vector<DecisionTree<data_t, pair<int,int>>> randomforest;


    // 正解となるクラス分け関数
    inline int func(const data_t& x) {
        return x[0] * x[0] + x[1] * x[1] < 0.49 ? 1 : 0;
    };


    // 正解となるクラス分け関数を元に教師データを作成する
    void generate_inputdata() {

        // 教師データ数
        const int data_size = 1000;

        data.reserve(data_size);
        for (int i = 0; i < data_size; ++i) {
            data_t x;
            for (auto& e : x) e = rand_real(value_lower, value_upper);
            data.emplace_back(func(x), x);
        }

    }
    
    // 教師データから学習する
    void learn() {

        const int tree_num = 100;

        function<void(decltype(randomforest)::value_type&, vector<int>&)>
            build_dfs = [&build_dfs](decltype(randomforest)::value_type& dt, vector<int>& selected){
            
            int cnt = accumulate(selected.begin(), selected.end(), 0);
            if (cnt < 100) {
                pair<int, int> p;
                for (int i = 0; i < selected.size(); ++i) {
                    if (selected[i]) {
                        (data[i].first == 0 ? p.first : p.second)++;
                    }
                }
                dt.generate_leaf(p);
            }
            else {
                int sel = rand_int(0, dimension - 1);
                double threshold = rand_real(value_lower, value_upper);
                dt.generate_branch([sel, threshold](data_t d) {return threshold <= d[sel]; });
                {
                    vector<int> selected_new = selected;
                    for (int i = 0; i < data.size(); ++i)
                        selected_new[i] &= threshold <= data[i].second[sel];
                    build_dfs(dt[1], selected_new);
                }
                {
                    vector<int> selected_new = selected;
                    for (int i = 0; i < data.size(); ++i)
                        selected_new[i] &= !(threshold <= data[i].second[sel]);
                    build_dfs(dt[0], selected_new);
                }

            }

        };

        vector<int> selected_all(data.size(), 1);
        for (int i = 0; i < tree_num; ++i) {
            randomforest.emplace_back();
            build_dfs(randomforest.back(), selected_all);
        }
    }

    int predict(data_t x) {
        pair<int, int> rate;
        for (auto& dt : randomforest) {
            auto y = dt.eval(x);
            rate.first += y.first;
            rate.second += y.second;
        }
        return rate.first < rate.second;
    }

    void test() {

        int correct = 0, incorrect = 0;

        for (int no = 0; no < 10000; ++no) {
            data_t x;
            for (auto& e : x) e = rand_real(value_lower, value_upper);

            int predicted = predict(x);
            int truth = func(x);

            (predicted == truth ? correct : incorrect)++;
        }
        cout << "correct   : " << correct << endl;
        cout << "incorrect : " << incorrect << endl;
    }
}



int main() {

    Program::generate_inputdata();
    Program::learn();
    Program::test();


    string s;
    cin >> s;

    return 0;
}