]> git.sesse.net Git - wloh/blobdiff - bayeswf.cpp
Remove dead function print_navbar().
[wloh] / bayeswf.cpp
index cc591832a556f023ea6aa364a44ce5422a0b0740..eee6882e4e18f59dcc9c2b8f37f950e22b3d6e13 100644 (file)
@@ -17,11 +17,32 @@ using namespace Eigen;
 #define PRIOR_WEIGHT 1.0
 #define MAX_PLAYERS 4096
 #define DUMP_RAW 0
+#define USE_DB 1
+
+#if USE_DB
+#include <pqxx/connection>
+#include <pqxx/tablewriter>
+#include <pqxx/transaction>
+#endif
 
 float mu[MAX_PLAYERS];
 float mu_stddev[MAX_PLAYERS];
-float global_sigma = 70.0f;
-float prior_sigma = 70.0f;
+float global_sigma;
+float prior_sigma;
+
+// Data waiting for insertion into the database.
+
+struct RatingDBTuple {
+       int player;
+       float mu, mu_stddev;
+};
+struct CovarianceDBTuple {
+       int player1, player2;
+       float covariance;
+};
+vector<RatingDBTuple> rating_db_tuples; 
+vector<CovarianceDBTuple> covariance_db_tuples;
+map<pair<string, string>, float> aux_params;
 
 #define EPSILON 1e-3
 
@@ -44,21 +65,16 @@ struct match {
 map<int, vector<match> > matches_for_player;
 vector<match> all_matches;
 
-void dump_scores(const vector<string> &players, const float *mu, const float *mu_stddev, int num_players)
+void dump_scores(const vector<int> &players, const float *mu, const float *mu_stddev, int num_players)
 {
-#if 0
-       for (int i = 0; i < num_players; ++i) {
-               printf("%s=[%5.1f, %4.1f] ", players[i].c_str(), mu[i], sigma[i]);
-       }
-       printf("\n");
-#elif 0
+#if USE_DB
        for (int i = 0; i < num_players; ++i) {
-               printf("%5.1f ", mu[i]);
+               RatingDBTuple tuple = { players[i], mu[i], mu_stddev[i] };
+               rating_db_tuples.push_back(tuple);
        }
-       printf("\n");
 #else
        for (int i = 0; i < num_players; ++i) {
-               printf("%f %f %s\n", mu[i], mu_stddev[i], players[i].c_str());
+               printf("%f %f %d\n", mu[i], mu_stddev[i], players[i]);
        }
 #endif
 }
@@ -227,7 +243,7 @@ void construct_hessian(const float *mu, int num_players)
 
 // Compute uncertainty (stddev) of mu estimates, which is sqrt((H^-1)_ii),
 // where H is the Hessian (see construct_hessian()).
-void compute_mu_uncertainty(const float *mu, const vector<string> &players)
+void compute_mu_uncertainty(const float *mu, const vector<int> &players)
 {
        // FIXME: Use pseudoinverse if applicable.
        Matrix<float, Dynamic, Dynamic> ih = hessian.inverse();
@@ -235,20 +251,49 @@ void compute_mu_uncertainty(const float *mu, const vector<string> &players)
                mu_stddev[i] = sqrt(ih(i, i));
        }
 
+#if USE_DB
        for (unsigned i = 0; i < players.size(); ++i) {
                for (unsigned j = 0; j < players.size(); ++j) {
-                       printf("covariance %s %s %f\n",
-                              players[i].c_str(),
-                              players[j].c_str(),
+                       CovarianceDBTuple tuple;
+                       tuple.player1 = players[i];
+                       tuple.player2 = players[j];
+                       tuple.covariance = ih(i, j);
+                       covariance_db_tuples.push_back(tuple);
+               }
+       }
+#else
+       for (unsigned i = 0; i < players.size(); ++i) {
+               for (unsigned j = 0; j < players.size(); ++j) {
+                       printf("covariance %d %d %f\n",
+                              players[i],
+                              players[j],
                               ih(i, j));
                }
        }
+#endif
 }
 
-int main(int argc, char **argv)
+void process_file(const char *filename)
 {
+       global_sigma = 70.0f;
+       prior_sigma = 70.0f;
+       matches_for_player.clear();
+       all_matches.clear();
+
+       FILE *fp = fopen(filename, "r");
+       if (fp == NULL) {
+               perror(filename);
+               exit(1);
+       }
+
+       char locale[256];
+       if (fscanf(fp, "%s", locale) != 1) {
+               fprintf(stderr, "Could't read number of players\n");
+               exit(1);
+       }
+
        int num_players;
-       if (scanf("%d", &num_players) != 1) {
+       if (fscanf(fp,"%d", &num_players) != 1) {
                fprintf(stderr, "Could't read number of players\n");
                exit(1);
        }
@@ -258,27 +303,27 @@ int main(int argc, char **argv)
                exit(1);
        }
 
-       vector<string> players;
-       map<string, int> player_map;
+       vector<int> players;
+       map<int, int> player_map;
 
        for (int i = 0; i < num_players; ++i) {
                char buf[256];
-               if (scanf("%s", buf) != 1) {
+               if (fscanf(fp, "%s", buf) != 1) {
                        fprintf(stderr, "Couldn't read player %d\n", i);
                        exit(1);
                }
 
-               players.push_back(buf);
-               player_map[buf] = i;
+               players.push_back(atoi(buf));
+               player_map[atoi(buf)] = i;
        }
 
        int num_matches = 0;
        for ( ;; ) {
-               char pl1[256], pl2[256];
+               int pl1, pl2;
                int score1, score2;
                float weight;
 
-               if (scanf("%s %s %d %d %f", pl1, pl2, &score1, &score2, &weight) != 5) {
+               if (fscanf(fp, "%d %d %d %d %f", &pl1, &pl2, &score1, &score2, &weight) != 5) {
                        //fprintf(stderr, "Read %d matches.\n", num_matches);
                        break;
                }
@@ -286,11 +331,11 @@ int main(int argc, char **argv)
                ++num_matches;
 
                if (player_map.count(pl1) == 0) {
-                       fprintf(stderr, "Unknown player '%s'\n", pl1);
+                       fprintf(stderr, "Unknown player '%d'\n", pl1);
                        exit(1);
                }
                if (player_map.count(pl2) == 0) {
-                       fprintf(stderr, "Unknown player '%s'\n", pl2);
+                       fprintf(stderr, "Unknown player '%d'\n", pl2);
                        exit(1);
                }
 
@@ -310,6 +355,8 @@ int main(int argc, char **argv)
 
                all_matches.push_back(m1);
        }
+       
+       fclose(fp);
 
        float mu[MAX_PLAYERS];
 
@@ -317,6 +364,7 @@ int main(int argc, char **argv)
                mu[i] = PRIOR_MU;
        }
 
+       int num_iterations = -1;
        for (int j = 0; j < 1000; ++j) {
                float old_mu[MAX_PLAYERS];
                float old_global_sigma = global_sigma;
@@ -340,22 +388,105 @@ int main(int argc, char **argv)
                sumdiff += (global_sigma - old_global_sigma) * (global_sigma - old_global_sigma);
                if (sumdiff < EPSILON) {
                        //fprintf(stderr, "Converged after %d iterations. Stopping.\n", j);
-                       printf("aux_param num_iterations %d\n", j + 1);
+                       num_iterations = j + 1;
                        break;
                }
        }
 
-#if DUMP_RAW
-       dump_raw(mu, num_players);
-#else
        construct_hessian(mu, num_players);
+       aux_params[make_pair(locale, "num_iterations")] = num_iterations;
+       aux_params[make_pair(locale, "score_stddev")] = global_sigma / sqrt(2.0f);
+       aux_params[make_pair(locale, "rating_prior_stddev")] = prior_sigma;
+       aux_params[make_pair(locale, "total_log_likelihood")] = compute_total_logl(mu, num_players);
+
        compute_mu_uncertainty(mu, players);
        dump_scores(players, mu, mu_stddev, num_players);
-       //fprintf(stderr, "Optimal sigma: %f (two-player: %f)\n", sigma[0], sigma[0] * sqrt(2.0f));
-       printf("aux_param score_stddev %f\n", global_sigma / sqrt(2.0f));
-       printf("aux_param rating_prior_stddev %f\n", prior_sigma);
+}
+
+int main(int argc, char **argv)
+{
+#if USE_DB
+       pqxx::connection conn("dbname=wloh host=127.0.0.1 user=wloh password=oto4iCh5");
+#endif
+       
+       for (int i = 1; i < argc; ++i) {
+               process_file(argv[i]);
+       }
 
-       float total_logl = compute_total_logl(mu, num_players);
-       printf("aux_param total_log_likelihood %f\n", total_logl);
+#if DUMP_RAW
+       dump_raw(mu, num_players);
+       return 0;
 #endif
+
+#if USE_DB
+       pqxx::work txn(conn);
+       txn.exec("SET client_min_messages TO WARNING");
+
+       // Dump aux_params.
+       {
+               txn.exec("TRUNCATE aux_params");
+               pqxx::tablewriter writer(txn, "aux_params");
+               for (map<pair<string, string>, float>::const_iterator it = aux_params.begin(); it != aux_params.end(); ++it) {
+                       char str[128];
+                       snprintf(str, 128, "%f", it->second);
+
+                       vector<string> tuple;
+                       tuple.push_back(it->first.first);  // locale
+                       tuple.push_back(it->first.second);  // parameter name
+                       tuple.push_back(str);
+                       writer.push_back(tuple);
+               }
+               writer.complete();
+       }
+
+       // Dump ratings.        
+       {
+               txn.exec("TRUNCATE ratings");
+               pqxx::tablewriter writer(txn, "ratings");
+               for (unsigned i = 0; i < rating_db_tuples.size(); ++i) {
+                       char player_str[128], mu_str[128], mu_stddev_str[128];
+                       snprintf(player_str, 128, "%d", rating_db_tuples[i].player);
+                       snprintf(mu_str, 128, "%f", rating_db_tuples[i].mu);
+                       snprintf(mu_stddev_str, 128, "%f", rating_db_tuples[i].mu_stddev);
+
+                       vector<string> tuple;
+                       tuple.push_back(player_str);
+                       tuple.push_back(mu_str);
+                       tuple.push_back(mu_stddev_str);
+                       writer.push_back(tuple);
+               }
+               writer.complete();
+       }
+
+       // Create a table new_covariance, and dump covariance into it.
+       {       
+               txn.exec("CREATE TABLE new_covariance ( player1 smallint NOT NULL, player2 smallint NOT NULL, cov float NOT NULL )");
+               pqxx::tablewriter writer(txn, "new_covariance");
+               for (unsigned i = 0; i < covariance_db_tuples.size(); ++i) {
+                       char player1_str[128], player2_str[128], cov_str[128];
+                       snprintf(player1_str, 128, "%d", covariance_db_tuples[i].player1);
+                       snprintf(player2_str, 128, "%d", covariance_db_tuples[i].player2);
+                       snprintf(cov_str, 128, "%f", covariance_db_tuples[i].covariance);
+
+                       vector<string> tuple;
+                       tuple.push_back(player1_str);
+                       tuple.push_back(player2_str);
+                       tuple.push_back(cov_str);
+                       writer.push_back(tuple);
+               }
+               writer.complete();
+       }
+
+       // Create index, and rename new_covariance on top of covariance.
+       txn.exec("ALTER TABLE new_covariance ADD PRIMARY KEY ( player1, player2 );");
+       txn.exec("DROP TABLE IF EXISTS covariance");
+       txn.exec("ALTER TABLE new_covariance RENAME TO covariance");
+#else
+       //fprintf(stderr, "Optimal sigma: %f (two-player: %f)\n", sigma[0], sigma[0] * sqrt(2.0f));
+       for (map<pair<string, string>, float>::const_iterator it = aux_params.begin(); it != aux_params.end(); ++it) {
+               printf("%s: aux_param %s %f\n", it->first.first.c_str(), it->first.second.c_str(), it->second);
+       }
+#endif
+
+       txn.commit();
 }