From: Steinar H. Gunderson <sgunderson@bigfoot.com>
Date: Mon, 19 Mar 2012 13:44:58 +0000 (+0100)
Subject: When doing Monte Carlo, player strength should be constant within a simulated round.
X-Git-Url: https://git.sesse.net/?p=wloh;a=commitdiff_plain;h=d76a4688b44cd0b1fcef71ea52d5a70bc9df3d00

When doing Monte Carlo, player strength should be constant within a simulated round.
---

diff --git a/mcwordfeud.cpp b/mcwordfeud.cpp
index b48936d..ce9dc5f 100644
--- a/mcwordfeud.cpp
+++ b/mcwordfeud.cpp
@@ -14,7 +14,7 @@ using namespace std;
 
 #define MAX_PLAYERS 16
 
-float match_variance = 70.0f * 70.0f;
+float match_stddev = 70.0f;
 
 struct player {
 	int player_index;
@@ -67,15 +67,12 @@ float draw_gaussian(float mu, float stddev)
 int main(int argc, char **argv)
 {
 	int trials = atoi(argv[1]);
-	float match_stddev;
 
 	if (scanf("%f", &match_stddev) != 1) {
 		fprintf(stderr, "Could't read match stddev\n");
 		exit(1);
 	}
 
-	match_variance = match_stddev * match_stddev;
-
 	int num_players;
 	if (scanf("%d", &num_players) != 1) {
 		fprintf(stderr, "Could't read number of players\n");
@@ -90,7 +87,7 @@ int main(int argc, char **argv)
 	vector<string> players;
 	map<string, int> player_map;
 	float ratings[MAX_PLAYERS];
-	float ratings_variance[MAX_PLAYERS];
+	float ratings_stddev[MAX_PLAYERS];
 	bool has_scores[MAX_PLAYERS][MAX_PLAYERS];
 	for (int pl1 = 0; pl1 < num_players; ++pl1) {
 		for (int pl2 = 0; pl2 < num_players; ++pl2) {
@@ -117,7 +114,7 @@ int main(int argc, char **argv)
 		players.push_back(buf);
 		player_map[buf] = i;
 		ratings[i] = rating;
-		ratings_variance[i] = rating_stddev * rating_stddev;
+		ratings_stddev[i] = rating_stddev;
 	}
 
 	for ( ;; ) {
@@ -151,6 +148,12 @@ int main(int argc, char **argv)
 	}
 
 	for (int i = 0; i < trials; ++i) {
+		// draw true strength for all players
+		float drawn_ratings[MAX_PLAYERS];
+		for (int p = 0; p < num_players; ++p) {
+			drawn_ratings[p] = draw_gaussian(ratings[p], ratings_stddev[p]);
+		}
+
 		// draw the missing matches
 		for (int pl1 = 0; pl1 < num_players; ++pl1) {
 			for (int pl2 = pl1 + 1; pl2 < num_players; ++pl2) {
@@ -158,10 +161,9 @@ int main(int argc, char **argv)
 					continue;
 				}
 
-				float mu = ratings[pl1] - ratings[pl2];
-				float stddev = sqrt(match_variance + ratings_variance[pl1] + ratings_variance[pl2]);
+				float mu = drawn_ratings[pl1] - drawn_ratings[pl2];
 				
-				int score = lrintf(draw_gaussian(mu, stddev));
+				int score = lrintf(draw_gaussian(mu, match_stddev));
 				scores[pl1][pl2] = score;
 				scores[pl2][pl1] = -score;
 			}