From 247f646f02728477627274a883348f1d4e734d09 Mon Sep 17 00:00:00 2001
From: maberet <maberet@ada.local.isima.fr>
Date: Wed, 29 Jun 2022 21:09:44 +0200
Subject: [PATCH] correction formules qlearn
---
.../jeu_appren_par_renfo/src/main.c | 7 ++--
.../jeu_appren_par_renfo/src/qlearn.c | 33 +++++++++----------
.../jeu_appren_par_renfo/src/qlearn.h | 6 +++-
3 files changed, 25 insertions(+), 21 deletions(-)
diff --git a/travail_de_groupe/jeu_appren_par_renfo/src/main.c b/travail_de_groupe/jeu_appren_par_renfo/src/main.c
index 94fa800..bd35d29 100644
--- a/travail_de_groupe/jeu_appren_par_renfo/src/main.c
+++ b/travail_de_groupe/jeu_appren_par_renfo/src/main.c
@@ -6,12 +6,13 @@ int game_state;
int main(){
float ***** Q = allocateAndInitiateQ();
+ int i = 10000;
srand ( time(NULL));
-
- traningAgent(1000000, 15, Q);
+ while (i>0){
+ traningAgent(10000000, 40, Q);
+ i--;}
-
writeQ(Q);
// running = 1;
// game_state = GAME;
diff --git a/travail_de_groupe/jeu_appren_par_renfo/src/qlearn.c b/travail_de_groupe/jeu_appren_par_renfo/src/qlearn.c
index 38fce48..d5a9aa7 100644
--- a/travail_de_groupe/jeu_appren_par_renfo/src/qlearn.c
+++ b/travail_de_groupe/jeu_appren_par_renfo/src/qlearn.c
@@ -115,15 +115,15 @@ int argmax(float * arr){
int i;
float max = arr[0];
int maxIndex = 0;
- printf("argmax: %f ", arr[0]);
+ //printf("argmax: %f ", arr[0]);
for(i = 1; i < NUMBER_ACTION; i++){
- printf("%f ", arr[i]);
+ //printf("%f ", arr[i]);
if (arr[i] > max){
max = arr[i];
maxIndex = i;
}
}
- printf("\n");
+ //printf("\n");
return maxIndex;
}
@@ -361,7 +361,8 @@ void traningAgent ( int numberRun, int numberStep, float *****Q) {// pour avoir
zoneAngleH=converterIntoAngleH(angleH);
dropZone=convertIntoZone(dropPoint.x,dropPoint.y);
canonZone= convertIntoZoneCanon(canon.x,canon.y);
- reward=0;
+ reward=0;
+ //printf("%d %d %d %d \n",dropZone, canonZone,zoneAngleH,zoneAngleF);
for (i=0; i<numberStep-1;i++){
action = takeAction(agent->x,agent->y,Q,canonZone,zoneAngleH,zoneAngleF,greedy);
@@ -374,28 +375,25 @@ void traningAgent ( int numberRun, int numberStep, float *****Q) {// pour avoir
line.reward=0;
actionStack(stack,line);
moveAgent(agent, action);
- printf("wtf%d \n ",i);
}
- action = takeAction(agent->x,agent->y,Q,canonZone,zoneAngleH,zoneAngleF,greedy);
+ action = takeAction(agent->x, agent->y,Q,canonZone,zoneAngleH,zoneAngleF,greedy);
agentZone = convertIntoZone(agent->x, agent->y);
line.receiverZone=agentZone;
line.shooterZone =canonZone;
line.angleHZone= zoneAngleH;
line.angleFZone= zoneAngleF;
line.action= action;
- line.reward=0;
- actionStack(stack,line);
- moveAgent(agent, action);
+
if (agentZone==dropZone){
- reward=1;
+ line.reward=1;
}
- else{reward= 0;}
-
- printf("wtf2\n ");
+ else{line.reward= 0;}
+ actionStack(stack,line);
+ moveAgent(agent, action);
Q[line.receiverZone][line.shooterZone][line.angleHZone][line.angleFZone][line.action] +=
- + greedy* ( reward- Q[line.receiverZone][line.shooterZone][line.angleHZone][line.angleFZone][line.action] );
+ + LEARN_RATE* ( line.reward - Q[line.receiverZone][line.shooterZone][line.angleHZone][line.angleFZone][line.action] );
while (!emptyStack(stack)){
maxAction= argmax(Q[line.receiverZone][line.shooterZone][line.angleHZone][line.angleFZone]);
@@ -403,12 +401,13 @@ void traningAgent ( int numberRun, int numberStep, float *****Q) {// pour avoir
line=unStack(stack);
Q[line.receiverZone][line.shooterZone][line.angleHZone][line.angleFZone][line.action] +=
- + greedy* ( reward + LEARN_RATE *Q[line.receiverZone][line.shooterZone][line.angleHZone][line.angleFZone][maxAction]
+ + LEARN_RATE* ( reward + DISCOUNT*Q[line.receiverZone][line.shooterZone][line.angleHZone][line.angleFZone][maxAction]
- Q[line.receiverZone][line.shooterZone][line.angleHZone][line.angleFZone][line.action] );
}
- printf("wtf1 \n ");
+ //printf("is empty : %d\n ", emptyStack(stack));
numberRun--;
greedy=greedy-1/((float)numberRun);
- printf("wtf1 \n ");
+
+ if ( numberRun%1000000==1){printf (" %d \n ", numberRun);}
}
}
\ No newline at end of file
diff --git a/travail_de_groupe/jeu_appren_par_renfo/src/qlearn.h b/travail_de_groupe/jeu_appren_par_renfo/src/qlearn.h
index 5d4fc87..5714908 100644
--- a/travail_de_groupe/jeu_appren_par_renfo/src/qlearn.h
+++ b/travail_de_groupe/jeu_appren_par_renfo/src/qlearn.h
@@ -10,7 +10,11 @@
#define M_PI 3.14159265358979323846
-#define LEARN_RATE 0.9
+#define LEARN_RATE 0.8
+#define DISCOUNT 0.5
+
+
+
#define NUMBER_ZONE_SHOOTER 4
#define NUMBER_ZONE_RECEIVER 4
--
GitLab