Qlearn draft

js/ai/qlearn/changelog.md

@@ -0,0 +1,8 @@
+# changelog
+
+## 6.0.0
+
+ * Start a changelog
+ * Change the learn calculation
+ * Up to ecma2021
+ * rename actionNames into previousActions inside learn

js/ai/qlearn/examples/dodgeShoot/dodgeShoot.js

@@ -30,6 +30,7 @@ const actions = {
         actor[0] = futureX;
     },
 };
+// do not change over time
 const actionNames = Object.keys(actions);
 
 const updateGame = (action, state, reward) => {
@@ -102,7 +103,7 @@ const start = (options) => {
                     previousStateActions,
                     stateActions,
                     previousAction: actionName,
-                    actionNames,
+                    previousActions: actionNames,
                     reward: scoreDifference,
                 });
             }

js/ai/qlearn/examples/dodgeShoot/results2.txt

@@ -1,31 +1,31 @@
 random positive reward 2000 frames score: 123  
 qLearn(reduceStateAndActionSeeAll)(learn) positive reward 2000 frames score: 463  
 qLearn(reduceStateAndActionSeeAll)(learnWithAverage) positive reward 2000 frames score: 465  
-qLearn(reduceStateAndActionSeeAllDistance)(learn) positive reward 2000 frames score: 420  
-qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) positive reward 2000 frames score: 214  
+qLearn(reduceStateAndActionSeeAllDistance)(learn) positive reward 2000 frames score: 266  
+qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) positive reward 2000 frames score: 240  
 qLearn(reduceStateAndActionSeeNearestOnly)(learn) positive reward 2000 frames score: 233  
-qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) positive reward 2000 frames score: 129        
 random negative reward 2000 frames score: -120  
-qLearn(reduceStateAndActionSeeAll)(learn) negative reward 2000 frames score: -18  
-qLearn(reduceStateAndActionSeeAll)(learnWithAverage) negative reward 2000 frames score: -23  
-qLearn(reduceStateAndActionSeeAllDistance)(learn) negative reward 2000 frames score: -83  
-qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) negative reward 2000 frames score: -128  
-qLearn(reduceStateAndActionSeeNearestOnly)(learn) negative reward 2000 frames score: -49
-qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) negative reward 2000 frames score: -60  
+qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) positive reward 2000 frames score: 137        
+qLearn(reduceStateAndActionSeeAll)(learn) negative reward 2000 frames score: -14  
+qLearn(reduceStateAndActionSeeAll)(learnWithAverage) negative reward 2000 frames score: -18  
+qLearn(reduceStateAndActionSeeAllDistance)(learn) negative reward 2000 frames score: -71  
+qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) negative reward 2000 frames score: -109  
+qLearn(reduceStateAndActionSeeNearestOnly)(learn) negative reward 2000 frames score: -19
+qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) negative reward 2000 frames score: -15  
 random positive reward 20000 frames score: 1241  
 qLearn(reduceStateAndActionSeeAll)(learn) positive reward 20000 frames score: 4963  
 qLearn(reduceStateAndActionSeeAll)(learnWithAverage) positive reward 20000 frames score: 4965
-qLearn(reduceStateAndActionSeeAllDistance)(learn) positive reward 20000 frames score: 4920  
-qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) positive reward 20000 frames score: 4168  
+qLearn(reduceStateAndActionSeeAllDistance)(learn) positive reward 20000 frames score: 3266  
+qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) positive reward 20000 frames score: 4740  
 qLearn(reduceStateAndActionSeeNearestOnly)(learn) positive reward 20000 frames score: 2333
-qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) positive reward 20000 frames score: 1438      
+qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) positive reward 20000 frames score: 2165      
 random negative reward 20000 frames score: -1235
-qLearn(reduceStateAndActionSeeAll)(learn) negative reward 20000 frames score: -180
-qLearn(reduceStateAndActionSeeAll)(learnWithAverage) negative reward 20000 frames score: -255
-qLearn(reduceStateAndActionSeeAllDistance)(learn) negative reward 20000 frames score: -554
-qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) negative reward 20000 frames score: -983      
-qLearn(reduceStateAndActionSeeNearestOnly)(learn) negative reward 20000 frames score: -549
-qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) negative reward 20000 frames score: -621      
+qLearn(reduceStateAndActionSeeAll)(learn) negative reward 20000 frames score: -28
+qLearn(reduceStateAndActionSeeAll)(learnWithAverage) negative reward 20000 frames score: -28
+qLearn(reduceStateAndActionSeeAllDistance)(learn) negative reward 20000 frames score: -95
+qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) negative reward 20000 frames score: -135      
+qLearn(reduceStateAndActionSeeNearestOnly)(learn) negative reward 20000 frames score: -19
+qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) negative reward 20000 frames score: -15      
 random positive reward 200000 frames score: 12442
 qLearn(reduceStateAndActionSeeAll)(learn) positive reward 200000 frames score: 49963
 qLearn(reduceStateAndActionSeeAll)(learnWithAverage) positive reward 200000 frames score: 49965
@@ -39,4 +39,4 @@ qLearn(reduceStateAndActionSeeAll)(learnWithAverage) negative reward 200000 fram
 qLearn(reduceStateAndActionSeeAllDistance)(learn) negative reward 200000 frames score: -4823
 qLearn(reduceStateAndActionSeeAllDistance)(learnWithAverage) negative reward 200000 frames score: -9479    
 qLearn(reduceStateAndActionSeeNearestOnly)(learn) negative reward 200000 frames score: -5707
-qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) negative reward 200000 frames score: -6299
+qLearn(reduceStateAndActionSeeNearestOnly)(learnWithAverage) negative reward 200000 frames score: -6299

js/ai/qlearn/examples/eatFoodAvoidDanger/eatFoodAvoidDanger.js

@@ -93,7 +93,7 @@ const start = ({ display, MAX_FRAMES }) => {
             previousStateActions: stateActions,
             stateActions: stateActionsAfter,
             previousAction: actionName,
-            actionNames,
+            previousActions: actionNames,
             reward,
         });
         frame += 1;
@@ -105,4 +105,4 @@ const start = ({ display, MAX_FRAMES }) => {
     };
 
     step();
-};
+};