tweak

js/agent.js

@@ -10,9 +10,9 @@ function Agent(opt, world) {
     this.infos = []
     this.maxInfos = 2000
 
+    this.steps = 0
     this.timer = 0
     this.timerFrequency = 60 / this.frequency
-    this.resetFrequency = 30 / this.frequency
 
     if (this.options.dynamicallyLoaded !== true) {
     	this.init(world.brains.actor.newConfiguration(), null)
@@ -39,19 +39,16 @@ Agent.prototype.init = function (actor, critic) {
 
         temporalWindow: temporal, 
 
-        discount: 0.97, 
-        rate: 0.002,
+        discount: 0.97,  // time discount
+        rate: 0.001,  // learning rate,
         theta: 0.05, // progressive copy
         alpha: 0.1, // advantage learning
 
-        experience: 75e3, 
         // buffer: window.neurojs.Buffers.UniformReplayBuffer,
+        experience: 3e3, 
 
         learningPerTick: 40, 
         startLearningAt: 900,
-
-
-
     })
 
     // this.world.brains.shared.add('actor', this.brain.algorithm.actor)
@@ -69,6 +66,7 @@ Agent.prototype.step = function () {
     this.timer++
 
     if (this.timer % this.timerFrequency === 0) {
+        this.steps++
         var [state, reward, done, info] = this.walker.simulationStep()
 
         if (done) {
@@ -80,6 +78,7 @@ Agent.prototype.step = function () {
         // train
         info.loss = this.brain.learn(reward) 
         this.action = this.brain.policy(state)
+        info.x = this.steps
         this.infos.push(info)
     }
     if (this.action) {