VPG it learns at a steady pase, by is it very slow.

I still like it better than DQN, because is more related to the dynamics of the model. or a least so far it seem that way. trying 10 m steps,
MADEAPPS · Oct 16, 2023 · e52c56e · e52c56e
1 parent 1171f2a
commit e52c56e
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/newton-4.00/applications/ndSandbox/demos/ndCartpoleDiscrete.cpp b/newton-4.00/applications/ndSandbox/demos/ndCartpoleDiscrete.cpp
@@ -99,9 +99,9 @@ namespace ndCarpole_0
 				,m_bestActor(m_actor)
 				,m_model(nullptr)
 				,m_timer(ndGetTimeInMicroseconds())
-				,m_maxGain(ndFloat32(- 1.0e10f))
+				,m_maxGain(ndFloat32(-1.0e10f))
 				,m_maxFrames(5000)
-				,m_stopTraining(4000000)
+				,m_stopTraining(10000000)
 				,m_modelIsTrained(false)
 			{
 				SetName("cartpoleVPG.dnn");