[ADD]Advanced kl

nsm/env_factory.py

@@ -161,7 +161,7 @@ def step(self, action, debug=False):
     # Only when the proram is finished and it doesn't have
     # extra work or we don't care, its result will be
     # scored, and the score will be used as reward. 
-    reward_compile = 0.0
+    reward_compile = 1.0
     if self.done and not (self.punish_extra_work and self.interpreter.has_extra_work()):
       reward = self.score_fn(self.interpreter.result, self.answer)
     else:
@@ -173,7 +173,6 @@ def step(self, action, debug=False):
     if self.done:
       if self.interpreter.result is None:
         self.error = True
-        print('isNone')
       else:
         if len(self.interpreter.result) == 0:
           self.error = True
@@ -208,9 +207,9 @@ def step(self, action, debug=False):
       self.error = True
 
     if self.error == True:
-      reward_compile = -10.0
+      reward_compile = 0.0
 
-    reward = reward + reward_compile
+    reward = reward * reward_compile
     self.rewards.append(reward)
     ob = (tf_utils.MemoryInputTuple(
       read_ind=mapped_action, write_ind=new_var_id, valid_indices=self.valid_actions),