Put reward scaling back in place

This is a temporary measure until #273 is resolved.

week5_policy_based/atari_util.py

@@ -7,7 +7,7 @@
 
 class PreprocessAtari(Wrapper):
     def __init__(self, env, height=42, width=42, color=False,
-                 crop=lambda img: img, n_frames=4, dim_order='theano'):
+                 crop=lambda img: img, n_frames=4, dim_order='theano', reward_scale=1,):
         """A gym wrapper that reshapes, crops and scales image into the desired shapes"""
         super(PreprocessAtari, self).__init__(env)
         assert dim_order in ('theano', 'tensorflow')
@@ -16,6 +16,7 @@ def __init__(self, env, height=42, width=42, color=False,
         self.color = color
         self.dim_order = dim_order
 
+        self.reward_scale = reward_scale
         n_channels = (3 * n_frames) if color else n_frames
         obs_shape = \
             [n_channels, height, width] \
@@ -34,8 +35,8 @@ def step(self, action):
         """plays breakout for 1 step, returns frame buffer"""
         new_img, r, done, info = self.env.step(action)
         self.update_buffer(new_img)
-        return self.framebuffer, r, done, info
 
+        return self.framebuffer, r * self.reward_scale, done, info
     ### image processing ###
 
     def update_buffer(self, img):