search

floodsung / DQN-Atari-Tensorflow

Simplest Version of playing Atari with Deep Q Learning in Tensorflow
160 stars 56 forks source link

I found a bug. #3

Open shimaokasonse opened 8 years ago

shimaokasonse commented 8 years ago

Line 138 : newState = np.append(nextObservation,self.currentState[:,:,1:],axis = 2) This should be as follows: Line 138 : newState = np.append(nextObservation,self.currentState[:,:,:-1],axis = 2)

Fabse92 commented 6 years ago

Either this or newState = np.append(self.currentState[:,:,1:],nextObservation,axis = 2)

It is really important to fix this! Otherwise the first 3 images on the stack will always be the same.