vy007vikas / openai-gym-solutions Goto Github PK

When I try to run the code carRacing.py. I get the error (No registered env with id: CarRacing-v0).
Following version are used.
gum - 0.15.4
python - 3.6
tensorflow - 2.0.0

Kindly help me in resolving this issue.

Hi, I was going through your code and I have a few doubts regarding a few parts of the code.

I know you must be busy but I am writing my own code for breakout and could really use some help ASAP, so if you could take some time to help me out then that would be great

I want to ask:
Our neural net is taking in the state of the game and will have 6 ('OUT') output nodes corresponding to the six possible actions the agent can take, each of these values represent the q value of being in that state and taking that particular action. Now when I take a particular action from a state, lets say I take action=2. I now have the q_predicted for each state-action pair(i.e 6 values). However I can only produce the td-error for one particular state-action(i.e action=2). For the remaining actions I dont have a td target and hence wont have a td error that can be used for backpropagating and optimizing the neural net.

In line 103 yval = np.zeros((1,OUT)) and then yval[0][action] = re + GAMMA*y you are saying that the td target for all other state actions is zero. Could you please explain this part??

Your help is much appreciated