ardavans / dsr Goto Github PK

desired_samples is not defined and this gives an error. I guess this can't be self.num_samples because even with that it throws an error in sample() function (TransitionTable ) because the batchSize is greater than the bufferSize.

Cheers

Hello,

I'm currently working on my master thesis and was wondering if you know of a repository that has implemented the same in keras or python. I'm trying to apply that on a simple 3D gridworld example. However, it does perform really poor and the network is really instable. Every run performs really different....

I have uploaded an example how I have rebuild the network. If you have time I would be very thankful if you can take a quick look into it. Maybe you see something wrong?

Here is the Github Link