itaicaspi / keras-dqn-doom Goto Github PK

Hello,

I'm trying to run your code for prioritized experience replay and during the training part, often times the minibatch is an empty list which is caused by your indices being empty. Do you have any idea why this could be happening?

Thank you

For the video demo in the Readme for defend_the_center scenario, do you use single_agent or multi_agent? LSTM Deep Recurrent Q network is used right?

ValueError: Negative dimension size caused by subtracting 3 from 1 for 'conv2d_2/convolution' (op: 'Conv2D') with input shapes: [?,1,29,16], [3,3,16,32].

I don't seem to be able to get results mentioned in this repos README.md with the simple dqn over doomSimple with training from 10k episodes.
Please mention the hyperparameters used to obtain the results in this repos README.md

The following is the graph I obtain by training using only dqn for 10k episodes with hyperparameters provided in main.py last lstm variable