search

calclavia / rl

Reinforcement learning algorithms implemented using Keras and OpenAI Gym
13 stars 1 forks source link

Training using recurrent networks? #7

Open kfeeeeee opened 7 years ago

kfeeeeee commented 7 years ago

Hi, I see that you have already thought of implementing some kind of "recurrentness" using the time_steps parameter.

Is it possible to train a stateful model (i.e., having data in shape as (batch_size, time_steps, features)) and resetting the state of, e.g., an lstm layer, for experience replay?

calclavia commented 7 years ago

Hmm. I don't have that implemented, but feel free to make a PR to add that feature! It think you should be able to by just using a one-step memory.