Create DQN framework - Githubissues

weidler / RLaSpa

Reinforcement Learning in Latent Space

MIT License

5 stars 1 forks source link

Create DQN framework #7

Closed weidler closed 5 years ago

adrigrillo commented 5 years ago

I created a better implementation of the dqn, more decoupled to use different environments. Anyway, after reading about this topic I have planned to improve the algorithm in different ways:

double q-learning: reduce the biased estimations. Paper
prioritized replay experience: to give importance to plays where the agent learn more. paper

I think this to things can be done this weekend. Explaining it could be more difficult.

adrigrillo commented 5 years ago

Implemented... For what I tried in the gym tasks the one I liked the most is the dueling-ddqn