search

yuchen071 / DQN-for-LunarLander-v2

Implementation of reinforcement learning algorithms for the OpenAI Gym environment LunarLander-v2
MIT License
1 stars 0 forks source link

Not good result #2

Closed dugduy closed 1 year ago

dugduy commented 1 year ago

uhm, i also try to train this based on pytorch dqn tutorial. But the problem is that it is not stable.Eg: at the episode of 1000, some time it did pretty good, and others were too bad!!! i see your plot results and i think you also have that problem?

yuchen071 commented 1 year ago

Testing different parameters certainly would help! Did you train with my parameters? Or did you use your own? Batch sizes, learning rate, episodes, target score for early stopping, and gamma all have an effect on it Play with those and train until it you have fairly consistent landings You might have just landed on a bad training output 1000 episodes might not be enough too Good luck!