search

spaceuniverse / QLSD

Q-learning sandbox with agent, enemies, shooting and fun 2.0 (Python)
2 stars 0 forks source link

LSTM #7

Open spaceuniverse opened 10 years ago

spaceuniverse commented 10 years ago

For now agent knows only current state, and does not remember any past. May be LSTM can improve this. For example if agent remembers that last move was around bullet, the next one will be on the optimal curve and not just by straight line.

Delicate-aRt commented 10 years ago

This required to develop a method, how NN or RNN could be trained as decision-makers. On 4 Sep 2014 15:33, "spaceuniverse" notifications@github.com wrote:

For now agent knows only current state, and does not remember any past. May be LSTM can improve this. For example if agent remembers that last move was around bullet, the next one will be on the optimal curve and not just by straight line.

— Reply to this email directly or view it on GitHub https://github.com/spaceuniverse/QLSD/issues/7.