Commit Graph

3 Commits

Author SHA1 Message Date
cbc385ca10 Change training dataset storage method
save training dataset by it target type.
while training NN use single target training set to backward NN.
this improve at least 20 times faster than last update!
2022-12-03 07:54:38 +09:00
895cd5c118 Add EndReward Broadcast function
while game over add remaintime/15 to every step's rewards. to improve this round's training weight.
fix get target from states still using onehot decoder bug.
2022-12-03 03:58:19 +09:00
3930bcd953 Add Multi-NN agent
Add Multi neural network in output layer
use different nn while facing to different target.
2022-12-01 19:55:51 +09:00