Learning the attack phase

Edouard360 / Halite-Python-RL

A fork of the Halite Starting Kit, aimed at providing an interface and debugging tools and for RL strategies (reinforcement learning).

6 stars 2 forks source link

Learning the attack phase #19

Open Edouard360 opened 7 years ago

Edouard360 commented 7 years ago

We should observe that a bot that learns to fight, learn some specific strategies:

To wait for the enemy to come before attacking himself !

Be careful, because the existing reward system will need to be changed. Indeed when a square is attacked their is a loss in production, but it is not attributed to any square...

leonardhussenot commented 7 years ago

cf learning Issue : in order to learn the attack phase, we need to train on terminating games (until last turn), and use final reward for highly rewarding victory/defeat . We should also rapidly train against enemies