keithgw / wingspan

A Wingspan AI that uses reinforcement learning to learn to play the wingspan board game.
2 stars 1 forks source link

set game state #45

Closed keithgw closed 7 months ago

keithgw commented 7 months ago

For MCTS, the agent will need to be able to "playout" the game from the current game state. This will require that a game can be initialized from the current game state.

keithgw commented 7 months ago

There is a difference between setting the game state for the real game and for the simulation. For the simulation, one would not want to presume perfect knowledge of which cards are in the deck or other players' hands.

keithgw commented 7 months ago

There is a difference between setting the game state for the real game and for the simulation. For the simulation, one would not want to presume perfect knowledge of which cards are in the deck or other players' hands.

This should be handled in the feature vector -> game state translation. As long as the translated game state is a valid and possible game state from the known information, then the exact game state should be set for the purpose of simulation.

keithgw commented 7 months ago

Closed with #54