I stumbled across this forum in midst of the COVID-19 pandemic and thought I'd use my free time at home to learn about CFR in hopes to ultimately write a script for poker bot.
I've used this
http://modelai.gettysburg.edu/2013/cfr/cfr.pdf to get me started and have already done the RPS and Kuhn poker examples.
Since then, I'm trying CFR out for a popular hand game in Taiwan we dub "Five Ten Fifteen" (FTF). The game is usually played in a drinking session where one person challenges another, where the loser needs to drink. The rules are as follows
Quote:
1. Two players RPS (Rock Paper Scissors) to see who goes first.
2. Players can either throw one of 3 moves with their HANDs simultaneously:
- 0 (both fists closed)
- 5 (one hand open, one closed)
- 10 (both hands open)
3. The "active" player (who won the RPS) needs to GUESS one of 5 possible counts that are possible from adding the total count of your hand + the opponents (16 total combos):
- 0 (both players have their fist closed / 1 combo)
- 5 (only one hand open / 4 combos)
- 10 (two hands open / 6 combos (2 + 4) )
- 15 (three hands open / 4 combos )
- 20 (all hands open / 1 combo)
5. If the guess is correct, the active player has to guess again. If it's incorrect, it switches to the opposing player.
6. The player who guesses right twice in a row wins.
So where I'm stuck is: starting from the root game node, the player first to act have actually:
3 (NUM_HANDS) * 5 (GUESSES) = 15 Actions branching out of the root. ( 9 if you disqualify invalids)
Hand 0 : Guesses: 0, 5, 10, (15), (20)
Hand 5 : Guesses (0), 5, 10, 15, (20)
Hand 10 : Guesses (0), (5), 10, 15, 20
Where as for the opponent who's defending, there is only 3 possible actions:
Actions:
Hand 0, Hand 5, Hand 10So they players actually take turns between selecting 15 actions or 3 actions. My question is how do you set things up to calculate the cumulative regret for the active player (hero) and the defending player (villain) for each rounds?
Do you need two sets of probability matrix to represent hero / villain ? If so how to arrive at a general strategy? Or is the only way is to have two sets of strategy for each situation?
Thanks in advance!
Login
Register
FAQ
Search
