In Indian Poker, a card is dealt to each player. They do not see this, but instead see the other player's card. They players then make bets in the hope that they have the higher card. At the end of the round the cards are revealed and the higher card wins the pot. PPL-oker combines card counting with learning the opponent's perception of risk to propose and call/fold on bets in this game.
We used a probabilistic programming language to model the bayesian network governing the opponents bet. A player will see the opponent's card and either think that they have a good chance of winning or believe that they will lose the round. If they think a loss is likely then they could either bet to minimise this loss or could bluff, betting high to scare the opposition into folding. These three outcomes have been modelled.
Integration between the website, the game and the learning was tough within the 24 hours. Also, learning to use PyMC3 for the probability modelling has been challenging.