Poker Expected Hand Strength Generalization Competition
Poker Hands have been used extensively as datasets – they provide both a fun and interesting way of linking purely categorical data to classes – an example is here.
We are taking this a step further and asking you to predict the mean strength of a 2-card hand plus a five card board under any table size given as few data instances as possible. Each seven card combination plus the number of players in the table is used as input, while a real number is given as output.
The competitor’s goal is to train an algorithm that learns to generalise as well as possible, given as little data as possible and we will thus provide the data-generating functions. Ranking will be multi-objective, taking into account both the amount of data used and the regression accuracy, with generalization ability always tramping pure performance!
1st 15th September, 2015