FARKLE NOW HOW TO
FARKLE NOW CODE
FARKLE NOW SERIES

It is essentially table for mapping the optimal choice for a given dice roll. So now I have a way to model player behavior and a way to evolve it, but how do I find the top players? You may recall from the first blog that I noted that the player move table is simplistic it does not allow for the consideration of changing player strategy depending on where they are in the game and how many points they’ve already scored in their turn. In other words, maybe we can make the next generation of players by randomly combining the move tables of the top players from the previous generation.

FARKLE NOW CODE

To further extend the genetic metaphor, we can then “breed” a new generation of Farkle players by mixing the genetic code of the best players of the current generation.

In our case, the behavior we are trying to optimize for is an effective Farkle strategy and the “genetic code” of a player is their table of move choices. This is a genetic algorithm: a programmatic approach to trying to replicate the effects of evolution and natural selection to optimize for a behavior we are seeking.

FARKLE NOW SERIES

With 853 separate parameters to tune, the parameter optimization loop approach I covered in my Making the Pass blog series won’t work – it is simply impossible without near infinite compute resources.īut maybe we can take a similar approach as the parameter tuning nodes where we start with a series of random players and score each of their performances? Then we could pick the best scoring players and attempt to extract the best aspects from them for a new set of players to test.

FARKLE NOW HOW TO

But if we consider that each unique scoring dice roll has on average 3.68 possible choices for how to proceed and an associated Yes/No choice for if the player continues rolling, the total number of players required is in the order of 2 x 3.68 ^852 or approximately 2.53 × 10 22. Ideally, we could try every combination of these choices to determine which one results in the best overall win rate. In KNIME, we can represent this mapping as a set of rows with just a few columns: the dice roll, the points scored, the number of dice to re-roll, and whether or not to just take the score or continue to roll. The combination of these discoveries led me to the realization that a Farkle player’s behavior can be simplistically modeled as a mapping from each of these possible dice combinations to one of the valid moves for that combination and the continue rolling choice. Furthermore, there are only 3,150 possible moves for this set of combinations and after each move there is a Yes or No choice to continue rolling. From this initial analysis, I discovered that there are only 852 unique dice combinations that score points. In the first part of this blog series, I introduced the dice game called Farkle and explored some of its basic statistics related to the combinatorics of rolling 6 dices and choosing how to score them in the optimal manner.

Farkle now