monta@cmu-cs-g.ARPA (Peter Monta) (08/14/85)
I have seen several books about casino blackjack that present strategy by referring to the results of computer simulations. Very few details, though. Are there accessible accounts of how these simulations are carried out? Peter Monta monta@cmu-cs-g.arpa ..!rochester!cmu-cs-pt!cmu-cs-g!monta
massar@think.ARPA (JP Massar) (08/15/85)
In article <300@cmu-cs-g.ARPA> monta@cmu-cs-g.ARPA (Peter Monta) writes: >I have seen several books about casino blackjack that present >strategy by referring to the results of computer simulations. >Very few details, though. Are there accessible accounts of how >these simulations are carried out? The best way is to simply carry out the computations directly: e.g., suppose one wishes to calculate what to do with a (9,6) vs a dealer's ten. Given the rules the dealer plays by and the number of decks, calculate the dealer's probabilities of getting various totals. Given this information, calculate the optimal play for the player based on the distribution of the remaining cards in the deck (shoe). Programmed recursively, the algorithm is simple to write, except for subtlties due to various obscure rule options one finds in different casinos. Actual simulation is incredibly slow and converges slowly to boot. A direct calculation in the worst cases takes only a few minutes on a standalone VAX. There are a few decisions which are very close calls and a simulation would not be guarenteed to come up with the correct answer for a very long time! One can then iterate over all possible dealer-player combinations of initial cards to generate optimal strategy given an initial deck configuration. I am pretty sure Thorpe essentially did this. Others may have done actual simulations, shuffling cards, dealing out hands, etc. -- -- JP Massar, Thinking Machines Corporation, Cambridge, MA -- ihnp4!godot!massar, massar@think.com.arpa -- 617-876-1111