I don't think I fully understand the "perfect EORs"
, but I don't think I explained myself well. The concepts of Playing Efficiency and Betting Correlation as defined in Theory of Blackjack are dated and we currently have better ways to calculate them.
We know that the best theoretical (theoretical because you can't use a computer at BJ table) gain from card counting that could be achieved is using Combinatorial analysis, where a) pre-deal advantage is calculated based on the shoe composition b) the playing decision is recomputed at every step for the player based on the dealer up card, the player's hand composition and the shoe composition.
In order to quantify that advantage for a number of decks, a set of the rules and a given penetration. In order to compute playing efficiency, we can run 2 sims one using Combinatorial Analysis to recompute the playing strategy every time the player needs to make a decision, and one using indices for a given counting system. For betting strategy, we will use flat betting for both sims, because this way we can compute the gain from using Combinatorial Analysis solely to "different and better" playing decisions compared to the ones dictated by index play. Comparing the SCOREs of those two sims we will be able to compute the Playing Efficiency of the counting system which is by definition how well a system can handle changes in playing decisions.
Calculating betting correlation is bit trickier and cannot be calculated using the same method because we cant compute the pre-deal EV using one method and then use a different playing strategy. For example, we cant use Combinatorial analysis to compute the pre-deal EV and then play using Hi-Low indices. I will need to think a little bit more about computing betting correlation, but one method would be to calculate the pre-deal EV and compare it to expected EV using True Count units for that system for a given penetration.
As far as the sims, you are right they are really slow, your best bet is to use heavy parallel computing, in the sim for 1 million shoes (not nearly enough shoes of course to get a good quantitative number), i ran on 256 cores for about 8 days! I would like to port the code to CUDA to run it on GPU, but really dont have a lot of time for BJ and would like to spend the free time to do something with higher return
Bookmarks