Strategy for rDrama blackjack

I was feeling understimulated, but I'm also too poor to spend much time in the casino. So I created a genetic algorithm for creating the "optimal" strategy, specifically for the rDrama casino. Note that I'm not much of a goombler, so I may get some terms wrong.

Using this strategy, over a million games, the house's edge is about 1.10±0.06%. That is an improvement over this one and this one, which net an edge of 1.63% and 1.52±0.07%, respectively.

There may still be better strategies, but this one summed up:

• Stand on a hard 17 and up (duh);

• Stand on a hard 16 through 12, if the dealer has a 6 or lower, except if you have 12 and the dealer has a 2;

• Stand on a soft 17 except if the dealer has a 9, 10 or ace;

• Double down on a 10 or 11 except if the dealer has an ace, or if both you and the dealer are showing 10;

• Hit otherwise.

details

I can't share the code as it's based on a proprietary algorithm unfortunately.

The GA, a real-valued EDA to be precise, evolves a 241-valued genome. 160 values for the hard-hand table, 80 values for the soft table, and one for insurance. Especially the hard table could have been smaller and insurance is well-known to not be worth it in the long run, but 1) it's good to have some easy parts for sanity checking and 2) maybe the algorithm could find an exploit But it didn't.

Because I only had a real-valued GA at hand, each value mapped to a probability to stand, hit, double-down-or-hit or double-down-or-stand. The probabilities are roughly sinusoidal and therefore repeat to let the algorithm reach "far away" actions easier (such as when going from double-down-or-stand to stand).

Each strategy played 100 000 randomly-generated games. Population sizes started at 40 and doubled with every new population up to 640. The GA used a scheme that makes counting actual evaluations difficult, but I'd estimate about 500 billion games have been played.

Why did I go through all this effort again? Like I'm on vacation from precisely this kind of shit. Goddarn