calculating EV from sim

[Super Natural]

I wrote a simulator in c# with the intention of analyzing blackjack variants.

As a baseline I wanted to sim a standard bj game to check the programs EV matches with expectation, using WoO and CVdata to compare results. 6D, H17, DAS, DOA, SPL3, NSA, LS, 1 deck cut off .. basic strategy EV -0.55%

To keep it simple I counted the number of payoff events and summed their p*EV to find the total EV. Is this not the way to calculate EV? After a 10b round sim the results are shown below.. EV -0.52% .. close but I figured it'd be dead on?

One potential problem I am having is 2.7% of the hands are split, compared to CVdatas 2.5%. I have walked through all split scenarios including handling of aces and I cant explain this one. I must be missing something obvious.

[DSchles]

I wrote a simulator in c# with the intention of analyzing blackjack variants.

As a baseline I wanted to sim a standard bj game to check the programs EV matches with expectation, using WoO and CVdata to compare results. 6D, H17, DAS, DOA, SPL4, NSA, LS, 1 deck cut off .. basic strategy EV -0.55%

To keep it simple I counted the number of payoff events and summed their p*EV to find the total EV. Is this not the way to calculate EV? After a 10b round sim the results are shown below.. EV -0.52% .. close but I figured it'd be dead on?

One potential problem I am having is 2.7% of the hands are split, compared to CVdatas 2.5%. I have walked through all split scenarios including handling of aces and I cant explain this one. I must be missing something obvious.

Attachment 4379

The precise house edge for this game, analytically, and not by sim is -0.529% (see BJA3, pp. 492-493). The difference you are seeing is due to the cut-card effect, which is used by the calculators of WoO and CVData, which do not provide traditional off-the-top values.

Also, am assuming that by SPL4 you mean split THREE times (SPL3) to make four hands total, right?

Don

[DSchles]

I wrote a simulator in c# with the intention of analyzing blackjack variants.

As a baseline I wanted to sim a standard bj game to check the programs EV matches with expectation, using WoO and CVdata to compare results. 6D, H17, DAS, DOA, SPL4, NSA, LS, 1 deck cut off .. basic strategy EV -0.55%

To keep it simple I counted the number of payoff events and summed their p*EV to find the total EV. Is this not the way to calculate EV? After a 10b round sim the results are shown below.. EV -0.52% .. close but I figured it'd be dead on?

One potential problem I am having is 2.7% of the hands are split, compared to CVdatas 2.5%. I have walked through all split scenarios including handling of aces and I cant explain this one. I must be missing something obvious.

Attachment 4379

Two more things: 1) Your attachment led nowhere; 2) you mentioned "one deck cut off." Did you run your sim by dealing through the cut card? If so, then you introduced the same cut-card effect, but it is different for different levels of penetration. When comparing to another sim, you'd have to make sure that the levels of pen were identical. This is one reason that we usually quote BS edge as off the top. That way, there is uniformity in expressing the value, even if it doesn't conform to real-world play. It standardizes the approach.

Don

[Super Natural]

Thanks Don, those pages are helpful.

I thought WoO and CVdata are including the cut card affect .. especially CVdata considering it is a sim and not a calculator.

WoO optimal result = 0.527%
Your analytical value = 0.529%

WoO basic strategy with cut card = 0.550%
CVdata sim = 0.565% (10b rounds shuffled after 260 cards)

Am I correct in assuming my sim should also be closer to -0.565? I am also shuffling after 260 cards.

Yes SPL3 to 4 hands, my bad. Could you point me towards any figures to check that my total number of splits is correct? Mine are consistently higher than the sims in CVdata. 270 million compared to 250 million (10b rounds).

I fixed the image in my first post.

[Norm]

Yes, CVData is including the CC effect.

[DSchles]

Am I correct in assuming my sim should also be closer to -0.565? I am also shuffling after 260 cards.

Yes.

Could you point me towards any figures to check that my total number of splits is correct?

Unfortunately, no. It would have to be done by simulation, because the analytical frequencies given in the appendixes of my book or Wong's just give the initial frequency of each pair vs. each dealer's upcard, but this would not give the extra info for resplits. I wonder if that isn't the problem. Maybe one way of doing it is counting only the original pair splits and not the resplits.

Don

[Dog Hand]

<snip>Could you point me towards any figures to check that my total number of splits is correct? Mine are consistently higher than the sims in CVdata. 270 million compared to 250 million (10b rounds).

I fixed the image in my first post.

I


Super Natural,

I just ran a 1-billion-round CVData sim for a 6D H17 DA2 DAS NoRSA game with 75% pen and a single Complete B.S. player, and found that he played 1,027,870,016 hands, which seems consistent with your program's result.

By the way, your handle sounds like a new BJ sidebet ;-)

Hope this helps!

Dog Hand

[Dog Hand]

Oops... I missed that your game allows LS.

With LS (and 5/6 pen), the billion rounds consist of 1,025,126,320 hands.

Maybe your code failed to account for the LS hands?

Dog Hand

[PinkChip]

Thanks Don, those pages are helpful.

I thought WoO and CVdata are including the cut card affect .. especially CVdata considering it is a sim and not a calculator.

WoO optimal result = 0.527%
Your analytical value = 0.529%

WoO basic strategy with cut card = 0.550%
CVdata sim = 0.565% (10b rounds shuffled after 260 cards)

Am I correct in assuming my sim should also be closer to -0.565? I am also shuffling after 260 cards.

Yes SPL3 to 4 hands, my bad. Could you point me towards any figures to check that my total number of splits is correct? Mine are consistently higher than the sims in CVdata. 270 million compared to 250 million (10b rounds).

I fixed the image in my first post.

Two questions, just to be sure:
1.) NSA means "No Split of Aces" (aces may not be split at all) or "No Resplit of Aces" (aces may be split once)?
2.) "Shuffling after 260 cards" means that after dealing 260 cards, the current round is still finished, so e.g. 265 cards might be dealt?

[DSchles]

Two questions, just to be sure:
1.) NSA means "No Split of Aces" (aces may not be split at all) or "No Resplit of Aces" (aces may be split once)?
2.) "Shuffling after 260 cards" means that after dealing 260 cards, the current round is still finished, so e.g. 265 cards might be dealt?

Not the OP, but I'm assuming he meant NRA for the first, and that the cut card is inserted at 5 decks (260 cards) and once reached, the round is completed.

Don

[Norm]

On penetration, CVData has five settings:

-Fixed cut card
-Fixed rounds
-Variable cut card
-Cut card and rounds
-Cut card or rounds

with additional settings for some of these.

Also, there are settings for:
-Dealer must complete hand (required for some bonus/side bets)
-Dealer shows burn cards
-Bottom card dealt

[Super Natural]

Maybe one way of doing it is counting only the original pair splits and not the resplits.

Yes thanks I will test this.

Maybe your code failed to account for the LS hands?

Yeah I tried to account for LS but maybe incorrectly, thanks Dog.

Two questions, just to be sure:
1.) NSA means "No Split of Aces" (aces may not be split at all) or "No Resplit of Aces" (aces may be split once)?
2.) "Shuffling after 260 cards" means that after dealing 260 cards, the current round is still finished, so e.g. 265 cards might be dealt?

Ace split once, I should have said NRA. Before each round I check if the cards remaining in the shoe are less than the cut card placement (52), 265 cards could be dealt.


Thanks everyone for your insight I will keep working on this. I did match the number of hands between my code and CVdata when allowing ace resplits/draws and LS off .. so I know it's related to those.

[Super Natural]

By the way, your handle sounds like a new BJ sidebet ;-)

Dog Hand

I like it, sounds very countable..