Data

Hi,

I am a serious player of BJ and I am interested in carrying out my own research.

However, I am a tad skeptical of computer generated data as one mistake in the software will cause the whole if the data set to be skewed.

I wondered if there was any manually generated data around using six decks and basic strategy.

Anthony

First, the number of permutations is so huge that you couldn't gather enough data manually in a lifetime to even make a dent.

Second, you should look at it the way science looks at everything else: independent researchers operating independently arriving at the same conclusion(s).

No, I am sorry you are wrong.

It is a popular falicy in the BJ world that anything less than 2 billion hands will not do.

Statistics has been around for hundreds of years and professional statiticians have spent a great deal of time developing ways of obtaining results on a small amount of data to express limitless real data.

For example: AIDS is a real worldwide threat. The use of condoms can help prevent its spread. Imagine the number of condoms used in the world every year.

Now WHO (WORLD HEALTH ORGANISATION) wanted to know if the AIDS virus could pass through condoms. So they experimented and tested condoms. They didn't test 2,000,000,000 or 2,000,000: they only tested 500.

That's right the results from 500 would be sufficient to demonstrate the effectiveness of the billions of real life condoms used in the next few decades. And these are some of the best minds in the world who won't be fooled by erroneous data.

If this is possible it is also possible to identify BJ trends with only say 100,000 hands!!

Anthony

If you can find it, and you probably can't, get hold of The Casino Gamblers Guide, out of print for going on four decades, for a lexicon of statistical shortcuts Wilson used to analyze the 21 game before computers were available. But don't get your hopes up.

I once set a goal, three decades ago, to confirm, by manual iterative process, at least a portion of single deck basic strategy. In reviewing every document in my possession as a function of my recent divorce, I recently ran across a portion of these work papers, some of which bore the Crayola arts of my kids. As I recall, Ed Thorp stated in Beat the Dealer that manual calculation of what he had done with early IBM computers would require 10,000 man years. On the other hand, Jess Marcum, the little dark-haired guy from Southern California, constructed his Ten Count methods, which apparently threw off a 3% advantage, in a fraction of a decade.

But the blackjack soil is much less fertile now and you should be assigning some realistic value to your time.

the obvious question is, do you understand variance? The variance on AIDS penetrating a condom is far less than the variance on the game of 21. The more variance the more trials are needed. Simple concept. And since 21 is not a low-variance game, 200-300 hands is not going to tell you _anything_ worthwhile. That's why Thorp used a computer in the 60's, and why everyone else is using them today. If you don't trust computer results, there's little one can say in response. But if you try to produce manual results, they are going to make those computer results look impossibly accurate because the manual results are going to be bad...

Thankyou: having a degree in maths I do understand the concept of variance. And when I mentioned around 100,000 hands I didn't mean 200 or 300 hundred!!

Believe me, there are proven mathematical methods to eliminate the need for billions of hands played. If something will work or not, it will show itself very quickly.

A player would have to start playing before the pyramids were built to play 2 billion hands - does that really sound feasible?

Perhaps the methods that have been produced previously require so much data because they themselves are not very good techniquies.

BJ seems to be stuck in a rut; I have developed an excellent alternative.But being a scientist I am concerned about any systematic errors in my own data set.

That is why I am asking for independent data.

Anthony

Yes, it is true that software errors can have big consequences in Gaming.

I had an interesting chat with Stanley Ko a few years ago. Ko is a first rate gaming analyst, who unfortunately works for the Dark Side.
He told me that he worked in the software industry, his programs were expected to have bugs. If clients found bugs, you would fix them and release another version. It went with the territory.

But in Gaming, your software cannot have bugs. If you tell your followers that they have a 1.8% edge, they are going to be Mad as Hxxx if you later discover a bug, and they had a negative 1% advantage. Ko told me he would laya wake at night worrying about bugs.

That is why it is essential to develop strong anti-bugging techniques. You need to thoroughly test all the modules, and build internal checks into them. Have them actively look for internal consistency.
Another thing you can do is try to solve the same problem two different ways. For example, after doing a Combinatorial Analysis, also do a Sim to verify it.

This does not catch conceptual errors. If you have made an error in understanding in formulating the problem, that error may effect both the CA and the Sim.

This could also effect your �manual� Simulation as well. In fact, it is much easier to create errors with a manual sim than with a program.

> Now WHO (WORLD HEALTH ORGANISATION) wanted to know if the AIDS
> virus could pass through condoms. So they experimented and tested
> condoms. They didn't test 2,000,000,000 or 2,000,000: they only
> tested 500.

Number of permutations in the above example: Two. Either the virus will pass through condoms or it won't. Thus the example is hardly relevant here.

Now consider how may different ways a player can get a hand of 15, and how many different hands the dealer can have with an upcard of Seven, and calculate the probability of each. The number of permutations and associated probabilities will run into the tens if not hundreds of millions, and each must be taken into account to get an absolutely accurate determination of how the player should play the hand of 15 vs. 7.

Now, can some shortcuts be taken? Sure. You could get approximate probs of the result of hitting, standing, doubling the 15 at each count, and average probs for each final hand for the dealer at each count, and then cross-multiply to get approximate win, push, lose probs.

But the key word there is "approximate." With today's computing power, even using brute force methods, getting EXACT information is really rather easy for any competent programmer.

Thanks Oldcoot, but I simply used that example to point out that if the statsman knows his job, it is possile to use a small amount of data against a large real world population.

I would also point out that there are only 3 outcomes for the player at the table - W,L or Tie. Really, by looking at these it is possible to identify where there are more wins than losses.

I prepared my method for autoshuffle machines originally but have applied it to manual shoes.It is possible to win without counting.

My point was this: the fewer rounds you play, the bigger the error term in estimating your EV. If you play enough (in the sim) then you know very accurately what the EV actually is. Will you as a player reach this EV? Hard to say, but N0 of 40K rounds is pretty common, which means you only need 40K rounds in a good game to be reasonably certain of landing close to the expected value...

If your sim is based on 100K or 200K rounds, your derived EV will have a larger error component. I'd personally prefer to know the EV as accurately as possible, and have knowledge of the variance so that I can estimate with some degree of accuracy how close to that EV I can actually come, if I play a more reasonable number of rounds (rather than 10B or whatever)...

Personally I'd like my sims to be based on a trillion or more rounds, to ensure I am getting the most accurate answer possible... I still have to deal with error when I estimate how close my actual EV will be to the simulated EV since my actual EV will be based on far fewer hands. But now I only have one error term, since the simulated EV will be quite accurate.

You put your life in the hands of simulations every day. Your automobile has components that are sized according to simulation results. If they are wrong, you lose a wheel at 70mph and the outcome will likely be fatal. No need to mistrust these kinds of results at all. I guess, basically, I don't see where you are coming from. I will add that I am looking at this from a computer scientist type perspective, so I've done tons of simulations myself...

TD strategies are known and can be calculated exactly without sims, including splits now. If the values have been arrived at independently by multiple people, I can't imagine why you're so worried about it.

That being said, most sources I've seen gives EV's in terms of CD play and splits are estimated. BJA3 has TD values for splits and overall game EV's and my CA gives TD values for individual hands as well.

Anyways, if you really don't trust other people's software - look at Eric Farmer's and customize it to be your own and you can do your own debugging.