Perfect BJ machine

[mleancole]

I'm guessing if a player has a BJ machine that can do the following:

1) Before each hand is dealt, it has to compute the player's advantage taking into account the undealt cards; with this advantage the player must bet according to a predefined ramp.

2) If the dealer has an ace as the upcard, it has to compute the number of tens and not tens so as to tell the player if he has to buy insurance or not.

3) Before each player decision, it has to evaluate the expected value of hitting, standing, etc. according to the undealt cards in order to recommend the appropiate move.

Can anyone tell me what will be the aproximate advantage the player will have? Let's assume any combination of decks and rules, what I want to know is the gap between this machine and any of the most used strategies.

Thanks in advance,

Mike LC

 

[itakeyourmoney]

Maybe I'm missing something here, but it seems to me that this is what you need:

A counting system, basic strategy, indexes.

Edit: If you're looking for a computer program that does all of this for you, you can look at Casino Verite at http://www.qfit.com.

 

[rrwoods]

Nono, I know what he means, and it would be really difficult.

He's looking for a system that knows *exactly* what cards are left in the deck, and *exactly* what plays to make. Such a system would have 100% betting correlation and 100% playing efficiency. It would also be impossible to use by a human, but I'm not sure that's what he's looking for anyway.

To answer the question: No, I don't think that's possible. I don't know much about combinatorial mathematics myself, but from what I do know, today's hardware isn't fast enough to give you decisions in a reasonable amount of time.

 

[johndoe]

This is usually called "perfect play", and if I remember correctly is only marginally better than playing with a reasonable number indicies. (Depending on game, of course.)

 

[mleancole]

rrwwods: That's exactly what I'm asking.
johndoe: Do you think this perfect machine would only be marginally better than a good counting system? Regarding betting correlation it's clear that it's a marginal increment, but what about playing efficiency? And insurance correlation? Wouldn't it be a big gain?

By the way, I use hi-lo with I18.

Thanks,

Mike

 

[johndoe]

No, I don't think the difference is very much, except if pen is very deep on a SD/DD game. I remember seeing the graphs somewhere, perhaps on QFIT's site.

In general, and especially in shoe games, there's so much uncertainty already that perfect play doesn't provide you much more information than you have already.

It definitely does help, don't get me wrong, but not by a whole lot.

 

[itakeyourmoney]

May I ask why you are asking in the first place?

Is it just curiousity? Because you'll be playing online blackjack and want the best chance you can get of winning? Are you planning on bringing some kind of device into the casino?

 

[mleancole]

For Itakeyourmoney:

It's just theorical curiosity. I'm planning to develop a software for this "Perfect BJ machine" and I wonder about the final result.

I do not play online blackjack and bringing a device like this into a casino it's not of my interest, not to say it's ilegal (or not?).

I'd appreciate your opinions.

Mike

 

[Pro21]

I believe Keith Taft's BJ computers used Thorp's Ultimate count and of course new exact number of cards played and perfect insurance decisions.

They were very strong playing single deck and almost worthless on shoe games.

 

[k_c]

I'm guessing if a player has a BJ machine that can do the following:

1) Before each hand is dealt, it has to compute the player's advantage taking into account the undealt cards; with this advantage the player must bet according to a predefined ramp.

2) If the dealer has an ace as the upcard, it has to compute the number of tens and not tens so as to tell the player if he has to buy insurance or not.

3) Before each player decision, it has to evaluate the expected value of hitting, standing, etc. according to the undealt cards in order to recommend the appropiate move.

Can anyone tell me what will be the aproximate advantage the player will have? Let's assume any combination of decks and rules, what I want to know is the gap between this machine and any of the most used strategies.

Thanks in advance,

Mike LC

My program called Composition Dependent Combinatorial Analyzer or cdca for short does what you are asking. What can it do?

1. It can compute a pre-deal EV for any shoe composition assuming perfect play for non-split hands. Split hands are computed using a highly accurate but effcient algorithm. Computing perfect play EVs for splits pre-deal is possible but the only way I know of would be by a brute force algorithm requiring an exhaustive amount of computer time. My program could play splits perfectly as they are dealt but it couldn't calculate an exact EV for perfect play beforehand, though, and settles for an accurate estimate of perfect play to quickly compute pre-deal split EVs. Non-pair hands are always computed using perfect play.

2. Pre-deal insurance EV using optimal insurance decisions could be computed. cdca doesn't presently do this but my other program, tdca (Total Dependent Combinatorial Analyzer) does.

3. It could play any hand perfectly, including splits, as they are dealt. Insurance is simple for any given hand. If dealer has an ace up and the shoe composition consists of more than 1/3 tens, take insurance.


What is the overall EV using perfect play? I'd like to know this myself.

1. It will definitely depend on penetration.

2. A mathematical solution to the problem would require the listing of all possible shoe compositions and their probabilities for a given penetration. There are far too many possible compositions for this to be practical.

3. A traditional sim would not be effective because it would not have the definition of perfect play for any given composition it encountered.

4. I am working on another version of cdca. If I combine simulation to deal and combinatorial analysis to play perfectly, that should give a reasonable answer.

Nono, I know what he means, and it would be really difficult.

He's looking for a system that knows *exactly* what cards are left in the deck, and *exactly* what plays to make. Such a system would have 100% betting correlation and 100% playing efficiency. It would also be impossible to use by a human, but I'm not sure that's what he's looking for anyway.

To answer the question: No, I don't think that's possible. I don't know much about combinatorial mathematics myself, but from what I do know, today's hardware isn't fast enough to give you decisions in a reasonable amount of time.

cdca works pretty quickly. It works best with a fast processor and the more RAM the better. I am dissatisfied with it because it saves each shoe state it calculates. It does this to save calculation time. The problem is that if the program is used long enough without a reset there are so many possible shoe states in blackjack that eventually all RAM will be used and virtual memory using the hard drive will be needed and slow the program a lot. The problem may never be noticed on newer computers with lots of RAM but on my old computer this happens. The new version of cdca doesn't have this problem and neither does tdca.

rrwwods: That's exactly what I'm asking.
johndoe: Do you think this perfect machine would only be marginally better than a good counting system? Regarding betting correlation it's clear that it's a marginal increment, but what about playing efficiency? And insurance correlation? Wouldn't it be a big gain?

By the way, I use hi-lo with I18.

Thanks,

Mike

Playing perfectly has nothing to do with counting. Perfect play just considers whatever composition exists at the time a playing decision needs to be made.

For Itakeyourmoney:

It's just theorical curiosity. I'm planning to develop a software for this "Perfect BJ machine" and I wonder about the final result.

I do not play online blackjack and bringing a device like this into a casino it's not of my interest, not to say it's ilegal (or not?).

I'd appreciate your opinions.

Mike

I hope you're prepared to take this up as a time consuming hobby. There's not much demand for this type of information.

 

[mleancole]

What is the overall EV using perfect play? I'd like to know this myself.

1. It will definitely depend on penetration.

2. A mathematical solution to the problem would require the listing of all possible shoe compositions and their probabilities for a given penetration. There are far too many possible compositions for this to be practical.

3. A traditional sim would not be effective because it would not have the definition of perfect play for any given composition it encountered.

4. I am working on another version of cdca. If I combine simulation to deal and combinatorial analysis to play perfectly, that should give a reasonable answer.

My idea is to run a sim that plays the "perfect play" for a defined set of rules and penetration and compare it to a predefined play as sims do normally, that would give me the difference between the two plays.

Playing perfectly has nothing to do with counting. Perfect play just considers whatever composition exists at the time a playing decision needs to be made.

I know it, that's what I'm looking for.

Do you think that the latest hardware can cope with taking playing decisions in a very brief time, to allow "online playing" (nothing to do with online blackjack)?

Mike

 

[k_c]

My idea is to run a sim that plays the "perfect play" for a defined set of rules and penetration and compare it to a predefined play as sims do normally, that would give me the difference between the two plays.

Yes perfect play versus basic strategy, index play, or whatever. My cdca program offers the option of computing using full shoe composition dependent basic strategy or best strategy. It's not a simultaneous calculation, though, but you can compute the same composition twice using either the basic strategy or best strategy option. That's as far as I've gotten with comp dependent strategy so far. Computing using user defined indices would be relatively simple but I'd want to simultaneously compute basic strategy so plays with no index revert to basic strategy.

I know it, that's what I'm looking for.

Do you think that the latest hardware can cope with taking playing decisions in a very brief time, to allow "online playing" (nothing to do with online blackjack)?

Mike

I think you're asking if computations can be done in somewhat real time. The answer is yes, but it's not that easy. Links to downloads of both of my programs are in my signature. They are free to try so you can see for yourself.