Sims VS Real World Results

[blackjack avenger]

I think I have noticed a trend among posters. Many seem to look at their stats vs sims.

How much do you make an hour?

If you answer this question by looking at your hours played and $ amount won then this is probably not quite right.

A better idea of what you make an hour can be gotten from sims. The reason being if one has not played many many hours then their results are just variance, which can be misleading.

Now, we don't play perfectly like a sim and casino conditions can affect our results but with a sim we eliminate variance, which is probably the wildest variable. We can also sim what the actual conditions are we did face and know what our EV was for a session. This would be more accurate then actual results which are just variance.

If one were to play a game they have never faced before then looking at their past results would have no meaning.

Real results are variance and is just smoke and mirrors
Sims allow one to clear the air and see the truth

:joker::whip:
good cards

 

[Southpaw]

Agreed, you cannot base what you are expected to win on your past, real-life results.

If you remember this thread that you participated in, I showed that after 3 million rounds (which represents a very active playing career), it is still highly uncertain as to how many units per hour that one would have averaged:

http://www.blackjackinfo.com/bb/showthread.php?t=20523&highlight=expected+fluctuation+lifetime+tables

Thus, one cannot approximate their expected win based on their previous results because there will, in all but the rarest cases, be little statistical significance associated with one's real-life results.

Spaw

 

[Canceler]

There’s at least one fairly frequent poster I’d like to ask if he knows the difference between “playing blackjack on the computer” and “running a sim”.

 

[blackjack avenger]

Delusional Real World

Conclusions some have made based on their results that don't seem quite right.

I used a higher level count for a few months, I did not win more so I stopped using it.

I once/have made X amount in Y amount of time so I am a great player.

I had X results so I was cheated.

My results are better using a simpler count.

I had X results so the game was great/bad.

Counting does not work or is weak because I had X results.

Sims are probably a better way to deal with the above issues.

 

[ohbehave]

There’s at least one fairly frequent poster I’d like to ask if he knows the difference between “playing blackjack on the computer” and “running a sim”.

If you are talking about me the answer is a resounding YES!. I thought I made it clear I'm just doing that for fun while keeping up some skills. Those few hands don't mean a thing, I know that.

I am not a voodoo BJ player by any means. I play fairly aggressive AP blackjack with a dose of cover to help dissuade heat.

 

[Smoke8]

I agree with what your saying about the true comparisons of Real World vs Sims, but for me it gets a bit confusing when evaluating my play against the sim results that I am expecting to see. When thinking " Long Run" are we to consider it as being 1,000,000 hands played? This is where I get confused! I play approximately 16 hrs each week end and when I do the math it works out to about 83,200 hands played per year, it will take over 10 years to reach the long run equivelant at the current rate of play. I win an average of 44% of the hands played and this generally equates a dollar amount greater than where I started. I attribute this to my using an 1 to 8 spread, DD game, and playing very effiecently. I know I won't get rich because of my reluctance to wager higher stakes, ( at least for now ). I have seen the ugly down slides of variance and when it's really bad I've learned to walk away. Other times I bet min and ride it out until it turns, when it does, I play until the conditions change again. I'm not certain this is the right approach, but, I'm having fun and still learning. It might be the wrong question but can someone tell me their interputation of the "Long Run" ?

 

[ohbehave]

And I do run my own sims quite frequently. I'm very anal about knowing exactly how my game should play. The practice sessions help with the "feel" of the game, ie bet spreads, cover, count strategy, etc

 

[Canceler]

ohbehave: You're not the person I had in mind.

Smoke8: Here's your chance to practice deleting duplicate posts! (Edit: Never mind, it's gone now.)

 

[ohbehave]

Ok thanks.

 

[Sonny]

It might be the wrong question but can someone tell me their interputation of the "Long Run" ?

From the Frequently Asked Questions thread:

Q: What is the “long run” that I keep hearing people talk about?
A: The long run is the point where you overcome the variance (luck) of the game and can be assured of reaching your EV. It is the point where your results are no longer affected by luck and only your skill remains. Here are some threads about the long run:
http://www.blackjackinfo.com/bb/showthread.php?t=4891
http://www.blackjackinfo.com/bb/showthread.php?t=5913

-Sonny-

 

[sagefr0g]

There’s at least one fairly frequent poster I’d like to ask if he knows the difference between “playing blackjack on the computer” and “running a sim”.

probably mean me, maybe, Canceler?
i could see where you would wonder.
my answer would be sure i see the difference, like say one is 'playing on the computer' same way as a 'sim plays out hands', then that 'playing on the computer' could be viewed as if it was some random sub set of the overall 'sim hands played'. the simulation's results can be wrapped up into statistically meaningful information, since such a large number of hands and rounds are played. a simulation i suppose it can be said pretty much overcomes the 'shroud' of the 'mystery' of the law of large numbers but the same can't be said for the puny amount of play anyone is gonna rack up 'playing on the computer', sorta thing.
so what meaning, if any does 'playing on the computer' or for that matter ones results of live play have? far as i can see it must be judged in light of the knowledge gleaned from sims, and answers to questions such as how closely has such play strategy and betting matched the play strategy and betting strategy of the simulation.

 

[Canceler]

probably mean me, maybe, Canceler?

Nope, not you. Not even close. I know you know what a sim is!

I have to say it’s a little entertaining to see who thinks I might have been referring to them. And I’m not sure what I’m going to do if the person I was referring to shows up.

so what meaning, if any does 'playing on the computer' or for that matter ones results of live play have? far as i can see it must be judged in light of the knowledge gleaned from sims, and answers to questions such as how closely has such play strategy and betting matched the play strategy and betting strategy of the simulation.

Exactly. And speaking of such things, whatever happened to Kasi?

 

[sagefr0g]

Nope, not you. Not even close. I know you know what a sim is!
I have to say it’s a little entertaining to see who thinks I might have been referring to them. And I’m not sure what I’m going to do if the person I was referring to shows up.

lol, glad i'm not a betting man, i thought for sure it was me.

Exactly. And speaking of such things, whatever happened to Kasi?

Kasi? heh, heh, well, i dunno last i corresponded with him i believe he was taking on some work, but right you are he would have been right on the target of the subject matter in question.

 

[21gunsalute]

Agreed, you cannot base what you are expected to win on your past, real-life results.

If you remember this thread that you participated in, I showed that after 3 million rounds (which represents a very active playing career), it is still highly uncertain as to how many units per hour that one would have averaged:

http://www.blackjackinfo.com/bb/showthread.php?t=20523&highlight=expected+fluctuation+lifetime+tables

Thus, one cannot approximate their expected win based on their previous results because there will, in all but the rarest cases, be little statistical significance associated with one's real-life results.

Spaw

In that case I have to question the usefulness of sims.

 

[Southpaw]

In that case I have to question the usefulness of sims.

I think if anything that this shouldn't make one question the usefulness of sims. They still indicate the most likely result, assuming you play the strategy to a tee. Rather, I think one might have to question the desirability of a straight-counting game when looking at in the short-run (let's for instance call this less than NO rounds).

Spaw

 

[k_c]

There’s at least one fairly frequent poster I’d like to ask if he knows the difference between “playing blackjack on the computer” and “running a sim”.

I'm 99% half sure you're referring to me.

In the Combinatorial Analysis thread I was trying to suggest that the missing ingredient in using a CA to compute indices is how to represent shoe composition. I have a method of getting a representative composition that is relatively simple. However simming to include include data from many compositions/pen levels may be better. Assume somehow computed data from a simulated game has been entered into a database. Suppose you were interested in hard 16 v T. You could query the database something like Select * From PlayerHands Where hardTotal=16 And UpCard=10 And HiLoTC=someValue And Pen=* to get pertinent data for all pen levels or Select * From PlayerHands Where hardTotal=16 And UpCard=10 And HiLoRC=someValue And Pen=someValue to get data for a specific pen level. Actual query may be different depending on design of database and this is just to show the idea. With computed data part of the puzzle is already solved so less data would be required to converge on an answer at the expense of requiring more computer time to get the data. If data is included from a player playing a practice game according to whim instead of correct strategy then data could possibly be corrupted to some extent so that may not be the greatest idea.

 

[MangoJ]

I'm 99% half sure you're referring to me.

In the Combinatorial Analysis thread I was trying to suggest that the missing ingredient in using a CA to compute indices is how to represent shoe composition. I have a method of getting a representative composition that is relatively simple. However simming to include include data from many compositions/pen levels may be better.

Hm, I don't think creating a representational deck ensemble with fixed penetration and count is so much of a problem. My first try would be to shuffle a new shoe, discard to the desired penetration, and count the discard. Then save by count. This could also be incorporated right into a Fisher-Yates shuffle algorithm, which will iterate on all cards only once. Just iterate for the desired penetration (so effectively discarding the rest), and sum up the count of each random card drawn.
One might try to avoid sampling the same shoe at different penetrations, because then the data might get biased if your ensemble isn't large enough.

Any thoughts about this ?

 

[Canceler]

I'm 99% half sure you're referring to me.

No no no no! No way am I going to even imply that you don’t quite know what you’re doing. That you would think it was you is almost scary.

 

[k_c]

Hm, I don't think creating a representational deck ensemble with fixed penetration and count is so much of a problem. My first try would be to shuffle a new shoe, discard to the desired penetration, and count the discard. Then save by count. This could also be incorporated right into a Fisher-Yates shuffle algorithm, which will iterate on all cards only once. Just iterate for the desired penetration (so effectively discarding the rest), and sum up the count of each random card drawn.
One might try to avoid sampling the same shoe at different penetrations, because then the data might get biased if your ensemble isn't large enough.

Any thoughts about this ?

It would be possible to sim count/pen data but I have a program that can calculate this by considering all of the possible subsets for a given count with given running count and given pen and optionally any number of specifically removed ranks. I wouldn't say the data translates to a representative composition, though, but just to the probabilities of drawing any rank. A translation to a shoe composition would result in numbers of each rank present being non-integral numbers.

Also count based math can have some differences from the math in a normal CA. For example an ace is an ace, obviously, but in HiLo it is also part of the group (T,A). If an ace is specifically removed from the first half of a shoe then probability of drawing a T on the next card increases when probs are based on HiLo. However specifically removing an ace in the second half of a shoe results also in the decrease in probability of drawing a T on the next draw. Using a representative composition would ignore the math resulting from the grouping of cards.

Below is output for 6 decks HiLo count with arbitrary values input. Probabilities of each rank are in the array toward the bottom. In my view extending this program into a count based CA offers the best potential of count analysis.

Please input tags relative to what remains in shoe
No input defaults to tag = 0

Example: HiLo tags (1-10) {1,-1,-1,-1,-1,-1,0,0,0,1}
	rank: A tag: 1
	rank: 2 tag: -1
	rank: 3 tag: -1
	rank: 4 tag: -1
	rank: 5 tag: -1
	rank: 6 tag: -1
	rank: 7 tag: 
	rank: 8 tag: 
	rank: 9 tag: 
	rank: T tag: 1

Decks (no input defaults to 1 deck): 6

Dealing all subsets for input count ..... Please wait ..... 

Press any key to continue

Count tags {1,-1,-1,-1,-1,-1,0,0,0,1}
Decks: 6

Display data for cards remaining, RC, specific removals, input below:
	Cards remaining (defaults to previous input; 1/2 shoe if none): 225
	Natural initial running count: 0
	Running count (no input defaults to 0): 6
	Number of each rank specifically removed (no input defaults to 0):
		A: 
		2: 
		3: 
		4: 
		5: 
		6: 
		7: 
		8: 
		9: 
		T: 


Count tags {1,-1,-1,-1,-1,-1,0,0,0,1}
Decks: 6
Cards remaining: 225
Initial running count (full shoe): 0
Running count: 6
Specific removals
	A: 0
	2: 0
	3: 0
	4: 0
	5: 0
	6: 0
	7: 0
	8: 0
	9: 0
	T: 0

Number of subsets for 225 cards: 3796
Prob of running count 6 from 6 decks: 0.0395526

p[1] 0.0795957  p[2] 0.0742624  p[3] 0.0742624  p[4] 0.0742624  p[5] 0.0742624
p[6] 0.0742624  p[7] 0.0769032  p[8] 0.0769032  p[9] 0.0769032  p[10] 0.318383

Press x or X to exit program (it may take some time to close,)
any other key to enter more data for same count tags/decks:

 

[MangoJ]

So there are only ~4k combinatorial decks with with the specific penetration and a running count of 6. The number is probably rather low because the RC is rather high. Yes I agree, with this condition a full combinatorial analysis is very feasible. At least for those higher RCs.

I still try to get my head around your "super-compositional deck distribution". I mean I see the technique, you average the decks composition over all compositions with the specific penetration and count, so each card will be represented with a real-value number in the super-composition.

I would really be eager to learn how to calculate dealer or player EVs from such a super-compositon. I'm sure one could use the same full combinatorial deck analysis as in integer-valued compositions, and probabilities of drawing specific cards should be the same even for real-valued numbers. But I'm not quite sure how the super-compositional deck evolves after drawing a card.
Say, the super-compositional deck contains 1.4 Aces, so after drawing an Ace it should contain 0.4 Aces. But how one draws an Ace on a 0.4 Ace left - obviously the super-compositional deck cannot contain less Aces than zero, but on the other hand should always contain a inter-number of total cards.
(With normal compositional decks this problem does not arise, drawing an 0.0 Ace has zero probability, but drawing an 0.4 Ace has a non-zero probability).