An interesting practice exersize

ohbehave

Well-Known Member
#1
This may be a helpful exercize for your practice sessions, especially for those of us who are new to counting.

Take a sheet of lined or graph paper and down the left hand side number the page starting from the highest true count you normally get for your preferred counting system down the page to the lowest count you usually see. I don't use HiLo but for HiLo it might be +8 down through -5 or so.

Now, deal practice rounds and record each round as a Win(W), Loss(L), or Push(P) next to the corresponding TC for that round.

After a while you'll start to see the distribution of hands dealt at each TC.

This will give you a visual feel for how often each count occurs as well as a visual for the difference in wins and losses at positive, negative, and neutral counts.

Note, this is only an exersize. Its not meant for any specific statistical purposes.
 

Renzey

Well-Known Member
#2
ohbehave said:
This may be a helpful exersize for those of us who are new to counting.
Take a sheet of lined or graph paper and down the left hand side number the page starting from the highest true count you normally get for your preferred counting system down the page to the lowest count you usually see.
Now, deal practice rounds and record each round as a Win(W), Loss(L), or Push(P) next to the corresponding TC for that round.
After a while you'll start to see the distribution of hands dealt at each TC.
This will give you a visual feel for the difference in wins and losses at positive, negative, and neutral counts.
After reading Thorp's "Beat the Dealer" in 1975, I felt the need to perform several experiments to verify that blackjack strategy really worked the way the book said. One ongoing experiment was to manually deal 100,000 practice hands, registering total number of hands won/lost vs. the total number of units won/lost. At the end, I had lost slightly more hands than I won, but had won many more units than I lost (while betting multiple units in positive TC's). This satisfied me that high cards do indeed favor the player, and that a fair, random game can be beaten.
 

Katweezel

Well-Known Member
#3
Sims

Renzey said:
After reading Thorp's "Beat the Dealer" in 1975, I felt the need to perform several experiments to verify that blackjack strategy really worked the way the book said. One ongoing experiment was to manually deal 100,000 practice hands, registering total number of hands won/lost vs. the total number of units won/lost. At the end, I had lost slightly more hands than I won, but had won many more units than I lost (while betting multiple units in positive TC's). This satisfied me that high cards do indeed favor the player, and that a fair, random game can be beaten.
Renzey, you are a pearl! In one brief paragraph, you managed to distill the essence - the very core - of Blackjack Advantage Play. And you managed to do that in 1975 without a computer or a sim of one billion. There is a school of thought that says someone who sits and plays 100,000 hands of BJ with... themselves, is, well... different. But you demonstrated what benefits persistence, determination, motivation and manual accurate record-keeping can provide. This is not to say there is anything wrong at all with sims. I think your paragraph simply makes it clear that perhaps sims are not the be-all and end-all of Blackjack analysis. :cat:
 

Renzey

Well-Known Member
#6
EasyRhino said:
How long does it take to deal yourself 100,000 hands?
Did it over the course of about 2 years -- an hour here and an hour there, while precticing counting. Some were single deck 1-to-3, some double deck 1-to-6 and some 4 deck 1-to-8.
 

Katweezel

Well-Known Member
#7
Clear

Renzey said:
Did it over the course of about 2 years -- an hour here and an hour there, while precticing counting. Some were single deck 1-to-3, some double deck 1-to-6 and some 4 deck 1-to-8.
Now that is a meticulous, detailed and precise answer. I bet your records were like that, as well... :cat:
 

QFIT

Well-Known Member
#9
With 100,000 hands, assuming they were all recorded correctly, the standard deviation for EV is about 1.2%. That is, the SD is in the neighborhood of the EV. This means that the EV is meaningless. Sorry, that's what stats say. And why I sim 2,000,000,000 hands.
 

EasyRhino

Well-Known Member
#10
Isn't N0 is when SD is roughly = EV?

Fred, I think you should go back to hand dealing until you get at least two standard deviations. :)
 

Katweezel

Well-Known Member
#11
Science V ...mystery

Fred, your measly 100,000 from the dark ages does not look real flash against Qfit's 2,000,000,000 hands' sample, now does it, in black and white... However, your relative (small) sample contained something that a computer can never get: the mysterious personal human touch factor. Only those who have experienced what you did - all those recorded hands - understand this. What you got from doing that, by 1977, was your reward for your focused attention and diligence. That made it all worthwhile. Correct? :cat:
 

DonR

Well-Known Member
#12
QFIT said:
With 100,000 hands, assuming they were all recorded correctly, the standard deviation for EV is about 1.2%. That is, the SD is in the neighborhood of the EV. This means that the EV is meaningless. Sorry, that's what stats say. And why I sim 2,000,000,000 hands.
Some of us might have to find a way of how to live 10-15 thousand years to reach this long term goal. :)
 

Renzey

Well-Known Member
#13
QFIT said:
With 100,000 hands, assuming they were all recorded correctly, the standard deviation for EV is about 1.2%. That is, the SD is in the neighborhood of the EV. This means that the EV is meaningless. Sorry, that's what stats say. And why I sim 2,000,000,000 hands.
Norm, I certainly support the value of sample sizes large enough to virtually eliminate the "luck factor" and reveal the "real" odds. However, doing a quick SD calculation, the SD for 100,000 hands with a 1-to-6 spread seemed to be much lower than the 1.2% you stated.

So I cranked up my 12 year old BJ simulator and ran off several series of 100,000 hands -- first with 1 deck and a 1-to-3 spread -- then with 2 decks and a 1-to-6 spread -- and finally with 4 decks and a 1-to-8 spread. The standard errors in Gain for each were 0.40%, 0.47% and 0.53% respectively. Then I used BJRM2002 to tell me the SD for the 4 deck run and it indicated it to be around 0.65% for the EV. Can you clarify the 1.2% figure??
 

Sonny

Well-Known Member
#14
Katweezel said:
However, your relative (small) sample contained something that a computer can never get: the mysterious personal human touch factor.
Nonsense. All of the programs that I write include detailed code for the mysterious personal human touch factor. I have a function that calculates the precise value of this and returns a number between 0 and 1 which I call the Human Mystery Coefficient. I’ll even share my pseudo-code with you:

Code:
int CheckMystery (string previousShoe)
{
  int tableDemeanor, humanTouch, tableBias, tableTrend, tableScore;

  CheckAshTray (); //external function for checking the volume of ash

//choose a seat
  for (i=1; i<7; i++){
    if (Seat(i)==Cold) then << 1 
  }

  if (previousShoe==Bad) then nextShoe=Bad;

//calculate player parameters
  for (i=1; i<numPlayers; i++){
    tableDemeanor+=player(i).Attitude;
    if player(i).takesDealersBustCard || player(i).disturbsCardFlow then
       player(i).Skill=0
    else 
       player(i).Skill=777;
    totalSkill+=player(i).Skill; 
  }

  if (Dealer==Hot) ||(Table==Cold) || (Players==Unhappy) || (lostThreeHandsInARow==true) then tableScore= -tableScore;

  humanTouch=(totalSkill+tableDemeanor)/numPlayers;
  tableBias=clumping+dealerBustRate;
  tableTrend=(streakLength+numStreaks)*paintFollowsPaint;
  
//calculate the precise value
  mysteriousHumanTouch=humanTouch+tableBias+tableTrend+tableScore;

//return the value
  return rand();
}
-Sonny-
 

QFIT

Well-Known Member
#15
Renzey said:
Norm, I certainly support the value of sample sizes large enough to virtually eliminate the "luck factor" and reveal the "real" odds. However, doing a quick SD calculation, the SD for 100,000 hands with a 1-to-6 spread seemed to be much lower than the 1.2% you stated.

So I cranked up my 12 year old BJ simulator and ran off several series of 100,000 hands -- first with 1 deck and a 1-to-3 spread -- then with 2 decks and a 1-to-6 spread -- and finally with 4 decks and a 1-to-8 spread. The standard errors in Gain for each were 0.40%, 0.47% and 0.53% respectively. Then I used BJRM2002 to tell me the SD for the 4 deck run and it indicated it to be around 0.65% for the EV. Can you clarify the 1.2% figure??
Actually, CVData cannot sim that small a sample. At 20,000,000 hands a second, I can't stop it quickly enough.:) I didn't keep the sim, but I'm sure I had a higher spread. In any case, .65% is still a problem.
 

QFIT

Well-Known Member
#16
Katweezel said:
However, your relative (small) sample contained something that a computer can never get: the mysterious personal human touch factor. Only those who have experienced what you did - all those recorded hands - understand this.
Not sure what this means. CVData supports player errors, dealer errors, real casino shuffles with varying degrees of shuffle precision, casino heat, players coming and going and varying penetration shoe by shoe.
 

bj bob

Well-Known Member
#19
That's freaking hilarious!!

Sonny said:
Nonsense. All of the programs that I write include detailed code for the mysterious personal human touch factor. I have a function that calculates the precise value of this and returns a number between 0 and 1 which I call the Human Mystery Coefficient. I’ll even share my pseudo-code with you:

Code:
int CheckMystery (string previousShoe)
{
  int tableDemeanor, humanTouch, tableBias, tableTrend, tableScore;

  CheckAshTray (); //external function for checking the volume of ash

//choose a seat
  for (i=1; i<7; i++){
    if (Seat(i)==Cold) then << 1 
  }

  if (previousShoe==Bad) then nextShoe=Bad;

//calculate player parameters
  for (i=1; i<numPlayers; i++){
    tableDemeanor+=player(i).Attitude;
    if player(i).takesDealersBustCard || player(i).disturbsCardFlow then
       player(i).Skill=0
    else 
       player(i).Skill=777;
    totalSkill+=player(i).Skill; 
  }

  if (Dealer==Hot) ||(Table==Cold) || (Players==Unhappy) || (lostThreeHandsInARow==true) then tableScore= -tableScore;

  humanTouch=(totalSkill+tableDemeanor)/numPlayers;
  tableBias=clumping+dealerBustRate;
  tableTrend=(streakLength+numStreaks)*paintFollowsPaint;
  
//calculate the precise value
  mysteriousHumanTouch=humanTouch+tableBias+tableTrend+tableScore;

//return the value
  return rand();
}
-Sonny-
Hey Sonny, is your code formula copyrighted? And even if it is, can you make it available in our bookstore in laminated form? I just can't wait to whip one out and proudly show it to my favorite PB's when I color in at the table. Guaranteed they won't bother me for the rest of the evening. :laugh::laugh:
 

Sonny

Well-Known Member
#20
bj bob said:
That's freaking hilarious!!
Thanks. It's a terribly slow and inefficient way to generate a random number, but it’s 100% accurate. I’m not sure if anyone caught the “shift left” gag though.

bj bob said:
Hey Sonny, is your code formula copyrighted?
No it's open source, although it may infringe on some of Ion Saliu's formulas. :rolleyes:

-Sonny-
 
Top