MJ: Any Response Don?

[MJ]

I posted a question on 8/9 in reference to your BJA3 ODP study. Not sure if you read it. In the event you did read it but are not interested, that is fine.

There is one additional question. Why is the ODP while observing a shoe the same as the ODP while playing? Is the ODP ever different for the former and latter case?

Table 13.1 gives a WR of $105.13 for WiWo and $104.40 for Mr. Perfect. Shouldn't Perfect's WR be ever so slightly higher than WiWo?

MJ

[Don Schlesinger]

> I posted a question on 8/9 in reference to your BJA3
> ODP study. Not sure if you read it. In the event you
> did read it but are not interested, that is fine.

I read it somewhere, but it didn't seem like I had to answer. Refresh my memory please.

> There is one additional question. Why is the ODP while
> observing a shoe the same as the ODP while playing? Is
> the ODP ever different for the former and latter case?

I think I wrote about that somewhere -- that they were, technically, slightly different, but not enough to really matter. Give me a minute and I'll find the reference.

> Table 13.1 gives a WR of $105.13 for WiWo and $104.40
> for Mr. Perfect. Shouldn't Perfect's WR be ever so
> slightly higher than WiWo?

Not sure about that. Don't remember if it has been addressed before. Could just be standard error and rounding, or there could be somethng else going on. I'd have to reread.

It's actually been a long time since I've read the book! :-) I used to reread Griffin almost every year. I think it's time to bite the bullet and try to get through BJA3 ... again. :-)

Don

[MJ]

> I read it somewhere, but it didn't seem like I had to
> answer. Refresh my memory please.

My question was regarding how JA used BJRM to calculate the optimal bets for the three different wonging styles. BJRM is a post sim calculator. It cannot run simulations although you can load simulations from SBA into it. Did John load sims from SBA or take sim data from another source and input it into BJRM to derive the optimal bets for the study?

To put things into perspective, if I wanted to run an ODP simulation on CVData, I could use an "optimal" bet schedule taken from CVCX. But the problem here is that the bet schedule would not truly be optimal because CVCX can only give an estimate of SCORE when the DA is set. Thus, I am inserting suboptimal bets into CVData. Consequently, this could result in ODP values that are suboptimal.

MJ

[Don Schlesinger]

> My question was regarding how JA used BJRM to
> calculate the optimal bets for the three different
> wonging styles. BJRM is a post sim calculator. It
> cannot run simulations although you can load
> simulations from SBA into it. Did John load sims from
> SBA

Yes.

>or take sim data from another source and input it
> into BJRM to derive the optimal bets for the study?

Ultimately, as you may have read, it was James Grosjean's program that we used for the final version of the study.

> To put things into perspective, if I wanted to run an
> ODP simulation on CVData,

You can't. Don't even go there. :-)

> I could use an
> "optimal" bet schedule taken from CVCX. But
> the problem here is that the bet schedule would not
> truly be optimal because CVCX can only give an
> estimate of SCORE when the DA is set. Thus, I am
> inserting suboptimal bets into CVData. Consequently,
> this could result in ODP values that are suboptimal.

There is no program anywhere that can run what we did in BJA3. You should realize that from reading the chapter. What was done is extremely complicated, and you might get some rough estimate from a simulator, but it isn't going to be doing what we did.

Don

[Norm Wattenberger]

I will add lag time calculations and observation departure points by depth to CVData. This will be defined the same as CVData currently handles wong-in and out points by depth. These are simple changes and will allow you to run your own sims with different WI, WO, departure points by shoe depth. That is, you can find ODPs by trial and error sims.

The question of automatically running many sims to determine the set of ODPs by depth is interesting. This kind of problem generally requires a specifically designed set of code as opposed to a general purpose simulator. However, I could probably modify the current Multi-Sim feature to handle this. Question is - is it worth the trouble? In the end you will arrive at an answer that is only correct for one set of circumstances. One number of players with specific lag times. The problem is that the number of players at the table substantially alters the TC frequency effect of entering and leaving a table.


CVCX Online

[MJ]

> I will add lag time calculations and observation
> departure points by depth to CVData. This will be
> defined the same as CVData currently handles wong-in
> and out points by depth. These are simple changes and
> will allow you to run your own sims with different WI,
> WO, departure points by shoe depth. That is, you can
> find ODPs by trial and error sims.

Also consider increasing the number of penetration levels the user can input. Right now it is 6, but 10 would be nice for a 6D game. For 8D 13 penetration levels would be about right. This way the ODPs can be determined for 1/2 deck penetration levels. The way it is currently configured is more or less for full deck increments.

> The question of automatically running many sims to
> determine the set of ODPs by depth is interesting.
> This kind of problem generally requires a specifically
> designed set of code as opposed to a general purpose
> simulator. However, I could probably modify the
> current Multi-Sim feature to handle this.

IMHO, this is the way to go. Can you imagine trying to use CVCX to determine index values via trial and error? Surely it can be done but it is an arduous, time consuming process .

> Question is
> - is it worth the trouble? In the end you will arrive
> at an answer that is only correct for one set of
> circumstances. One number of players with specific lag
> times.

That is true. But the same holds true if you were to figure out the ODPs by trial and error. Either way the ODPs are correct for only one particular set of circumstances.

> The problem is that the number of players at
> the table substantially alters the TC frequency effect
> of entering and leaving a table.

I knew that # of players affects TC freq, but I was not aware that this can substantially affect the entry and exit points.

The only logical course of action is for the player to settle upon some type of compromise ODP values based on one number for players and avg lag time; unless you endorse the notion of memorizing different ODP values based upon different number of players and different lag times. Absurd! :-)

MJ

[Norm Wattenberger]

> Also consider increasing the number of penetration
> levels the user can input. Right now it is 6, but 10
> would be nice for a 6D game. For 8D 13 penetration
> levels would be about right. This way the ODPs can be
> determined for 1/2 deck penetration levels. The way it
> is currently configured is more or less for full deck
> increments.

Hard to believe that half-decks makes a difference. Particularly when you take into account that you are estimating the lag times. Generally speaking, if one part of a calculation has a certain precision, it isn't worth the effort to use a greater precision for other parts of the calc. Kind of like hooking a $20,000 speaker up to a $300 amplifier.

> IMHO, this is the way to go. Can you imagine trying to
> use CVCX to determine index values via trial and
> error? Surely it can be done but it is an arduous,
> time consuming process .

I used CVData in this manner to find the REKO index set.

> I knew that # of players affects TC freq, but I was
> not aware that this can substantially affect the entry
> and exit points.

Remember that pulling Tens out of the shoe affects TC frequencies. TC frequencies are very important.

> The only logical course of action is for the player to
> settle upon some type of compromise ODP values based
> on one number for players and avg lag time; unless you
> endorse the notion of memorizing different ODP values
> based upon different number of players and different
> lag times. Absurd! :-)

Yes absurd. Though it would be useful to know how far off the win rate can be when the variables are different.

[MJ]

> Hard to believe that half-decks makes a difference.
> Particularly when you take into account that you are
> estimating the lag times. Generally speaking, if one
> part of a calculation has a certain precision, it
> isn't worth the effort to use a greater precision for
> other parts of the calc.

Well the ODP study conducted in BJA3 incorporated ODPs for half deck precision. Whether or not it really makes a difference in WR, I cannot say. It would be interesting to see a study comparing WR for full deck vs half deck precision.

Don, did anybody on Team ODP raise any objections to using half deck precision for the study? Any thoughts on how WR will be affected if full deck precision is used instead for the exit points? I realize exact card resolution was used for the TC conversion, but that is a separate matter.

The other important point that should be raised is that most card counters are trained to estimate the discard rack to the nearest 1/2 deck for the TC conversion. Why not use this same precision when you determine the ODPs? It certainly is not requiring any extra mental effort on the counter's part.

It certainly cannot hurt to give the user the option to specify more levels of penetration with CVData.

> I used CVData in this manner to find the REKO index
> set.

But what is the benefit of using a trial and error process if you can code the software to determine the ODPs with a single simulation?

> Yes absurd. Though it would be useful to know how far
> off the win rate can be when the variables are
> different.

I agree with you there. In so far as altering the lag time, we know how it affects the WR and ODPs. Just look at the tables 13.1 and 13.3 in BJA3. When a lag time of even 6 rounds is imposed, the WR drops from $105.13 to $59.95 for WiWo. That is about a $7.50 loss per round of lag.

I am curious how the ODPs and WRs would be affected if the number of players were switched from 4 to say, 2. Any thoughts?

MJ

[Norm Wattenberger]

First, when I think of the problem, I think of WI, WO and departure points by depth. So there are numerous combinations of possibilities.

As I understand it, the Chapter 13 calculations assumed that a player left whenever you entered. This means that TC frequencies are not altered. This means that you can run one sim and then make calculations from that one sim for any set of departure points. Yes in such a circumstance the calculations can be optimized to find the correct set of ODPs by penetration. That is, since it takes a fraction of a second to calculate the effect of any one set of ODPs; the software itself can iterate numerous guesses until it closes in on the right answer.

Now suppose we want to take into account the effects of entering a table. Now we have to run separate sims for most every set of possible ODP's by penetration. Now it no longer takes a fraction of a second per iteration. Instead you run a billion or so hands per iteration. (Actually many billions if you want ODP?s in fractions TCs.)

Now if we wish to look at different wong-in points by penetration and different wong-out points by penetration and different ODP points by penetration; we are talking about thousands of billon+ round sims.

But I'm getting carried away. If we look at the Chapter 13 work already performed; we find that the ODP's by depth really don't vary that much until the end of the shoe. Practically speaking, are you really going to use a departure point of -.1 at .5 depth and -.3 at 3.0 depth? If you are looking to find the best set of practical ODPs, WIs and WOs instead of a theoretical comparison; I think the best way forward is to make the mods to CVData to allow ODPs by depth and definition of various lag times and to use these to manually run sims to come to a set of values with which you are comfortable. And I don?t think it will take that many trials to find a practical set of numbers.

There is also an advantage to this method. You can keep the trials in the archives. Then later, you can look at the trials with a different set of lag times. Lag times could be changed after a sim.

[Don Schlesinger]

> First, when I think of the problem, I think of WI, WO
> and departure points by depth. So there are numerous
> combinations of possibilities.

> As I understand it, the Chapter 13 calculations
> assumed that a player left whenever you entered. This
> means that TC frequencies are not altered. This means
> that you can run one sim and then make calculations
> from that one sim for any set of departure points. Yes
> in such a circumstance the calculations can be
> optimized to find the correct set of ODPs by
> penetration. That is, since it takes a fraction of a
> second to calculate the effect of any one set of ODPs;
> the software itself can iterate numerous guesses until
> it closes in on the right answer.

> Now suppose we want to take into account the effects
> of entering a table. Now we have to run separate sims
> for most every set of possible ODP's by penetration.
> Now it no longer takes a fraction of a second per
> iteration. Instead you run a billion or so hands per
> iteration. (Actually many billions if you want ODP?s
> in fractions TCs.)

> Now if we wish to look at different wong-in points by
> penetration and different wong-out points by
> penetration and different ODP points by penetration;
> we are talking about thousands of billon+ round sims.

That's why we kept the number of players at the table the same! :-)

> But I'm getting carried away. If we look at the
> Chapter 13 work already performed, we find that the
> ODP's by depth really don't vary that much until the
> end of the shoe.

Right. Fine-tuning will never change by a full TC point.

> Practically speaking, are you really
> going to use a departure point of -.1 at .5 depth and
> -.3 at 3.0 depth?

See above. :-)

>If you are looking to find the best
> set of practical ODPs, WIs and WOs instead of a
> theoretical comparison, I think the best way forward
> is to make the mods to CVData to allow ODPs by depth
> and definition of various lag times and to use these
> to manually run sims to come to a set of values with
> which you are comfortable. And I don?t think it will
> take that many trials to find a practical set of
> numbers.

They're all going to look strikingly similar.

Don

[MJ]

Thanks for the in depth explanation. Allow me to clarify what I am trying to figure out. Note: I am not sure if I should proceed in the order mentioned.

*** I need a one size fits all Wi point for KO-P (not dependent on depth) for the 6D game.

*** Then I need to find the Wo point for different depths (increments of full or half deck) up to 4.5 decks that take into account lag time.

*** Then I need to establish the departure point when observing the table which take into account lag time.

OK, that was a pretty clear and concise explanation.

After I determine the Wi point, I can use that number when I run the sims to determine the Wo points. Once I determine the Wo point for the first deck, I use that number when I run the sims to determine the Wo point for the second deck; and so on and so forth.

Of course, Force Shuffle On Exit will be in affect and an estimate of lag time will be imposed.

Once you add the feature for departing while observing a shoe, how do you suggest incorporating it into my algorithm?

> There is also an advantage to this method. You can
> keep the trials in the archives. Then later, you can
> look at the trials with a different set of lag times.
> Lag times could be changed after a sim.

Good to hear lag time can be altered post sim.

MJ

[MJ]

Do you think using ODPs optimized for half deck increments is overkill? Can I get a similar WR using full deck increments? The system I am using is KO.

I realize in your study Team ODP tried to be as precise as possible so that no errors were made due to rounding and such. Ultimately, I am not certain if the study used half deck increments because it produces a better WR or just to be as precise as possible.

Other question: For the study 4 players were assumed to be playing. If there were only 2 players at the table, would the counter have to be more or less patient before departing?

Thanks,
MJ

[Don Schlesinger]

> Do you think using ODPs optimized for half deck
> increments is overkill?

Yes, surely.

Can I get a similar WR using
> full deck increments? The system I am using is KO.

I have no doubt. Just look at the graphs, at the end, according to pen level. How much do they really change, as you move through the pack?

> I realize in your study Team ODP tried to be as
> precise as possible so that no errors were made due to
> rounding and such. Ultimately, I am not certain if the
> study used half deck increments because it produces a
> better WR or just to be as precise as possible.

Probably the latter and because, if you do it that way, no one can ask what would have happened if you had tried to be more precise. :-)

> Other question: For the study 4 players were assumed
> to be playing. If there were only 2 players at the
> table, would the counter have to be more or less
> patient before departing?

I would assume that the better the game, the more inclined you would be to stay. Just compare 4.5/6 numbers to the 5/6 ones.

Don