vpFREE2 Forums

accuracy and practicality of some computer-generate strategies?

#1

Looking for some information on a certain VP strategy, I found one
such strategy which had reportedly been computer-generated (before
anyone asks, it was not FVP, though I've seen some output from FVP
that also had some similarly questionable bits). Am I smoking
something or are there "bugs" in this strategy?

Reading through the strategy, though, a few details caught my eye as
odd, and I'd like a second opinion on those. The game is JoB with a
custom pay table.

Here's one part of the strategy:
-3 to RF, 3 high cards, A high
-4 to inside SF
-QJTs, KQJs

I am surprised than AQJs is rated higher than KQJs. It seems to me
that KQJs can form 2 straights and AQJs can only form 1, and that
everything else is equal. I would expect KQJs to rate higher than
AQJs. The only thing I can think of is a fine case of penalty cards,
but the difference in expected values that the site mentions is really
high. I'm also surprised that QJTs and KQJs would rate exactly the
same.

Similar question about K being rated higher than Q, itself rated
higher than J. My insinct rates them in the opposite order (in this
case the actual ranking is irrelevant as in this strategy you never
keep only 1 face cards when dealt 2 or more).

Here are some other interesting parts:

-QJ
[...]
-QJ9s
-AKQJ
[...]
-9TJQ

and

-K
[...]
-KQ9s, KJ9s

Is there any point listing the later plays since the earlier ones
always apply and offer a better value?

Thanks,
JBQ

#2

Jean-Baptiste Queru wrote:

Looking for some information on a certain VP strategy, I found one
such strategy which had reportedly been computer-generated (before
anyone asks, it was not FVP, though I've seen some output from FVP
that also had some similarly questionable bits). Am I smoking
something or are there "bugs" in this strategy?

Reading through the strategy, though, a few details caught my eye as
odd, and I'd like a second opinion on those. The game is JoB with a
custom pay table.

Here's one part of the strategy:
-3 to RF, 3 high cards, A high
-4 to inside SF
-QJTs, KQJs

I am surprised than AQJs is rated higher than KQJs.

Come on, guy. You're intelligent enough that if you want a savvy
answer you've got to provide all the facts.

You don't say where you got the strategy from (for 2 bucks on a dark
street corner?) and you reference a "custom pay table" without detail.
Anything can fly. But you're right, in standard Jacks you can't do
any better on a 3 cd. royal hold than KQJ. Does this paytable feature
a bonus on trip Aces, or something like that?

At any rate, why are you hunting around for rogue strategies when
reputable and reliable ones are readily available.

I suppose there's some satisfaction in pursuing vp as a analytic
pursuit, but I find it a whole lot more interesting as a practical one
and there's every reason to believe that's largely the case as a whole
around here.

Any angle on the game is welcome. But the most productive ones are
those that interest most.

- Harry

#3

you've got to provide all the facts.

Sorry, yes, you're right, re-reading what I had written I realized
that there wasn't quite enough information. But I also didn't want to
name the source, which I normally consider to be reasonably reliable,
without seeking reliable independent advice first. I also didn't want
to make a fool of myself (again), but it looks like I failed on that
one.

The strategy is Mike "Wizard of Odds" Shackleford's for Multistrike
Level 1 9/6 JoB, supposedly as generated by VPSM by adding 6 to each
paying hand in 9/6 JoB (a fairly reasonable assumption as far as I can
tell, at least as a first approximation). No fancy trip-ace bonus or
anything of the kind, at least nothing I can identify that would skew
AQJs over KQJs. I can't tell at this point whether Mike made mistakes
feeding data into VPSM or copying its output, or whether VPSM is buggy
or was buggy in the version that Mike used and has been fixed since
then. Certainly I don't intend to buy a piece of software when the
only sample output I have from it doesn't seem reliable, though I
guess that those here who use VPSM and have time to waste could have a
look. http://wizardofodds.com/multistrike/appendix2.html

At any rate, why are you hunting around for rogue strategies when
reputable and reliable ones are readily available.

Point taken.

I suppose there's some satisfaction in pursuing vp as a analytic
pursuit, but I find it a whole lot more interesting as a practical one
and there's every reason to believe that's largely the case as a whole
around here.

Got it. Unfortunately for the poor souls who live far away from the
nearest casino, VP as a practical pursuit isn't that practical. I'll
skip the personal details. I the meantime and as a substitute the
mathematician in me enjoys the analytic aspects of studying and
practicing the game, while definitely lacking confidence. At least the
practicing part should serve me well the next time I get to go to a
casino (work, wife, health...)

[joke] I guess that I need a
vpFREE_I_Live_Far_Away_From_Casinos_And_I_Want_To_Scratch_My_VP_Itch_By_Following_Pointless_Pursuits
list :wink:

Thanks,
JBQ

#4

Jean-Baptiste Queru wrote:

The strategy is Mike "Wizard of Odds" Shackleford's for Multistrike
Level 1 9/6 JoB.

Michael Shackleford is, of course, a very reputable source. So the
task at hand (if you have concerns about his strategy) is to analyze
it with the tools available for this purpose.

You express concern over the strategy placing a higher value on AQJs
than KQJs. A couple of comments:

From a practical stand point, this odd ranking is of no consequence.
If you have a choice between both holds, then clearly you have at
least a 4 card RF hold available as well -- thus you wouldn't opt for
either.

I'm not defending the strategy if it were inaccurate in valuing the
holds ... just saying that the inaccuracy would prevent you from
playing the related hands properly.

So, a word about his source. As you note, he's derived the strategy
from VP Strategy Master (VPSM). VPSM (like any other strategy source)
groups similar hands together in it's rankings. Those hands often
don't have identical values, but from a practical sense can be
evaluated in the strategy at their average value and weighed
accordingly. Most every source has a slightly different methodology
for doing this. However, the end results are almost entirely in
agreement (and any differences are relatively insignificant).

In the case of these VPSM rankings, your original post notes the
strategy as follows:

Here's one part of the strategy:
-3 to RF, 3 high cards, A high
-4 to inside SF
-QJTs, KQJs

Note that this doesn't place the A-high 3 to RF above just KQJ, but
above the combined holds QJT & KQJ. In truth KQJ is a stronger hold,
QJT (because of the non-face card) is a weaker hold. (I've evaluated
the non-penalty cases of these holds with WinPoker.)

I can't say exactly why VPSM combines these two holds yet separates
out the A-high hold. It may be that it's a broad based technique that
proves of value when evaluating games where quad Aces earn a bonus.
In any case, the stratgy produced is accurate.

Certainly I don't intend to buy a piece of software when the
only sample output I have from it doesn't seem reliable, though I
guess that those here who use VPSM and have time to waste could have
a look. http://wizardofodds.com/multistrike/appendix2.html

"Time to waste" is an interesting way to put it in the context of this
discussion :wink:

If I read you right, you're not willing to purchase VPSM because the
strategy you glanced at wasn't intuitively accurate. However, you
haven't armed yourself with other tools by which to assess that
accuracy. Puts you in a catch-22 bind, no?

My defense of the strategy should carry no more weight than other
opinions expressed here in the archives (and elsewhere) that VPSM is a
strong tool. Thus, I don't see that you're any further toward a
resolution unless you simply pop the $30 and buy it so that you can
evaluate it for yourself thoroughly. (Alternatively, you can pick up
Frugal VP and score both a very thorough strategy generator/evaluator
and practice tutor for just $10 more.

Got it. Unfortunately for the poor souls who live far away from the
nearest casino, VP as a practical pursuit isn't that practical.

Ok ... admittedly I hadn't chilled over my nightly cup of Bosco before
drafting my reply to you :wink:

What I'll stress is that if you're going to post more fron a
theoretical aspect than a practical one (not saying that theory can't
be put into practice, mind you ;), you want to do your homework
thoroughly and get as many ducks lined up in a row as possible -- the
objective simply to ensure that you receive as strong and useful
response as possible.

- Harry

#5

I agree with you here. The oddity probably comes from the fact that
VPSM assumes that it can clump KQJs and QJTs together, which may be a
reasonable assumption in regular JoB (where such hands typically end
up between high-pair and 4-to-a-flush if I'm not mistaken), but seems
to fall apart when the pay table changes dramatically (increasing the
relative value of a high pair to unusual levels). As a software
architect, that's not quite what I call "solid" software.

I also agree that the difference is of no practical significance
except maybe for the most hard-core players (who probably won't be
playing Multi-strike 9/6 JoB in the first place, I guess).

However, you
haven't armed yourself with other tools by which to assess that
accuracy. Puts you in a catch-22 bind, no?

I trust my brain. I'm not sure if it's a good idea, though. And I
admit that I didn't do an exact quantitative analysis of that specific
situation (though it's definitely within the realm of what can be
achieved without a computer or a calculator).

Thus, I don't see that you're any further toward a resolution

Actually, I think I am. I believe that I have a reasonable explanation
for what I've been seeing, thanks to your help. I now have me a much
better idea of how VPSM generates its strategies (and I feel that I
don't like it).

What I'll stress is that if you're going to post more fron a
theoretical aspect than a practical one (not saying that theory can't
be put into practice, mind you ;), you want to do your homework
thoroughly and get as many ducks lined up in a row as possible -- the
objective simply to ensure that you receive as strong and useful
response as possible.

Yup. I started from a false assumption (mea culpa), which was that the
clustering of hands was done programmatically instead of being
partially hard-coded. I'm not claiming that doing it programmatically
is an easy task, far from it (if it was easy I'd have written it for
myself and I'd be done). It's probably an interesting task, though.

I was primarily hoping to get confirmation that my instinct was
correct (i.e. than KQJs is better than AQJs), and you gave me much
more than that, for which I am grateful. All that gives me much more
food for thought, which I will keep for later as I now need to get
some real work done.

Thanks a lot,
JBQ

#6

Jean-Baptiste Queru wrote:

The oddity probably comes from the fact that VPSM assumes that it
can clump KQJs and QJTs together, which may be a reasonable
assumption in regular JoB, but seems to fall apart when the pay
table changes dramatically (increasing the relative value of a high
pair to unusual levels). As a software architect, that's not quite
what I call "solid" software ...

I now have me a much better idea of how VPSM generates its
strategies (and I feel that I don't like it).

I find these statements extraordinarily curious.

We've established in one case that the strategy is an entirely
accurate guide to play. You haven't identified another case in which
it's inaccurate.

The author of the software is, again, a fully reliable source. If
anyone would have discredited the software, ample evaluation to date
would have surfaced it.

Every strategy developer will condense strategy via some means in
order for the strategy to be a useful tool. If accuracy is sacrificed
in the process, the strategy is flawed. That's not the case here so,
you not having found a stronger tool for the time being, I find it
difficult why you'd label this software deficient for your needs.
(And you really need to lay your hands on some strong evaluation tools
to effectively continue your explorations of the subject.)

- Harry

#7

Wow. I had a quick run at the numbers. And I reached a pretty solid
conclusion and got the answer to my original question (I quote: 'Am I
smoking something or are there "bugs" in this strategy?'). Assuming
that I didn't make any gross mistake in my calculations (which
wouldn't be surprising at this point) the answer is that I must have
been smoking something. The strategy is sound. Simplified in an
"interesting" way, arguably, but sound.

Apologies for the disturbance. I got confused by the ordering of the
strategy, and didn't realize it until it was too late.

You gals and guys rule. I don't belong.
JBQ

#8

Jean-Baptiste Queru wrote:

You gals and guys rule. I don't belong.

Trust me, those who are frequent contributors sometime revel in their
misperceptions (I count heavily here :wink:

The most valued contributors are those who raise novel ideas at the
risk of exposing ignorance ... provided they have the sensibility to
improve upon that ignorance after presented with the intelligence and
insight of others more informed (rather than persistently defending
those misconceptions).

All evidence suggests you embrace this philosophy ... no need to be
apologetic for your posts (even if you "take it on the chin"
occasionally :wink:

Just keep in mind, no one loves a know-it-all and the display of a
modest degree of ignorance in life can endear you to many :wink:

- Harry

#9

What Harry says is very true, Jean Baptiste. I have certainly
endeared myself to many on this very erudite list, for that very
reason.

Please do not feel that you are alone in having posted an incorrect
assumption (whether modest or immodest). Most of us have done this
at one time or another. As Harry says, both the posting and then
admitting that the posted premise is fallacious, is what makes us
loveable, if not laughable.

Best regards,

Babe

···

---------------------------------
- In vpFREE@yahoogroups.com, "Harry Porter" <harry.porter@v...> wrote

Just keep in mind, no one loves a know-it-all and the display of a
modest degree of ignorance in life can endear you to many :wink:
- Harry

---------------------------------
Jean-Baptiste Queru wrote:
You gals and guys rule. I don't belong.

#10

JBQ,
In your post, you mentioned a few items.
KQJs is better than AKQs. The listing at wizard of odds is the basic
output from VPSM. If the advanced output had been run, you would seen
a much more (and much more messy) detailed breakdown of hand holdings,
and you would have seen KQJs listed ahead of AKQs.
However, for this game, AKQs is better than QJTs as a major goal is to
make at least one winning payback to advance to the next level.
  VPSM should have done a better job at housekeeping when it shows QJ
better than AKQJ. The AKQJ should not have been listed since QJ trumps
it. But as long as you read top to bottom and hold first match, you
would not be making an error.
  In making VPSM more user friendly, some hand combinations are listed
together in the basic printout that do not conflict. Thus KQJs and
QJTs can be shown on the same line as both cannot occur in the same
hand at one time.
  In general a lone J is better than lone K absent penalty cards. If
VPSM listing shows reverse order, one will see KJ above as even better
hold so thus one would not make an error by holing a K over a J, but
would rather hold KJ.
Welcome to the fun world of VP math. Best. TomSki

--- In vpFREE@yahoogroups.com, Jean-Baptiste Queru <jbqueru@g...>
wrote:

···

Wow. I had a quick run at the numbers. And I reached a pretty solid
conclusion and got the answer to my original question (I quote: 'Am I
smoking something or are there "bugs" in this strategy?'). Assuming
that I didn't make any gross mistake in my calculations (which
wouldn't be surprising at this point) the answer is that I must have
been smoking something. The strategy is sound. Simplified in an
"interesting" way, arguably, but sound.

Apologies for the disturbance. I got confused by the ordering of the
strategy, and didn't realize it until it was too late.

You gals and guys rule. I don't belong.
JBQ

#11

Thank you Tom (and Harry and Babe) for your kind words.

Tom: I guess that I was confused a bit by the output of VPSM. Not
seeing it merge certain other kinds of hands, I hadn't realized that
it had gone pretty deep in the analysis of merging and sorting some
3RF against some 4SF (and therefore thought that it had analyzed KQJs
behind AQJs).

Certainly it's a hard problem, and I have to admit that simplifying a
strategy is not necessarily as simple as it seems (with the goals of
removing the edge cases at the cost of a small bit of EV, or of ending
up with a human-readable equivalent of a given strategy, or both).
Thinking about it too much last night literally gave me a bad
headache.

That being said, I'm still somewhat surprised about the ranking of K
ahead of Q and J, with EVs listed in that explicit order. My instinct
would have listed J first. Obviously this part of the discussion has
no practical consequences (since there's never any choice between
those lines of the strategy). For the sake of avoiding a headache I
won't actually try to analyze it right now.

Without a doubt VP math is full of surprises, and has the distinct
advantage of having a direct real-world application.

Regards,
JBQ

#12

Jean-Baptiste Queru wrote:

That being said, I'm still somewhat surprised about the ranking of K
ahead of Q and J, with EVs listed in that explicit order. My
instinct would have listed J first. Obviously this part of the
discussion has no practical consequences (since there's never any
choice between those lines of the strategy). For the sake of
avoiding a headache I won't actually try to analyze it right now.

I'm not so quick to dismiss this puzzling order. (Amd I grasp your
avoidance of a detailed analysis likely stems for the most part from a
desire to avoid nitpicking based upon previous feedback.)

In the case of the {AKQ,AKJ,AQJ} over {QJT,KQJ} subset ranking
previously covered, it's reasonable that with the inclusion of QJT in
the second subset it would have a lower EV. I don't question the EV's
listed for these subsets (although, as previously noted, I find the
second subset grouping a little odd -- though I accept an explanation
that it's a consequence of the intracacies of optimizing strategy).
And, in any case (as also discussed) the strategy is still optimal.

However, I have difficulting accepting that a holding a lone K would
hold a higher EV than a lone J, all possible hands considered. In
examining a limited number of hands with WinPoker, I find the only
time that a K ranks higher as a substitute for a J is when I load the
balance of the hand up with at least 3 penalties to the Jack. Clearly
the higher number of straight possibilities will almost always give a
hand a higher EV with a Jack vs. a King held, all other cards constant.

So I'm at a loss as to how vpsm's strategy for this paytable assigns a
higher EV to the K only hold in a general case, even if the EV
reflects the exclusion of those hands where a higher EV hold is
present. If I understand, this is the basis under which EV's are
calculated in Frugal VP.

As a sidenote, while I stick to use of vpsm to generate my tailored
strategies (with editing to further condense the rankings), Frugal VP
does yield strategy subsets in its rankings that are a bit more
"rationale". Mind you, both produce what I believe are optimal
strategies in almost all cases. (If I understand correctly, in a very
limited number of cases, Frugal VP is more accurate in it's analysis
but I'd deem any differences in those cases insignificant and rarely
affects optimal play.)

- Harry

#13

I wrote:

So I'm at a loss as to how vpsm's strategy for this paytable assigns a
higher EV to the K only hold in a general case, even if the EV
reflects the exclusion of those hands where a higher EV hold is
present. If I understand, this is the basis under which EV's are
calculated in Frugal VP.

I was speaking off the cuff here, without reference to earlier
discussions when FVP was first released. I believe, upon further
reflection, this is an inaccurate representation of FVP. It is the
case that FVP uniquely considers more than just general case EV's in
determining rankings - namely (iirc) in the comparison of two holds
where higher ranked holds aren't present.

I'll hold off on digging through the archives for a more accurate
explanation of the FVP differences that result in modest adjustments
that yield more accurate strategies in a limited number of cases (with
negligible ER impact).

- H.

#14

I'll hold off on digging through the archives for a more accurate
explanation of the FVP differences that result in modest adjustments
that yield more accurate strategies in a limited number of cases (with
negligible ER impact).

I generate a lot of strategy charts from FVP for situations that I find, either interesting base games, or progressive situations that might prove to be good plays either currently or in the future. Often the strategies generated by FVP can be simplified a little by moving some of the holds around by hand with no loss in EV, or a not noticeable loss, but making the chart more condensed and easier to read.

I have occasionally found, though, that when I condense the charts, I can occasionally come up with situations where the new chart I've generated has a slightly higher EV than the one that FVP wants to generate.

The transitive property of "better than" does not always pertain to strategy holds in specific situations, for example you can have 3 holds, A, B and C, where if A and B are present in a hand, you always hold A for higher EV, if B and C are present you always hold B for higher EV, and therefore you'd assume that if A and C were present you'd always hold A for higher EV, but there can be cases where C is better than A. In this case, it's not possible to generate a simple linear chart ranking that will always tell you the best hold for your situation.

A chart that takes into account penalty card considerations can resolve some of these discrepancies, but FVP does not take into account penalty cards. In a non-transitive A,B,C holding case, FVP will have to pick some ordering of the holds, but it may not end up trying all the permutations of the hold rankings to determine which strategy chart (given the constraint of not specifying penalty card situations) will actually generate the highest EV.

Not using VPSM, I cannot comment on its algorithms or accuracies. I am sure that there are situations that may come up as games are analyzed that may show up quirks in its algorithms for determining best strategies too. Given that your computer doesn't have infinite computing power, clever algorithms have to be developed to take shortcuts when possible when analyzing the games, otherwise a truly exhaustive analysis of the games would take too long to be useful. These shortcuts can still provide accurate results given proper constraints about the games they are analyzing, but sometimes a game comes up (like TDB) which is outside the assumptions that were made, and adjustments need to be made.

I've seen games like "Add up Four of a Kind" (not sure what the real name of the game is, I believe it's a Sigma game) where any hand with 3 of a kind, 2 through 10, and two other cards that add up to the rank of the 3 of a kind qualifies as a 4 of a kind of that rank (Aces counting as 1 for adding), "In Between Four of a Kind" (again not sure of the real name of the game), or other odd games with weird possible pays. Games of these sorts may violate some of the simplifying assumptions made in the algorithms of the VP analysis software if attempts were made to handle these types of games.

···

_________________________________________________________________
FREE pop-up blocking with the new MSN Toolbar � get it now! http://toolbar.msn.click-url.com/go/onm00200415ave/direct/01/

#15

Given that your computer doesn't have infinite computing power, clever
algorithms have to be developed to take shortcuts when possible when
analyzing the games, otherwise a truly exhaustive analysis of the games
would take too long to be useful.

That's only true if you take "truly exhaustive" to mean "brute force". My
own algorithm is truly exhaustive in that it covers every possible case,
and uses no approximations, yet it can perform a complete and exact
analysis of JoB games in less than one second of CPU time on ancient
hardware (600MHz AMD Duron).

The real trick is to properly partition the problem into a relatively small
number of cases which are mathematically equivalent. For a 9/6 JoB
game, the 2.6 million distinct starting hands produces only 1292 different
cases. Evalutating those 1292 cases to find the best draw and summing
the results gives the 99.5439044% ER for the game.

I've seen games like "Add up Four of a Kind" (not sure what the real name
of the game is, I believe it's a Sigma game) where any hand with 3 of a
kind, 2 through 10, and two other cards that add up to the rank of the 3 of
a kind qualifies as a 4 of a kind of that rank (Aces counting as 1 for
adding), "In Between Four of a Kind" (again not sure of the real name of
the game), or other odd games with weird possible pays. Games of these
sorts may violate some of the simplifying assumptions made in the
algorithms of the VP analysis software if attempts were made to handle
these types of games.

Right. Game manufacturers will always devise new ways to play that
couldn't possibly be anticipated by software developers.

···

On Thursday 30 June 2005 02:11 pm, Part Timer wrote:

#16

(And I grasp your
avoidance of a detailed analysis likely stems for the most part from a
desire to avoid nitpicking based upon previous feedback.)

Well, let's say that I'd probably have taken any results privately
before discussing them on the list. But the #1 reason is that
analysing KQJs vs KQJ9s involved so few possibilities that I could
very easily do an exhaustive analysis "by hand". Analyzing J against K
is another story (there are many tens of thousands of possible hands
after keeping a lone J or a lone K).

No matter how I try to think about it, I can't find any good reason.
As far as I've been able to analyze, in any hand that has a J (and not
the K of the same suit), if the hand with the J is not played as a
single J, the same hand with the J replaced by a K of the same suit is
also not played as a single K; the same applies in the other
direction, if you try to replace a K by a J, therefore the set of
hands that are played as a single K or a single J are exactly the same
(same number of hands). I might have missed some hands, so don't take
what I just wrote for granted. The only kinds of results where the
difference matters is for straights and straight flushes, and in
either case there are more possibilities with a J than with a K.

I suspect that either my reasoning is wrong, or that there's a subtle
behavior in VPSM's simplifications that causes this results (e.g. the
removal of the playable JTs cases, which are more frequent than KTs,
or possibly the spurious listing of hands like QJ9s while KQ9s isn't
there).

JBQ