accuracy and practicality of some computer-generate strategies?

[Joe_Schober]

Jean-Baptiste Queru wrote:

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.

I suspect what you're encountering here is an optimization on the part of the strategy generator. Consider the hands you list above: they're either mutually exclusive, or have already been handled by a higher line item on the strategy chart. For example, you should never have to choose between 3-to-RF-Ace-high and QJTs/KQJs, because if you have both of those, you already have RF4 or RF5, and shouldn't even be looking this far down the chart. The strategy generator saves time by not trying to rearrange these line items when there's no value in doing so.

--Joe

[Steve_Jacobs]

I agree with Jean-Baptiste. It doesn't make sense for hands like AQJs to
be rated higher than hands like KQJs. The KQJs has more ways to make
straights and straight-flushes, and the same number of ways to make other
paying hands. QJTs has less value than KQJs because the QJTs has
fewer ways to make a high pair, so KQJs and QJTs probably wouldn't
be next to each other.

The strategy simply isn't optimal, at least not for maximizing EV.

···

On Tuesday 28 June 2005 10:12 am, Joe Schober wrote:

Jean-Baptiste Queru wrote:
>> 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.

I suspect what you're encountering here is an optimization on the
part of the strategy generator. Consider the hands you list above:
they're either mutually exclusive, or have already been handled by a
higher line item on the strategy chart. For example, you should
never have to choose between 3-to-RF-Ace-high and QJTs/KQJs, because
if you have both of those, you already have RF4 or RF5, and shouldn't
even be looking this far down the chart. The strategy generator
saves time by not trying to rearrange these line items when there's
no value in doing so.

[Jean-Baptiste_Queru]

Actually, I think that this specific part of the strategy is optimal;
it certainly looks counter-intuitive, but it's OK.

The ordering of 3RF, 3 High cards, A high (AKQs, AKJs, AQJs) and (4 to
inside SF) doesn't matter, because if you have both of those in the
same hand you automatically have a pat flush, which is higher up in
the strategy (notice that it's only possible with AQJs anyway).

Similarly, the ordering of (AKQs, AKJs, AQJs) and (KQJs, QJTs) doesn't
matter, because if you have both of those in the same hand you
automatically have 4 to a RF, which is higher up in the strategy.

Therefore, the only ordering that matters is that of KQJ9s against
KQJs, and that QJT8s against QJTs, and each time the 4-card one is
better if I'm not mistaken.

My guess is that you could easily modify the strategy by hand, moving
(AKQs, AKJs, AQJs) below (4 to inside SF), and clump together all the
cases of (4 to SF) and all those of (3 to RF except with AT).

Similarly, there are at least 9 lines in that strategy which I think
can be entirely removed (they appear to be unreachable), and at least
7 more that can be merged with other lines (2 suited high cards, 2
high cards, 1 high card).

Personally, I also simplify the top of my JoB strategies by merging
together at the top all the paying hands that are kept as such (RF,
SF, FH, 3oak, 2P).

A strategy doesn't have to sort the lines by EV, as long as for each
possible dealt hand it results in the choice (out of 32 possible
choices) that results in the best EV for that deal. being extreme, you
could list all 2.6 million deals and for each of them the ideal play,
and regardless of the ordering you use you'd still have an ideal
strategy (obviously, not a practical one for a human, but for a
computer it'd be very efficient if sorted the right way).

JBQ

···

On 6/28/05, Steve Jacobs <jacobs@xmission.com> wrote:

The strategy simply isn't optimal, at least not for maximizing EV.