Current computer bridge dealing in the U.S.A.

[Douglas]

After much searching on the Internet, I think the following is a fair
listing of the current generally accessible computer bridge dealing
sources:

1. Deal Master Pro 6. Computer software dating from the days of DOS.
It is limited to 99 deals in the usual non-batch mode.

2. BridgeComposer. Available in various versions. I think an older
version is the source of the CommonGame deals. It deals from its
built-in random source, and has the option to deal using BigDeal 1.2.

3. Dealer4. Freely available downloadable software which accompanies Dealer4
dealing machines. It used to have the same sort of choice as BridgeComposer,
but the latest version only uses BigDeal.

4. playBridge.com. An Internet download only provider. It has a limit of 64
deals per download.

5. bridgedealer.com. An Internet download only provider. It provides only 36
deals per download; no more, no less.

6. Bridge Analyser. Apparently now completely out of business. Due of its form
of copy protection, it probably will disappear from practice quickly.

7. Miscellaneous, often quirky, dealing programs. A long-time one requires one
to program the input to it in a computer language known as TCL.

8. Bridge dealing programs built-in to bridge playing programs. One usually
needs to purchase these programs to use their dealing program.

If you think I have missed something important here, I am open to correction.

Douglas

[KWSchneider]

Deal 3.1.9 from bridge.thomasandrews.com is the best available dealing program. I’ve used a version for almost 20 years.

[Ian Zimmerman]

Post by KWSchneider
Deal 3.1.9 from bridge.thomasandrews.com is the best available dealing
program. I’ve used a version for almost 20 years.

There is also redeal, which is a reimplementation (not necessarily a
clone) of deal:

https://github.com/anntzer/redeal

[Douglas]

Post by KWSchneider
Deal 3.1.9 from bridge.thomasandrews.com is the best available dealing
program. I’ve used a version for almost 20 years.

I do not want to learn another programming language at my advanced age.

I have a 5 x 99 deals comparative analysis project ready to go with the
three most successful commercial dealing programs today.

I note Deal 3 currently has a .pbn output ability. If it can do 99 at
a time, I invite you to send me 5 of those files to include for comparison.

Douglas

[KWSchneider]

Deal 3 can do as many deals as you want (until you overload the cache - around 2million, I think, on my computer). But then you simply can “batch” a series of 1million deal runs until the cows come home in pretty much any output format you want.

[Steve Willner]

Gold standard for producing deals for competition is Big Deal.

Hans van Staveren's 'dealer' is old, but I find its language very easy
to program. Kurt mentioned Thomas Andrews' program 'deal', which I
think is functionally equivalent but seems to me harder to program. It
has the advantage of an easier interface to double-dummy analyzers.
Both of these will generate as many deals as you want.

[Douglas]

My updated analytic format is six categories:

1. 4-4-3-2
2. 5-3-3-2
3. 5-4-3-1
4. 5-4-2-2
5. 4-3-3-3 + 6-3-2-2
6. Remaining 33 hand shape expected probabilities.

I've identified 4 currently available computer bridge dealing
programs which allow me to deal 99 deal hand records. I wish
playBridge.com deals could be included, but alas, they have
a limit of 64 deals maximum per dealing.

I initially dealt out ten 99 deal hand records from each of
the four dealing programs. I was left with several questions
after looking at their analysis results. So I extended my
dealing to thirty 99 deal hand records for each of the four
dealing programs.

1. Deal 319
2. Deal Master Pro 6
3. BridgeComposer 5.76
4. Dealer4 BigDeal

Douglas

[Douglas]

Post by Douglas
After much searching on the Internet, I think the following is a fair
listing of the current generally accessible computer bridge dealing
1. Deal Master Pro 6. Computer software dating from the days of DOS.
It is limited to 99 deals in the usual non-batch mode.
2. BridgeComposer. Available in various versions. I think an older
version is the source of the CommonGame deals. It deals from its
built-in random source, and has the option to deal using BigDeal 1.2.
3. Dealer4. Freely available downloadable software which accompanies Dealer4
dealing machines. It used to have the same sort of choice as BridgeComposer,
but the latest version only uses BigDeal.
4. playBridge.com. An Internet download only provider. It has a limit of 64
deals per download.
5. bridgedealer.com. An Internet download only provider. It provides only 36
deals per download; no more, no less.
6. Bridge Analyser. Apparently now completely out of business. Due of its form
of copy protection, it probably will disappear from practice quickly.
7. Miscellaneous, often quirky, dealing programs. A long-time one requires one
to program the input to it in a computer language known as TCL.
8. Bridge dealing programs built-in to bridge playing programs. One usually
needs to purchase these programs to use their dealing program.
If you think I have missed something important here, I am open to correction.
Douglas

Deal 319:

Pro: The price is right! None. Very informative PBN output.

Con: For MS Window's users, it operates in DOS mode only.

Several times now, I've seen BigDeal referred to as the
"gold standard." If so, I suggest Deal 319 be referred to
as the original "standard." In my thirty 99 deal hand records
there was only one close to unusual result obvious to me.


The grand total of all thirty hand records was very close
to being too even. That means the final six z_values were
had an almost too low a sampling standard deviation.

To those who find flat hands a problem, the 4-3-3-3 total
was almost exactly as predicted.

Douglas

[KWSchneider]

CON: Operates in DOS-mode only.

This is a huge plus. You can set up and run batch files...
Clearly you have no background in simulations.

[Peter Smulders]

Post by Douglas

Post by Douglas
After much searching on the Internet, I think the following is a fair
listing of the current generally accessible computer bridge dealing
1. Deal Master Pro 6. Computer software dating from the days of DOS.
It is limited to 99 deals in the usual non-batch mode.
2. BridgeComposer. Available in various versions. I think an older
version is the source of the CommonGame deals. It deals from its
built-in random source, and has the option to deal using BigDeal 1.2.
3. Dealer4. Freely available downloadable software which accompanies Dealer4
dealing machines. It used to have the same sort of choice as BridgeComposer,
but the latest version only uses BigDeal.
4. playBridge.com. An Internet download only provider. It has a limit of 64
deals per download.
5. bridgedealer.com. An Internet download only provider. It provides only 36
deals per download; no more, no less.
6. Bridge Analyser. Apparently now completely out of business. Due of its form
of copy protection, it probably will disappear from practice quickly.
7. Miscellaneous, often quirky, dealing programs. A long-time one requires one
to program the input to it in a computer language known as TCL.
8. Bridge dealing programs built-in to bridge playing programs. One usually
needs to purchase these programs to use their dealing program.
If you think I have missed something important here, I am open to correction.
Douglas

Pro: The price is right! None. Very informative PBN output.
Con: For MS Window's users, it operates in DOS mode only.
Several times now, I've seen BigDeal referred to as the
"gold standard." If so, I suggest Deal 319 be referred to
as the original "standard." In my thirty 99 deal hand records
there was only one close to unusual result obvious to me.
The grand total of all thirty hand records was very close
to being too even. That means the final six z_values were
had an almost too low a sampling standard deviation.
To those who find flat hands a problem, the 4-3-3-3 total
was almost exactly as predicted.
Douglas

Read Hans van Staveren's website for why BigDeal came into existence.
https://sater.home.xs4all.nl/doc.html
Randomness is not the only consideration, but also security.
It is also a very robust program that has not been changed since 2000.

Nothing bad about Deal 3.1.9 but since it has options affecting the
shape of deals it is not suitable as a standard for tournaments.

[Bertel Lund Hansen]

Post by Peter Smulders
Read Hans van Staveren's website for why BigDeal came into existence.
https://sater.home.xs4all.nl/doc.html

I'm impressed. A very clear article and easy to read.

[KWSchneider]

While BigDeal is the “go to” program for dealing in tournament play, Deal 3.1.9 has the tools and an adequate level of pseudo-randomness to be the front-end in simulation work.

[Douglas]

...Deal 3.1.9 has the tools and an adequate level of pseudo-randomness
to be the front-end in simulation work.

That is the rub. What is "adequate" pseudo-randomness objectively? And
how do you know that objectively? Do I merely have to take your
pronouncement as fact?

I am sorry, but I do not find this satisfactory. I think this state
of affairs leads to myth and superstition.

Douglas

[Douglas]

Dealer4 BigDeal 4.82

Pro: Free download. Simple Windows interface. Excellent
hand shape stat presentation.
Con: Only a single stat is presented.

So here is the "gold standard."

Everything looks OK. Until I notice it is a full 10 points
less on average for the 30 hand records than Deal 319. And
Deal 319 was somewhat less than the expected value of 1 for
the standard deviations of the z-values.

Translated into meaning: Overall, somewhat less
randomness (volatility) on average then Deal 319.

This is early days yet.

Douglas

[Douglas]

BridgeComposer 5.76

Pro: Regularly updated Windows program. Nominal price.
Deals, records, and publishes almost seamlessly.
Con: Use of BigDeal dealing option is somewhat awkward.

Volatility is the same as Deal 319. A good thing. What is
jarring is the standard deviation of the 30 standard
deviations is 0.41.

Standard deviation for a uniform distribution is about 0.29.
You can think of it as the possible middle value for unknown
distributions.

For the wonks: Maximum standard deviation (the limit) for
the N divisor is 0.5. For N-1 it is anything less than 0.5.

In my 30 plus years of this kind of experimenting, this is
the only the second time I measured over 0.40.

It means my 30 standard deviations of the z_values were
either relatively high or low, with almost none in the middle.

I think it would be good if BridgeComposer made BigDeal
the de facto dealing non-choice just like Dealer4 did
some time ago.

Douglas

[Douglas]

Deal Master Pro 6

In spite of what Steve Willner thinks, random bridge
deals come as not random enough, too random, and OK.
OK apparently comes in degrees.

So far, the three dealing programs previously
reviewed are degree "merely OK."

Deal Master Pro 6 is definitely too random by quite
a bit. The mean of its 30 hand record standard
deviations of the z_values is 1.09, a full 17 points
greater than the Deal 319 and BridgeComposer means.

What makes it definite is the number of greater
than plus and minus 1.96 z_values. 17 out of the
possible 180 in 30 hand records. I make that out
to be the p value equivalent of positive 2.36 z.

Douglas

[Douglas]

So, I'm told BigDeal can deal all the possible bridge deals.
All 2^96 (29 decimal digits) worth of them.

What to do, what to do?

Well, Dealer4 is limited to dealing 640 bridge deals in one
PBN file. So I decided to do 100 of those for Dealer4, and
100 for Bridge Analyser with the same 640 deals size. I am
a fan of directly comparable sampling when possible.

Both dealers use cipher algorithms.

Dealer4 BigDeal uses a 160 bit generator reduced to a 2^96
number output if I understand correctly. So it apparently
matches one random number possibility to one unique bridge
deal. I am curious as to how that works in practice?

Bridge Analyser uses a ISAAC cipher algorithm with a claimed
2^8192 (2,467 decimal digits) worth of random numbers before
they repeat. This length is traditionally known as the "period."
Needless to say, it does not lack adequate numbers for the task
at hand in this lifetime; and many others thereafter.

Here are Bridge Analyser's cumulating bias numbers:

1.051
1.024
1.021
1.007
1.015
1.013
0.997
0.993

Optimal is 1.000. This is a near perfect imperfect. It is
also strong evidence that I over-sampled in this experiment.

Her are Dealer4's cumulating bias numbers:

0.867
0.889
0.948
0.924
0.942
0.959
0.954
0.957

This is a final minus 1.50 standard error from optimal. It is
strong evidence that BigDeal does not access all 2^96 different
bridge deals.

Also the standard error of the final standard deviation
exceeded the plus 1.96 limit by a bit. That is slightly
excessive volitility.

These are better results than my previous report to you. But
I am afraid your "gold standard" is a teeny bit tarnished.

Douglas

[Lorne]

Post by Douglas
So, I'm told BigDeal can deal all the possible bridge deals.
All 2^96 (29 decimal digits) worth of them.
What to do, what to do?
Well, Dealer4 is limited to dealing 640 bridge deals in one
PBN file. So I decided to do 100 of those for Dealer4, and
100 for Bridge Analyser with the same 640 deals size. I am
a fan of directly comparable sampling when possible.
Both dealers use cipher algorithms.
Dealer4 BigDeal uses a 160 bit generator reduced to a 2^96
number output if I understand correctly. So it apparently
matches one random number possibility to one unique bridge
deal. I am curious as to how that works in practice?
Bridge Analyser uses a ISAAC cipher algorithm with a claimed
2^8192 (2,467 decimal digits) worth of random numbers before
they repeat. This length is traditionally known as the "period."
Needless to say, it does not lack adequate numbers for the task
at hand in this lifetime; and many others thereafter.
1.051
1.024
1.021
1.007
1.015
1.013
0.997
0.993
Optimal is 1.000. This is a near perfect imperfect. It is
also strong evidence that I over-sampled in this experiment.
0.867
0.889
0.948
0.924
0.942
0.959
0.954
0.957
This is a final minus 1.50 standard error from optimal. It is
strong evidence that BigDeal does not access all 2^96 different
bridge deals.
Also the standard error of the final standard deviation
exceeded the plus 1.96 limit by a bit. That is slightly
excessive volitility.
These are better results than my previous report to you. But
I am afraid your "gold standard" is a teeny bit tarnished.
Douglas

You may have probably considered this, but just in case I will mention
that dealing all 2^96 possible hands is not the full story - you must
deal each with the same probability.

As there are 2^225 ways to deal 52 cards (with over 8 billion ways to
deal each 13 card hand) I think it is best to program on the basis that
you can create all 2^225 ways to deal 52 cards and hence hope to get all
2^96 different hand sets with equal probability.