[PrecisionL]

Our Tennessee Club in July started using pre-dealt hands using Deal Master Pro software.

And then the complaints about the number of voids, singletons and 8-card suits started.

Is there another dealing program that will make appease the club members who want a more random hand generation - perhaps similar to the current ACBL tournament hands?

Larry

[jnichols]

You can have Dealmaster generate a large number of hands and report the frequencies of the various suit distributions and of the high card points. It includes a table with the frequencies for all possible hands.
Try this for about 100,000 hands and I think you will find that the match is quite close.

Remember - most peoples hand shuffling does not give random results.

There are a number of other dealing programs out there. One well liked program - BIg Deal - is now included in Dealmaster. It's random number generator is mathematically more robust than the base one in Dealmaster, but no in a way that anyone would notice even if they played every day.

[PrecisionL]

After 90 days and 60 deals and 2340 hands, most players are not convinced that this program approaches statistical norms.

Thanks for the reply.

[rbforster]

 PrecisionL, on 2011-October-11, 14:58, said:

After 90 days and 60 deals and 2340 hands, most players are not convinced that this program approaches statistical norms.

Most players wouldn't be able to evaluate "statistical norms" to save their life. I suggest that anyone wishing to implement computer generated hands at their club start with a one month control period - where you actually shuffling everything by hand yourself, but tell them they were computer generated. Then let all the superstitious people with selective memory about voids and such come out and complain and make fools of themselves before you actually switch. "Hand shuffling is too erratic, and since everyone complained we're switching to a provable random solution".

[PrecisionL]

 rbforster, on 2011-October-11, 17:23, said:

Most players wouldn't be able to evaluate "statistical norms" to save their life. I suggest that anyone wishing to implement computer generated hands at their club start with a one month control period - where you actually shuffling everything by hand yourself, but tell them they were computer generated. Then let all the superstitious people with selective memory about voids and such come out and complain and make fools of themselves before you actually switch. "Hand shuffling is too erratic, and since everyone complained we're switching to a provable random solution".

True, but my partner is a math major and I am a statistics minor and both of us are suspicious of the non-randomness of the 2340 hands so far.

[wyman]

 PrecisionL, on 2011-October-11, 17:54, said:

True, but my partner is a math major and I am a statistics minor and both of us are suspicious of the non-randomness of the 2340 hands so far.

Ok I'll bite.

Can you provide some statistics of what you've seen so far?

[Cascade]

 rbforster, on 2011-October-11, 17:23, said:

Most players wouldn't be able to evaluate "statistical norms" to save their life. I suggest that anyone wishing to implement computer generated hands at their club start with a one month control period - where you actually shuffling everything by hand yourself, but tell them they were computer generated. Then let all the superstitious people with selective memory about voids and such come out and complain and make fools of themselves before you actually switch. "Hand shuffling is too erratic, and since everyone complained we're switching to a provable random solution".

Indeed I did something similar when I introduced computer dealt hands at a club where I was directing.

I dealt computer hands for three weeks of a four week competition including hand records. The complaints mounted. On the fourth night I shuffled and dealt every hand by hand. I just made sure that every deck was thoroughly shuffled. I then entered every hand into the software and produced the appropriate hand records. I received more complaints on the fourth night about the nasty computer hands than I had on any other night.

[Hanoi5]

I have a player of international record tell me that each hand dealt by the computer has been played before. She's always saying how she knew this was gonna happen, how she should have done this or that, how his/her partner should have known, 'don't you remember this hand?' etc.

[aguahombre]

I am surprised after all these years of computer-generated hands at tourneys that people are still saying these same old things.

Time to get over it If someone feels a specific program is being less than random, there must be ways of checking for flaws and fixing them if found.

[Free]

People in my club also complained when the club introduced computer dealt hands. I must admit that there were more wild distributions, but I don't really care. After a while people accepted the fact that they were now playing 'really' random deals (yeah yeah, it's not truely random, I know), while they weren't in the past.

[han]

After using dealmaster pro for many years, I must say I really agree with Larry. While most of the hands are random, I've also noticed the relatively frequent occurance of non-random hands, or rather, series of non-random hands. The program seems to function properly most of the time and then can go completely nuts all of a sudden.

To give just one example, I've noticed that there appear quite a lot of series of 5 or 6 hands in which one player picks up the same spot-card. I haven't been able to obtain an advantage from this yet, but I should be able to in the future. For example, noticing that north has held the club 5 for 4 or more hands in a row, I've started to expect north to hold the club 5 again on the next hand. It's not certain of course (dealmaster pro is not completely non-random) but from experience it is certainly higher than the 1/4 chance you would expect.

Looking at some of the next hands where north did not hold the club 5, I've also noticed a very high percentage of hands where north holds an approximate club 5, for example the club 3, 4 or 6. I don't have hard numbers but I am convinced there is a correlation here.

BTW, I was also a math major.

[mikestar13]

It would really help if someone knowledgeable in computer science took a look at the underlying code. If the code, for example, uses the standard C library random number generator, these complaints are likely valid--it is a very poor random number generator intended for quick "randomish" values for testing--no one should use it for serious simulations (such as bridge dealing) where a good approach to randomness is desirable. If the code, on the other hand, uses a good random number generator such as the Mersenne Twister, I suspect we are seeing artifact of selective memories. By the way, I am not a math major, but I have a degree in Computer Science and Programming.

[Trinidad]

 mikestar13, on 2011-October-13, 03:25, said:

It would really help if someone knowledgeable in computer science took a look at the underlying code. If the code, for example, uses the standard C library random number generator, these complaints are likely valid--it is a very poor random number generator intended for quick "randomish" values for testing--no one should use it for serious simulations (such as bridge dealing) where a good approach to randomness is desirable. If the code, on the other hand, uses a good random number generator such as the Mersenne Twister, I suspect we are seeing artifact of selective memories. By the way, I am not a math major, but I have a degree in Computer Science and Programming.

Given that the number of bridge deals is a little less than 2⁹⁶, one would need a 96 bit Mersenne Twister. That requires a not so standard PC. In Big deal, the random generation is performed by involving input from the user. (As an example: In the front end by Kaj Backas, one needs to click a number of times with the mouse. The timing and location of the mouse clicks provide the randomness.) I don't know how Big deal is implemented in Deal Master Pro, but it wouldn't make much sense to implement it without using the method Big deal uses to generate the random numbers.

Rik

[Hanoi5]

 Trinidad, on 2011-October-13, 05:57, said:

Given that the number of bridge deals is a little less than 2⁹⁶, one would need a 96 bit Mersenne Twister. That requires a not so standard PC. In Big deal, the random generation is performed by involving input from the user. (As an example: In the front end by Kaj Backas, one needs to click a number of times with the mouse. The timing and location of the mouse clicks provide the randomness.) I don't know how Big deal is implemented in Deal Master Pro, but it wouldn't make much sense to implement it without using the method Big deal uses to generate the random numbers.

Rik

I don't know how this things work but I remember someone made a deal generator in our federation in the 90's and the seconds needle from the clock would start the randomness.

[Cascade]

 Trinidad, on 2011-October-13, 05:57, said:

Given that the number of bridge deals is a little less than 2⁹⁶, one would need a 96 bit Mersenne Twister. That requires a not so standard PC. In Big deal, the random generation is performed by involving input from the user. (As an example: In the front end by Kaj Backas, one needs to click a number of times with the mouse. The timing and location of the mouse clicks provide the randomness.) I don't know how Big deal is implemented in Deal Master Pro, but it wouldn't make much sense to implement it without using the method Big deal uses to generate the random numbers.

Rik

Indeed I just assumed it was a good generator, although I haven't used it that much for that purpose, since it claimed to be using Big Deal.

Are there not alternative options to Big Deal in DMPro?

[Free]

 Hanoi5, on 2011-October-13, 12:59, said:

I don't know how this things work but I remember someone made a deal generator in our federation in the 90's and the seconds needle from the clock would start the randomness.

The problem is that for a 2⁹⁶ bits random number generator, you should theoretically be able to start with 2⁹⁶ seed values. Using time and date doesn't get close to that. Having to click and/or type multiple times gets you easily to that many possibilities, although most people will only press a few buttons over and over again. If you can't start with that many seed values, then you'll miss out on a huge amount of deals!

Personally I prefer a TRNG which doesn't require the input of a human being (please leave me alone when I want to generate random deals). It's perfectly possible, but for some reason (time consuming perhaps) it's just not used. A hybrid method is to generate the seed with a TRNG and use the PRNG to generate the deals more quickly.

[mycroft]

Free, for the modern RNG-seeders, it doesn't matter *what* you type, just when. They use things like the 3rd or 4th least-significant-bit of the time difference between actions (the LSBs on certain hardware can be highly biased, so they're not used) where the clock speed is in the GHz range (so the 3rd or 4th LSB is still oscillating at 100+MHz). If you're able to control your typing to the microsecond, you'll be able to influence the RNG (though not much, because they also bring in things like "number of free sectors on the disk, system uptime, process id, ... that will be somewhat different every time). (Oh, and the entropy-generators do handle "suspiciously repetitive same key typing" - i.e. holding a key down and letting it autorepeat - sanely).

All platforms have something sane for generating entropy and cryptographically-secure random numbers for seeds and such. I don't know how much fiddling would have to be done while generating 96x36 bits of true randomness, but I do know that for SSL 2048-bit keys, it's about 5 seconds on my Core 2 Quad - at least the first time, when there's 4096 bits of entropy in the /dev/random pool.

Edit: did a test (opened /dev/random as a file in Python, dumped out consecutive 16-byte (96-bit) strings. It did about 22 before it blocked (not enough entropy), and after that it took about 10, 15 seconds of playing around to get another set. I'm sure if I was playing FreeDoom in another window, it would have gone faster :-)

I used to say that for big tournaments, they should be using a provably "true random", BigBook method. I don't, now. Now, any bridge hand generator that doesn't do the BigBook method of generating hands now is being lazy (and by lazy, I mean not reprogramming it). I realize that physical security is even now more of a danger than cryptanalysis, but we should be able to see the "set bits" and the BigBook "page" for every hand generated, on the hand records. It's not hard.

[paulg]

I have been in contact with the author of the program and he has asked me to post the following:


"From the author of Dealmaster Pro. We are one hundred percent certain that the deals created by DM Pro for club/tournament play are statistically accurate. The FAQ’s (in recent releases of DM Pro) has statistics of 750,000 DMPro deals where the shapes are compared to Standard Expected. The data shows no bias toward excessive voids or singletons. Help file topic “BigDeal” has information comparing the shapes of one million DM Pro deals to one million BigDeal deals. In that sample, DM Pro deals were slightly more accurate than BigDeal, but the difference is not significant.

Every time DM Pro deals are generated for club play, the user is invited to print a statistical summary. The data is accumulated. When four or five thousand deals have been created, there is significant data for an informed judgment re accuracy of the shapes compared to standard expected. In addition, the “BigDeal” help topic tells an easy way to create 50,000 deals and statistical summary quickly. Finally, we believe DM Pro can create every possible bridge deal. No one has proven it cannot.

DM Pro uses a unique deal generation process. No one “seed” value is required. The above mentioned data prove its effectiveness. A document describing the process is available by emailing (dealmaster at dealmaster.com)."

[randygen]

I had the same types of complaints at my club and so I did a test run of 2,000,000 hands. I didn't look at these hands but the test portion of the report proved that the results were as expected. You can do the same. That ended the controversy. Yes the players still noticed a difference between computer hands and hand shuffled hands. Have you ever watched your opponents shuffle in team games? Most shuffle only twice. Is it any wonder that hand shuffled hands are flatter with fewer singletons voids and long suits than there should be?

[Lobowolf]

 PrecisionL, on 2011-October-11, 14:34, said:

Our Tennessee Club in July started using pre-dealt hands using Deal Master Pro software.

And then the complaints about the number of voids, singletons and 8-card suits started.

Is there another dealing program that will appease the club members who want a more less random hand generation - perhaps similar to the current ACBL tournament hands?

Larry

FYP?!