Bridge Hands on Bridgebase.

[smerriman]

—The MP hands typically had multiple decisions per hand designed to test one’s technique and appetite for risk in pursuing overtricks, saving undertricks, ruffing losers, finesses and other cardplay devices, establishing side suits, etc. (NOTE: The MP hands compared each to another didn’t have the same decisions, and these decisions were only occasionally influenced by distributions, splits and the like. All good bridge players know that the game is not as simple as distributions and splits.)

—The IMP hands were ridiculously simple by comparison.

I think this is absolutely true. But it's also absolutely true for 100% random hands as well - playing MPs is far, far more complex than playing IMPs.

 

It's well known that when playing IMP tournaments, the majority of hands are completely meaningless - the whole tournament comes down to a small number of decisions in a small subset of hands - while when playing MP, every trick in every hand is important.

 

So the fact there are far more decisions to be made in MP vs IMPs doesn't imply non-randomness - it's what you'd expect as a baseline.

 

Well then, do you know of an easy way to state a specific, simple quantitative factor that one could use to test hands for a “preponderance of scoring-method-specific decisions to be made”? I don’t.

 

Nope, I have no idea how you could possibly test this. The fact that neither of us do in a sense strengthens the argument against you - the only way BBO would be able to bias such hands is for them to have an algorithm for testing this, so they knew which hands to throw out, and do so in a way that doesn't affect the overall distributions, etc. I do not believe they are capable of having such an algorithm.

 

Since a large scale test is unfeasible, I would recommend you get a source of deals dealt via BigDeal, load them through a teaching table, and play 'as if it were a challenge'. I am confident that you would see exactly the same 'obvious' signs that the MP versions involve considerably more decisions.

 

If the manipulation is indeed obvious, and you go into the test with the right mindset (which is tough, since you'll be wanting to look for evidence you're right, rather than being unbiased), you should notice the difference straight away. You can then elaborate on specific differences that you noticed, with examples.

[smerriman]

Actually, scrub the last paragraph. Easy to run that test without bias - just have someone else mix in groups of deals that come from BigDeal, and deals which come from challenges (or even simpler, BBO IMP vs BBO MP). Your task would be to guess which was which. If you are correct that there is an obvious difference, you should be able to guess consistently better than average.

 

Let me know if you're game and I can set up the deals for you.

[mythdoc]

Nope, I have no idea how you could possibly test this. The fact that neither of us do in a sense strengthens the argument against you - the only way BBO would be able to bias such hands is for them to have an algorithm for testing this, so they knew which hands to throw out, and do so in a way that doesn't affect the overall distributions, etc. I do not believe they are capable of having such an algorithm.

 

Actually, it would be quite easy for BBO servers to create this outcome. Thousands upon thousands of hands are played on BBO every day, generating scores in both MP’s and IMPs. Hands played at anonymous tables, hands played at live tables, there is no shortage.. All that is necessary is to recycle these hands, making sure not to deliver the same hand twice to the same user, and in the meantime, dropping out a few of the flattest and/or selecting out some of the boards generating wider swings.

 

I want to make one final point that I think is likely to have been missed in all this back and forth. I don’t think there is some nefarious plot afoot to make BBO less fair for the average user. My belief is that, over the long haul, better players will get better results and lesser players will get lesser results — perhaps even more so if there is a filter being used to select harder or suppress easier hands. I, for one, welcome this challenge, if you’ll pardon the pun. I do speculate, however, that there may well be (at least) two forms of motivation that a profit-making online bridge website would find worthwhile in using such a filter: 1) it could provide a more engaging, more valuable experience for the player spending .40 cents US, or whatever it costs in your local currency, to play more interesting hands as opposed to a daylong with 2 or 3 out of the 8 hands flat. 2) it could lessen instances when the robots generate flat boards by making an embarrassingly bad play (like leading an ace against certain slam contracts, and enabling a lay down claim.)

 

As to your other point — namely that MP’s hands and MP tournaments are inherently harder... sure! But surely you aren’t saying that an experienced bridge player can’t assess the difficulty of a a given hand and (more importantly) its likelihood for generating a swing at MP’s scoring vs. at IMPs scoring. I also won’t take you up on your offer send me hands to identify. Take the hour you would spend generating deals for me to look at, and look at them yourself. Remember, they are “just declare,” “non-best-hand” MP challenges and IMPs challenges.Thanks for reading. mythdoc out.

[smerriman]

But surely you aren’t saying that an experienced bridge player can’t assess the difficulty of a a given hand and (more importantly) its likelihood for generating a swing at MP’s scoring vs. at IMPs scoring.

Huh? I'm saying that if BBO is biasing the hands and you think this is obvious, then it should be *easy* for you to look at the hands and determine which are the biased sets. If there is no bias, then you will unable to do so.

 

If you're unwilling to take a simple test that will actually prove whether your claim is right or wrong, then I guess enough said, and there's no point anyone spending any more time on it.

[hrothgar]

Huh? I'm saying that if BBO is biasing the hands and you think this is obvious, then it should be *easy* for you to look at the hands and determine which are the biased sets. If there is no bias, then you will unable to do so.

 

If you're unwilling to take a simple test that will actually prove whether your claim is right or wrong, then I guess enough said, and there's no point anyone spending any more time on it.

 

Might make sense to do this with one set of hands that are mixed up and then labelling individual hands.

[mythdoc]

Why not let everyone play? Just please understand that it is not scientific and doesn’t prove that BBO conditions hand delivery for the two scoring types. It only suggests it.

 

These two sets of four hands (four board robot challenges, just declare, non best hand) were generated this morning. Only these two sets were dealt. I didn’t pick and choose among sets, lol. If an algorithm was being used to enhance scoring swings, which set do you think would be MP’s scoring, and which would be IMPs scoring?

 

SET 1

 

https://tinyurl.com/ygh6cpy7

 

https://tinyurl.com/yhp5l466

 

https://tinyurl.com/yh79nqt5

 

https://tinyurl.com/yzpxg35h

 

SET 2

 

https://tinyurl.com/ygrz9jj8

 

https://tinyurl.com/ygrsto4k

 

https://tinyurl.com/yetatguk

 

https://tinyurl.com/yge2hw3x

[mythdoc]

And also, I’d love a reply to this paragraph that I wrote above, that you guys ignored. smerriman, you said it would be impossible for BBO to have a computer sophisticated enough to generate conditioned hands. I wrote:

 

Actually, it would be quite easy for BBO servers to create this outcome. Thousands upon thousands of hands are played on BBO every day, generating scores in both MP’s and IMPs. Hands played at anonymous tables, hands played at live tables, there is no shortage.. All that is necessary is to recycle these hands, making sure not to deliver the same hand twice to the same user, and in the meantime, dropping out a few of the flattest and/or selecting out some of the boards generating wider swings.

 

Do you agree, then, it is quite doable?

[mycroft]

That would last exactly until someone pointed out on the forums to one person's post that "when they held that hand in the main room, the auction went..." It hasn't happened.

 

It's possible, but it fails on the corporation test: "how much money does it cost? How much money does it make? How expensive would it be to get caught? How easy would it be to find?" "Lots", "none", "very", and "I wouldn't bet against it falling out of Nic Hammond going full bore" respectively, IMHO.

 

David Stevenson once said approximately, on these forums:

 

There are three types of bridge sessions:

1. hand-dealt, exciting and swingy: "These hands sure are weird tonight."
   
2. computer-dealt, exciting and swingy: "Those damned computer hands again."
   
3. flat, normal hands, hand- or computer-dealt: "Thanks for the game, James."

[hrothgar]

Why not let everyone play? Just please understand that it is not scientific and doesn’t prove that BBO conditions hand delivery for the two scoring types. It only suggests it.

 

These two sets of four hands (four board robot challenges, just declare, non best hand) were generated this morning. Only these two sets were dealt. I didn’t pick and choose among sets, lol.

 

No offense, but this isn't how you set up this kind of experiment:

 

I propose the following

 

1. We establish a set of rules (in advance) regarding how sets of deals will be selected. For example we might chose the first 100 IMP boards and the first 100 MP boards on some given day. I don't much care what the rules are, but rather, that some objective method is chosen in advance.

 

2. That set of 200 boards (or however many might get chosen) get mixed up.

 

3. People vote on individual boards

 

4. We assess whether folks ability to identify MP or IMP hands differs from a coin tossing exercise.

[pilowsky]

This doesn't sound like a mathematics problem as much as a Turing test problem.

How do you define "interesting" mathematically? Who decides?

It really sounds as though the wrong question is being asked. In which case, no satisfactory answer is possible.

Why not climb back out of Alice-in-wonderland world?

What is needed is an 'interest test.'

Here's a simple example. Find 100 boards that mythdoc played 100 days ago and see if he bids/plays them in the same way.

The percentage variance from the original bidding and play would be an "interest index".

 

 

[hrothgar]

This doesn't sound like a mathematics problem as much as a Turing test problem.

How do you define "interesting" mathematically? Who decides?

 

I think that you are making things way too complicated.

 

Mythdoc made a specific claim:

 

He said that he could take a sip from a cup of tea and determine whether the tea had been added to the cup first or the milk had been added to the cup first (or something like that)

 

We have plenty of cups of tea

They have labels...

[pilowsky]

I think that you are making things way too complicated.

 

Mythdoc made a specific claim:

 

He said that he could take a sip from a cup of tea and determine whether the tea had been added to the cup first or the milk had been added to the cup first (or something like that)

 

We have plenty of cups of tea

They have labels...

 

It still sounds like you are trying to make tea in a cracked pot - so to speak.

Skim milk for me, please.

[smerriman]

Just please understand that it is not scientific and doesn’t prove that BBO conditions hand delivery for the two scoring types. It only suggests it.

Done properly, it is entirely scientific.

 

If I give you 100 sets, 50 that came from IMPs tournaments, and 50 that came from MP tournaments, and you only identify 50% of them, it completely disproves your statement that the difference is obvious.

 

If you are able to even correct predict, say 70 of the 100 (and you're claiming you'd be able to tell far more than that), that is enough proof for me to show that you are correct, as the probability that happens by chance is negligable.

[hrothgar]

It still sounds like you are trying to make tea in a cracked pot - so to speak.

Skim milk for me, please.

 

Take it up with Ronald Fisher...

 

He's the one who proposed it.

[AL78]

 

David Stevenson once said approximately, on these forums:

 

There are three types of bridge sessions:

1. hand-dealt, exciting and swingy: "These hands sure are weird tonight."
   
2. computer-dealt, exciting and swingy: "Those damned computer hands again."
   
3. flat, normal hands, hand- or computer-dealt: "Thanks for the game, James."

 

As far as I'm concerned there are three types of bridge sessions:

1. Below average HCP, struggele to contribute anything, poor score at the end. Feels like a waste of an evening.

2. Below average HCP, pick up hands where I can at least play an active role even if I have to defend three quarters of the evening. Final score anything from poor to reasonable, a fair or enjoyable evening.

3. I declare on at least a quarter of the hands and/or the hands are biased our way, final score is fair to decent. An enjoyable evening.

[AL78]

Done properly, it is entirely scientific.

 

If I give you 100 sets, 50 that came from IMPs tournaments, and 50 that came from MP tournaments, and you only identify 50% of them, it completely disproves your statement that the difference is obvious.

 

If you are able to even correct predict, say 70 of the 100 (and you're claiming you'd be able to tell far more than that), that is enough proof for me to show that you are correct, as the probability that happens by chance is negligable.

 

I believe there is a statistical test, Kolmogorov Smirnov I think, which can be used to test whether two samples come from the same distribution. It is one of those p-value tests where if the p-value is very small, the null hypothesis they are from the same distribution is rejected. If some property of the sets could be quantified numerically, it could be used (e.g. mean HCP North).

[hrothgar]

I believe there is a statistical test, Kolmogorov Smirnov I think, which can be used to test whether two samples come from the same distribution. It is one of those p-value tests where if the p-value is very small, the null hypothesis they are from the same distribution is rejected. If some property of the sets could be quantified numerically, it could be used (e.g. mean HCP North).

 

I think that it might be more appropriate to use FIsher's exact test here (or perhaps one of its more modern antecedents)

 

There is a reason that I was talking about tea a couple posts back

[mythdoc]

I think that you are making things way too complicated.

 

Mythdoc made a specific claim:

 

He said that he could take a sip from a cup of tea and determine whether the tea had been added to the cup first or the milk had been added to the cup first (or something like that)

 

We have plenty of cups of tea

They have labels...

 

No, I made no such specific claim. You interpreted my statement that the differences between the challenges I played on that occasion were obvious as amounting to a claim that I could, and would, do it over and over again like a magic trick. My interest is and has always been that others take a look at these hands and state if they see any patterns. I have no interest in submitting to self appointed lord high interrogators because your lack of good faith is beyond obvious at this point.

 

Done properly, it is entirely scientific.

 

If I give you 100 sets, 50 that came from IMPs tournaments, and 50 that came from MP tournaments, and you only identify 50% of them, it completely disproves your statement that the difference is obvious.

 

If you are able to even correct predict, say 70 of the 100 (and you're claiming you'd be able to tell far more than that), that is enough proof for me to show that you are correct, as the probability that happens by chance is negligable.

 

So, did you look at the hands I posted above? Could be a coincidence, but 3 out of the 4 IMP hands were game score contracts, whereas 4/4 of the MP hands were part score contracts. One of the trends I said one might see to tell one set from another in one of my earlier posts. Doesn’t prove anything, but certainly doesn’t disprove my hypothesis.

 

Meanwhile this is still a blunt tool to get at the question of whether flat boards are being removed from daylongs.

 

If you do want to pursue this in good faith, make it public, set up a poll, and post SETS of 12 or 16 board challenges (just declare, non-best hand) of the two types, and ask folks to pick which set is which. Then we might all judge for ourselves.

 

Or let it drop. I for one am tired of replying to your misrepresentations and playing your gotcha games.

[hrothgar]

No, I made no such specific claim. You interpreted my statement that the differences between the challenges I played were obvious as amounting to a claim that I could, and would, do it over and over again like a magic trick. My interest is and has always been that others take a look at these hands and state if they see any patterns.

 

1. You are making a bullshit claim

2. You refuse to do a modicum of work to let other people validate said claim

3. You're too stupid to realize that what I proposed is the same as what you are suggesting with a better experimental design

[hrothgar]

 

So, did you look at the hands I posted above? Could be a coincidence, but 3 out of the 4 IMP hands were game score contracts, whereas 4/4 of the MP hands were part score contracts. One of the trends I said one might see to tell one set from another in one of my earlier posts. Doesn’t prove anything, but certainly doesn’t disprove my hypothesis.

 

I didn't bother because I don't trust the way in which the the sets of hands were selected.

 

As I said before, pre commit to a way in which a reasonably sized sample of future hands can be selected and I'll do so

[pilowsky]

Take it up with Ronald Fisher...

 

He's the one who proposed it.

 

Fisher was famously irascible. You two aren't related by any chance?

[mythdoc]

I didn't bother because I don't trust the way in which the the sets of hands were selected.

 

As I said before, pre commit to a way in which a reasonably sized sample of future hands can be selected and I'll do so

 

Last I knew, smerriman could access my tourneys and could verify that the two sets were the only two run, as I said they were.

 

And as I just said, publicly publish sets of challenge hands of the types I specified (just declare, non-best-hand challenges), post them AS A SET (none of this dividing one by one bullshit), have a poll at the top with the following question:

 

“If, as a supposition, BBO was thought to be selecting sets of hands for competitions involving one human and three robots, to reduce incidence of flat boards, and was doing this both for sets of boards under MP scoring, and for separate sets of boards under IMP scoring, which of these sets (below) would you think is the MP set and which is the IMP set?”

 

Now that would be very interesting to me.

[hrothgar]

Last I knew, smerriman could access my tourneys and could verify that the two sets were selected as I said they were.

 

The issue is why you choose to post these, not whether they exist.

 

Tell you what, play 8 tournaments tomorrow (4 and 4) and post these...

I'll take a look

[mythdoc]

The issue is why you choose to post these, not whether they exist.

 

Tell you what, play 8 tournaments tomorrow (4 and 4) and post these...

I'll take a look

 

They were the only two I ran, so I should not have said “selected”. But, your offer is exactly what I have been asking, thank you. Almost as if you finally stopped misrepresenting my posts when I insisted upon it, :)

 

How many boards would you like the challenges to be? You have stated twice how precious your time is, lol. I’ll run them and post the hands sometime tomorrow.

[AL78]

I think that it might be more appropriate to use FIsher's exact test here (or perhaps one of its more modern antecedents)

 

There is a reason that I was talking about tea a couple posts back

 

My knowledge of statistical methods is not complete, and I am not familiar with Fisher's exact test. A quick look on Wikipedia and I see where the tea reference comes from.