Bridgebrowser Challenge

[han]

I have had quite a bit of fun playing with the Bridgebrowser program. The results you can get are often very interesting, and sometimes quite surprising. The occasional search that Ben shows here on the forum can often really help the discussion, in particular when you want to know how frequent a particular hand or auction comes up (and what kind of hands people have).

 

However, as awm has also pointed out, it seems almost impossible to run a bridgebrowser search that can tell you whether a certain treatment is better or worse than another. There are (at least) two reasons for this. (1) The people playing one kind of treatment may be generally better or worse players than the people playing another. (2) How you define one bid may greatly influence the definition of another bid.

 

For instance, if you want to find out whether the 10-12 NT is a winner, then it is not enough to just search for hands with 10-12 points that are opened 1NT. You would also have to search for the hands that the people who play a mini-notrump do on hands that they open 1m (for example).

 

It seems almost impossible to do this, so I pose a challenge:

 

 

Can you define and run a Bridgebrowser test that convincingly shows that one particular treatment is better than another.

 

Sorry, no cash prizes. :)

 

I started a similar thread on the homebase forum: http://forums.homebaseclub.com/index.php?s...&st=0entry842

[pigpenz]

:)

[joshs]

actuallly Hannie it should be pretty easy.

1.On the bidding screen mark the 1NT 1NT range, pick 10-12 hcp

2 go to bid analysis screen enable auxillary item

 

now do search

at end you can save your report at 1012.tpl

 

now go start over

1.on bidding screen mark 1club 1diamon pick 10-12hcp

2.go to bid analsysis screen enable auxillary item

at end you can save report as 1m1012.

 

you can comparre the CtrOS CtNo or whatever other final bids or bids that you want. It willgive you imp avg or mp%

I am missing something. The results from playing 10-12 NT have nothing to do with the results optained from opening 1N on hands with 10-12 HCP. Nothing.

 

Your sample will include:

a. people opening some of these hands, but playing a different range, perhaps 10-13 or 12-14 or 11-14.

b. people opening hands that aren't even in their NT range. For instance promoting a 12 count to a 13-15 NT opener. Or psyching a strong NT in 3'rd seat with a 10 count.

 

And furthermore, 10-12 NTers might have wide range rebids to deal with. For instance 13-16 (1N) and 17-19 (2N). How did they do on these hands? Again its impossible to tell unless you knew each players range. I am sure that a 10-12 NTer, who was not playing a strong club, and who rebid 1N on 13-16 did worse on balanced 13's than standard bidders did (whose 1m then 1N rebid was the tighter range of 12-14). The real question is, how much worse.

 

Unless the ranges the partnership use can be used to control the sample, you can't learn anything.

[han]

This may come as a surprise to you Pigpenz, but I actually thought before starting this thread. :)

 

 

Playing a mini-notrump influences many things. For example:

 

-what do you do with other balanced hands. If you open best minor with all of those, then you have to find a way to distinguish between 12-14, 15-17 and 18-19 on your next bid. Also, since you open 1m on a wider variety of hands, you may have more difficulty in competitive auctions.

 

Some would agree to open 1C with (say) all 15-17 balanced hands and 1D with 12-14 balanced hands. This means that both 1C and 1D show only 2+ cards in that suit, which means that in competitive auctions partner will be more uncomfortable raising. So playing a mini-1NT may even influence your results when you are not balanced at all!

 

I hope this is clear.

[han]

Unless the ranges the partnership use can be used to control the sample, you can't learn anything.

 

I absolutely agree Josh, and this is a big problem. Even if you can track the partnership, they might be playing a different system on another day.

[pigpenz]

:)

[cherdano]

I think it is safe to put up a cash prize. I would bet it would never be claimed :)

[pigpenz]

:)

[han]

Pigpenz, I'm not going to repeat myself, it seems useless. Of course I don't know you, but I get the impression that you are not willing to take the effort to read our posts and think about them before you write more.

[hrothgar]

♠the more variables you want to add in then the more searches you have to do.

As far as further influence in NT rebids you can also enter those into the bidding parmaters also.

 

so marking opponents as passing you can do a search where

auction goes

1♣ PASS 1♦/1♥/1♠ PASS

1NT(13-16)

1NT(15-18)

2NT(17-19)

2nt(18-19)

 

The data is there its just up to you how you want to mine it out.

Please look at Josh Sher's post.

The data is NOT there.

 

Assume that you see someone open 1NT in 1st seat holding

 

♠ AT9

♥ Q32

♦ KQT64

♣ T9

 

What is his NT range?

 

9-11?

10-12?

12-14?

 

You have no way of mapping an isloated data point onto a systemic definition...

 

To chose a more graphic example, suppose that you see a player make a 1NT overcall over RHO's 1♦ opening hold

 

♠ AQJ2

♥ K

♦ K42

♣ KT732

 

Is the NT bidder playing raptor or making a natural NT overcall?

[cherdano]

Ok, pigpenz, please explain how to analyze the effect of the NT range on the auction 1♣-(2♠) with bridgebrowser.

[pigpenz]

you people are cruel.

the original post asked how to fill in for doing a search and thats all i responded too. And I tried to explain how you can read the results. So all i can say is go buy your own copy of bridgebrowser and do it your self, OVER and OUT.

[han]

Arend, I'm glad I didn't offer any cash prizes :D.

[han]

♠the more variables you want to add in then the more searches you have to do.

As far as further influence in NT rebids you can also enter those into the bidding parmaters also.

 

so marking opponents as passing you can do a search where

auction goes

1♣ PASS 1♦/1♥/1♠ PASS

1NT(13-16)

1NT(15-18)

2NT(17-19)

2nt(18-19)

 

The data is there its just up to you how you want to mine it out.

Please look at Josh Sherr's post.

The data is NOT there.

 

Assume that you see someone open 1NT in 1st seat holding

 

♠ AT9

♥ Q32

♦ KQT64

♣ T9

 

What is his NT range?

 

9-11?

10-12?

12-14?

 

You have no way of mapping an isloated data point onto a systemic definition?

 

To chose a more graphic example, suppose that you see a player make a 1NT overcall over RHO's 1♦ opening hold

 

♠ AQJ2

♥ K

♦ K42

♣ KT732

 

Is the NT bidder playing raptor or making a natural NT overcall?

Good point Richard, I think that this difficulty is quite different from the two I already listed.

 

It seems that to make a succesful attempt, you would need to know what the definition of each bid is including a lot of negative inferences.

 

Of course, that still wouldn't solve difficulty (1).

[inquiry]

Well, the solution here is something quite different. First thing I would do is use a large Bridgebrowser database of at least 4 million hands, and probably one with 29 million.

 

Second, I would restrict the search to first seat opening bids, for simplicity

 

Third, I would search for balanced hands with 10-12 hcp in first seat. This would be set on the Bidding/HCP tab, along with shape from 4333 to 5332, because this will speed up the search which will be quite slow anyway. I would also go the VIEW tab and check the box labelled "suppress main display" as this will spead things up considerably. Then go to the Bid analysis tab and choose that as an auxillary term. Finally I would go to the Bidding tab and set shape to 4333, 4432, and 5332 (also select that as auxiallary term), also choose 4 or fewer spades and hearts.

 

The I will go back to the bid Shape/hcp tab and press search.

[Simplicity]

Second, I would restrict the search to first seat opening bids, for simplicity

 

thanks :D

[awm]

I think in order to do this you'll have to pick and choose your convention to evaluate very carefully. Ideally it should be one where:

 

(1) The ripple effect on the rest of the system is minimized.

(2) The caliber of player and whether they do this isn't very correlated.

(3) The "false positive" data arising from people actually playing a different method isn't too high.

 

The 10-12 notrumps fail pretty much all these tests.

 

One that might be a little closer to possible, is opening a weak 2♠ with 5332 distribution. It seems as though:

 

(1) Most people who open 2♠ on these hands still open 2♠ with six-card suits also. There aren't really any strong negative inferences based on failure to open 2♠ because not all hands with five cards necessarily qualify. The basic opening structure remains relatively intact.

 

(2) I see this kind of bid from really good players and really bad players. It's more a function of "aggressiveness" than skill I suspect.

 

(3) This is why I restricted to 5332, to avoid the fairly common muiderberg-type treatments. Of course it's possible someone plays multi 2♦ as a "six card weak two" and 2♠ as "five card weak two" or vice versa, and there's some weak two style out there where it shows 5-6 in the bid suit and 2-3 in every other suit and sort of intermediate values... but I'm willing to bet both of these are less frequent than people who just preempt aggressively (or not so aggressively).

 

Then again, I don't have a copy of bridge browser or a few million hands of data. I freely admit that this test is still "not perfect" but it seems a lot closer to accurate than the NT ranges.

[jdonn]

Second, I would restrict the search to first seat opening bids, for simplicity

 

thanks ;)

Now THAT is funny!

 

No one ever restricted a search for me :D

[han]

Well, the solution here is something quite different. First thing I would do is use a large Bridgebrowser database of at least 4 million hands, and probably one with 29 million.

 

Second, I would restrict the search to first seat opening bids, for simplicity

 

Third, I would search for balanced hands with 10-12 hcp in first seat. This would be set on the Bidding/HCP tab, along with shape from 4333 to 5332, because this will speed up the search which will be quite slow anyway. I would also go the VIEW tab and check the box labelled "suppress main display" as this will spead things up considerably. Then go to the Bid analysis tab and choose that as an auxillary term. Finally I would go to the Bidding tab and set shape to 4333, 4432, and 5332 (also select that as auxiallary term), also choose 4 or fewer spades and hearts.

 

The I will go back to the bid Shape/hcp tab and press search.

Just like Pigpenz, you seem to be answering a question you haven't read.

[inquiry]

Well, the solution here is something quite different. First thing I would do is use a large Bridgebrowser database of at least 4 million hands, and probably one with 29 million.

 

Second, I would restrict the search to first seat opening bids, for simplicity

 

Third, I would search for balanced hands with 10-12 hcp in first seat. This would be set on the Bidding/HCP tab, along with shape from 4333 to 5332, because this will speed up the search which will be quite slow anyway. I would also go the VIEW tab and check the box labelled "suppress main display" as this will spead things up considerably. Then go to the Bid analysis tab and choose that as an auxillary term. Finally I would go to the Bidding tab and set shape to 4333, 4432, and 5332  (also select that as auxiallary term), also choose 4 or fewer spades and hearts.

 

The I will go back to the bid Shape/hcp tab and press search.

Just like Pigpenz, you seem to be answering a question you haven't read.

Sadly you fail to understand. My search does not find balanced 10-12 hands that open 1NT. It finds ALL BALANCED hands that are 10-12 hcp, and with bid analysis separates them into each bid: pass, 1C, 1D, 1H, 1S, 1NT, etc (yes you find 2♦ bids and even more, just very few of those).

 

In addition, with bid analysis turned on, you can see how each bid did (on average), did. For example, "opening" with a pass averages a perfect 0.00 (i would bet), and you can "see" how the other bids do on the same hands. But more than that, you can "click" on the 1NT opening bids and see how opening 1NT with balanced 10 does, versus balanced 11 (the mean, sd are given). You can see after opening 1NT does the opening side do better declaring or defending, again with the plot routine?

 

Ben

 

BTW comparing me to pig probaby makes him angry with the comparison, and certainly does so to me.

[hrothgar]

Sadly you fail to understand. My search does not find balanced 10-12 hands that open 1NT. It finds ALL BALANCED hands that are 10-12 hcp, and with bid analysis separates them  into each bid: pass, 1C, 1D, 1H, 1S, 1NT, etc (yes you find 2♦ bids and even more, just very few of those).

Its perfectly obvious what you're doing.

Its also readily apparent that this approach is flawed.

 

1. If a player opens 1NT on a balanced 12 count is his range 10-12, 12-14, or 13-15 with frequent upgrades?

 

As Josh and I already noted, the fact that someone chose to open 1NT (or 1C or whatever) with a balanced hand and 10-12 HCP doesn't necessarily mean that this was their agreement. The player making the opening could have psyched or misbid or whatever. This probably isn't as important for a mini-NT but could be quite significant for other types of openings. I return to my earlier example... If a player makes a 1NT of a 1♦ opening overcall holding

 

♠ AQJ2

♥ K

♦ K42

♣ KT732

 

how do you tell whether they are playing raptor?

 

2. You are examining balanced hands with 10-12 HCP in isolation from the rest of the system. Its entirely possible that the 10-12 NT opening will score quite well, however, the system as a whole will be net negative because the 1m openings are badly overloaded. The analysis that you propose can't measure any kind of negative inferences.

[inquiry]

Richard,

 

I appreciate the impossibility to use such bids as part of a system and being a problem. I appreciate not only your complaint here, but other complaints as well. Nevertheless, the way to do approach this and similiar studies is to be very blunt about the search and very encompassing and then use bid analysis (and the new plotting feature) to examine what is happening.

 

A reasonable question is:

How good (or bad) is it to open 1NT with a balanced hand and 10 hcp, when Vul?

same but when not vul?

 

Here 1NT might be a psyche, might be part of 8-10 nt, or might be 10-12. But it gives you data on how the 1NT bid works out. Bridgebrowser data seems to suggest that four card majors work better at matchpoint than opening better minor, but less well at imps (a study i did once). This does not seperate out the 1M opening that are systemic or table feel bid.

 

But the key part of my post was how to phrase large database searches. You simply HAVE to use the Bidding/HCP tab to set initial parameters and search from that page. Then the other trick there was shape issues, making sure no five card major, and limiting to first seat. The limit to first seat is necessary,,, (you can search by any seat, btw), if you want to be sure that passes are included in the mix (otherwise you will only get data for hands where the hand with 10-12 did not pass). For what it is worth, I take a lot of the "overall statitics" from bridgebrowser with a grain of salt, until the number of hands found is quite large. I seriously doubt there will ever be enough 10 or 11 hcp 1NT opening bids to draw a good conclusion. There are lots of 12 pt 1NT opening bids (part of 12-14 i guess).

[han]

Well, the solution here is something quite different. First thing I would do is use a large Bridgebrowser database of at least 4 million hands, and probably one with 29 million.

 

Second, I would restrict the search to first seat opening bids, for simplicity

 

Third, I would search for balanced hands with 10-12 hcp in first seat. This would be set on the Bidding/HCP tab, along with shape from 4333 to 5332, because this will speed up the search which will be quite slow anyway. I would also go the VIEW tab and check the box labelled "suppress main display" as this will spead things up considerably. Then go to the Bid analysis tab and choose that as an auxillary term. Finally I would go to the Bidding tab and set shape to 4333, 4432, and 5332  (also select that as auxiallary term), also choose 4 or fewer spades and hearts.

 

The I will go back to the bid Shape/hcp tab and press search.

Just like Pigpenz, you seem to be answering a question you haven't read.

Sadly you fail to understand. My search does not find balanced 10-12 hands that open 1NT. It finds ALL BALANCED hands that are 10-12 hcp, and with bid analysis separates them into each bid: pass, 1C, 1D, 1H, 1S, 1NT, etc (yes you find 2♦ bids and even more, just very few of those).

 

In addition, with bid analysis turned on, you can see how each bid did (on average), did. For example, "opening" with a pass averages a perfect 0.00 (i would bet), and you can "see" how the other bids do on the same hands. But more than that, you can "click" on the 1NT opening bids and see how opening 1NT with balanced 10 does, versus balanced 11 (the mean, sd are given). You can see after opening 1NT does the opening side do better declaring or defending, again with the plot routine?

 

Ben

 

BTW comparing me to pig probaby makes him angry with the comparison, and certainly does so to me.

I do understand your search plan. However, I fail to see how it solves any of the 3 problems that have been mentioned. I know that you are smart enough to understand those, so I conclude that you haven't read them.

 

I don't mean to compare you and pigpenz with eachother, just your responses in this thread.

[kenberg]

I think the answer to the question "Can you define and run a Bridgebrowser test that convincingly shows that one particular treatment is better than another" is "No". However, maybe it is possible to do something. I'll simply trust that what I suggest is technologically possible. If it isn't, it will be soon as the technology advances.

 

 

Let's consider 12-14 nt versus 15-17 nt.

 

 

First, select events of a certain type, for example acbl pairs tourneys.

 

Next, identify pairs who play in many of these.

 

Third, select a hand pattern or patterns that pretty much everyone would agree is opened 1NT if the point count is within the range.

 

Fourth, search on each of the chosen pairs to find out whether, in practice, they open hands in the 12-14 range, the 15-17 range, or some other range. An exact criterion could be set for this that would allow the occasional out of range call. This would be an operational definition of "plays weak/strong nt", independent of what they may actually say they play.

 

 

This is the easy part. Now to evaluate how it fares. I think that really it cannot be done. But something can be done. You can see how pair X does in the tourneys. Say the average 58%. You can see how they did on the hands that had the shape chosen above, and hcps in each of the two ranges. Perhaps these figures would be above 58%, perhaps below. You could then, with perhaps some confidence, say things such as "On balanced hands in the 15-17 nt range, it appears to cost if you are playing weak nts, since the weak no-trumpers who scored x% overall scored only y<x percent on the 15-17 balanced hands". Since I am not in advance certain that y would be less than x, it seems that this is not a totally trivial statement. In particlar there would be some sort of quantitative estimate of the cost of various agreements in various situations.

 

 

Of course it is very limited. The auction on unbalanced hands is affected by the NT range that you play, and this is very serious. The above would not address that at all. But science advances in small steps usually, and it seems that something could be said.

 

 

I apologize in advance if you find this non-responsive. I did read all the posts, and I don't believe I am a total moron. I really doubt the overall objective can be accomplished, but I think what I suggest has not, as far as I know, been done and I think it would be interesting.

 

k

[cherdano]

I think in order to do this you'll have to pick and choose your convention to evaluate very carefully. Ideally it should be one where:

 

(1) The ripple effect on the rest of the system is minimized.

(2) The caliber of player and whether they do this isn't very correlated.

(3) The "false positive" data arising from people actually playing a different method isn't too high.

 

The 10-12 notrumps fail pretty much all these tests.

 

One that might be a little closer to possible, is opening a weak 2♠ with 5332 distribution. It seems as though:

 

(1) Most people who open 2♠ on these hands still open 2♠ with six-card suits also. There aren't really any strong negative inferences based on failure to open 2♠ because not all hands with five cards necessarily qualify. The basic opening structure remains relatively intact.

 

(2) I see this kind of bid from really good players and really bad players. It's more a function of "aggressiveness" than skill I suspect.

 

(3) This is why I restricted to 5332, to avoid the fairly common muiderberg-type treatments. Of course it's possible someone plays multi 2♦ as a "six card weak two" and 2♠ as "five card weak two" or vice versa, and there's some weak two style out there where it shows 5-6 in the bid suit and 2-3 in every other suit and sort of intermediate values... but I'm willing to bet both of these are less frequent than people who just preempt aggressively (or not so aggressively).

 

Then again, I don't have a copy of bridge browser or a few million hands of data. I freely admit that this test is still "not perfect" but it seems a lot closer to accurate than the NT ranges.

I admit this is pretty good. However, there is still the problem that

1. either partner doesn't know about/ignores this style, and therefore raises these weak twos too aggressively, or
   
2. does take it into account, and will thus lose a bit when partner has a 6-card weak two.
   

I suppose most players fall into category a.

 

Arend