[tnevolin]

Hi All.
I devised new hand evaluation method. I would be glad to listen to comments and critics about it. Even better if anyone would like to try it out in practical bidding. Here is the document link below.
https://drive.google...RzVOTjZJcXlfa0k

I've analyzed 400k deals on the computer to find all the parameters and adjusted values. It also took me few years to perfect the algorithm and find the right balance of precision vs. ease of memorizing. I can go on and on about history and benefits of this approach to anyone wanting to listen but do not want to clutter this post with my opinion. I'd rather seek yours.
This evaluation technique can be directly used with any existing bidding system. However, I also devised basic bidding system that directly utilizes the power of this hand evaluation method resulting in more precise and shorted bidding sequences. I would definitely love to give it a try with anyone interesting.

[helene_t]

Hi tnevolin, welcome to the forum

It looks like a fairly simple formula. I just wonder how you derived this from computer analysis - usually, when fitting regression models, you get noninterger coeffients, but your coefficients are all integers.

Do you have any benchmarks for how well it predicts the chance of making game or slam, compared to other methods?

[tnevolin]

helene_t, on 2016-July-10, 15:53, said:

Hi tnevolin, welcome to the forum

It looks like a fairly simple formula. I just wonder how you derived this from computer analysis - usually, when fitting regression models, you get noninterger coeffients, but your coefficients are all integers.

Do you have any benchmarks for how well it predicts the chance of making game or slam, compared to other methods?

People don't like memorizing fractions. That's why they are counting in points (1/3 of a trick) just to avoid non-integer values. You are right that regression resulting coefficients are non-integer. So I did a lot of manual tweaking trying to find an integer coefficients those still pretty close to the initial regression analysis. I've used residual squares to estimate match quality (i.e. how good/bad it deviates from best non-integer values). I also applied my values to the initial set of deals and estimated how MP and IMP would improve if people used my method for contract level decisions. It gave somewhere +5% MP and +1.5 IMP.
By the way, I also have an alternative approximation with half-integer values. However, my take on it is that it doesn't improve significantly enough to justify fiddling with half points.
https://drive.google...akFNbTloYTQ0X3M

[dave251164]

You may find these links of interest:

https://www.amazon.c...n/dp/0713479140

http://www.eclipse.c...5046/midmac.htm

[The_Badger]

Hello tnevolin (Tim)

First, let me congratulate you for all the input, time and effort you have put in to devise a new hand evaluation method

Anything that makes hand evaluation easier is very welcome, but....

a)....of those 400K hands analysed, how many (approx.) are for contested auctions, non-contested auctions?
b)....were the "opponents" on these auctions using the same NHE?
c)....was there any factor involved of reassessing hands during the auction?
d)....etc. etc.

I say this as obviously I don't know your background or age or bridge experience, but I too, when I was much, much younger tinkered with various formulae and evaluations to make this game easier. At least yours is far more straightforward and a lot less complicated than some of the things I devised!

At the cutting edge, bridge bidding has become a lot more cutthroat in recent years. I am currently re-reading a book published in 1993, Partnership Bidding at Bridge by Andrew Robson and Oliver Segal. Computer simulations are all well and good, and the GIB program is excellent, but would your method be foolproof in a competitive environment?

I ask this as there already many evaluation factors out there already: Losing Trick Count, Law of Total Tricks, etc. Does your program (maybe at a different level) incorporate those? Or is it based solely on a primary points hand evaluation?

I remember those "ask the experts" bidding contests that most bridge magazines had where experts used a complex (usually relay) bidding system to establish whether a 4-4 minor suit small slam was feasible after a dozen or so bids. Establishing key cards and distribution. Can your system do that?

The point I'm making is that your system is probably very good for teaching novices and beginners how to evaluate hands initially, and that's great, but until it is used successfully at higher level expert bridge, then it's a hand evaluation system that helps but might not be conclusively the ultimate solution to many a bridge player's woes. It's worth reading the Wikipedia entry (https://en.wikipedia...Hand_evaluation) to get a taste of what's currently out there, too. (Anyone using Zar points these days?)

Good luck!

[1eyedjack]

The_Badger, on 2016-July-11, 02:01, said:

c)....was there any factor involved of reassessing hands during the auction?

This I think would be a key component of any successful hand evaluation method, and should be a factor regardless of whether the auction is contested.
There is some indication that the OP has adopted this approach to a degree. You would not, for example, know the extent of your trump fit until later in the auction.
However other adjustments are lacking, such as how you would devalue a holding such as KQ for suit play purposes when partner has shown a shortage, or LHO has bid the suit. Likewise, for NT purposes, a KQ would be an asset opposite a shortage, as providing a guard against the suit running, while being a liability for suit purposes.

Another key component would be to specify what range of values, however adjusted, should be required for a specified target trick level. I did not see any reference to this in the posted doc.
That said, by the time that information is available to make adjustments of this nature I suspect that most experts ditch any formulaic approach and run with their gut.

Another aspect to consider is whether both parties to a partnership make adjustments to their own values for (eg) size of fit or just one party. Maybe just one party knows of the size of fit, and yet the potential of the hand will not be dependent on whether one or both partners is aware.

Actually I think that the method shown has a lot going for it, and mirrors the kind of subconscious thought processes that go through my mind in real life.

[Zelandakh]

1eyedjack, on 2016-July-11, 02:12, said:

However other adjustments are lacking, such as how you would devalue a holding such as KQ for suit play purposes when partner has shown a shortage,

Perhaps I am misreading it but this seems to be included in the Value Duplication section. KQ opposite a "singlet" is listed as -1. I am assuming this is not a reference to fashion trends!

[1eyedjack]

Zelandakh, on 2016-July-11, 08:28, said:

Perhaps I am misreading it but this seems to be included in the Value Duplication section. KQ opposite a "singlet" is listed as -1. I am assuming this is not a reference to fashion trends!

Yup good catch I totally missed that. I take it back, although the difference between NT and suit valuation may remain an issue.

[tnevolin]

Let me thank all contributors! Your opinion really matters.
These docs do not have any explanation in them and are just dry tables for direct usage only. I realize people at this forum are much more interested in "why" and "how". Looks like it would be a logical step for me to write a full scale article not about the method in essence but about how I got to it and why (I think) it is awesome.
That's a really good result of your feedback.
Obviously, I cannot fit whole article in forum format but I will try to address your concerns as much as possible in short replies.

Before I get down to specific question let me assure you that I hate wasting my time inventing something already existing. Therefore, I have searched and read tons of hand evaluation articles. Unfortunately, none of them was on the deep enough level to satisfy me. So I continued working on my own model. I will not discuss each and every model I studied. Most of them have same mistake. They are trying to summarize real playing experience in the head of the author not using the full scale mathematical methods. I saw some publication mentioning computer analysis but they mainly adjusted HCP values and not the whole hand.

My approach is to simply find a bunch of equally important visual parameters and find coefficients for them so that linear sum of two hand values would be as close as possible to average number of tricks. This way I detach the bidding and other factors from strength evaluation. This way bidding sequence becomes irrelevant. You can use this system to evaluate contract level as soon as you agree on denomination. Obviously, before you agree on denomination, evaluation is based on a large variety of unknown hands and no model can give exact answer.

The_Badger, on 2016-July-11, 02:01, said:

a)....of those 400K hands analysed, how many (approx.) are for contested auctions, non-contested auctions?
b)....were the "opponents" on these auctions using the same NHE?
c)....was there any factor involved of reassessing hands during the auction?
d)....etc. etc.

a) No idea and don't care because (see above) my intention was specifically detach hand evaluation from the bidding. So you may think of it as a baseline for all possible bidding scenarios and build on top of it based on current bidding situation.
b) Not familiar with term.
c) Definitely! You can see three sections in the doc. First one is base points, essentially HCP. Two following are adjustment on the contract type upon agreing on denomination.

1eyedjack, on 2016-July-11, 02:12, said:

However other adjustments are lacking, such as how you would devalue a holding such as KQ for suit play purposes when partner has shown a shortage, or LHO has bid the suit. Likewise, for NT purposes, a KQ would be an asset opposite a shortage, as providing a guard against the suit running, while being a liability for suit purposes.

Good observation! In fact, I'll tell you more, there are a lot of adjustments lacking. I analyzed ~100 of different adjustment types (probably all you can think of + many of their combination) and kept only those contributing 1 point or more for the sake of simplicity. Indeed, who needs adjustments worth quarter of a point? So if you don't see it there this is probably why.

1eyedjack, on 2016-July-11, 02:12, said:

Another key component would be to specify what range of values, however adjusted, should be required for a specified target trick level. I did not see any reference to this in the posted doc.
That said, by the time that information is available to make adjustments of this nature I suspect that most experts ditch any formulaic approach and run with their gut.

Don't see how this is an argument against any hand evaluation method whatsoever. By definition it is a tool to find a target trick level. If you at some point already know your level then the hand evaluation is of no use to you, is it? Did I understand your question right?

1eyedjack, on 2016-July-11, 02:12, said:

Another aspect to consider is whether both parties to a partnership make adjustments to their own values for (eg) size of fit or just one party. Maybe just one party knows of the size of fit, and yet the potential of the hand will not be dependent on whether one or both partners is aware.

Actually I think that the method shown has a lot going for it, and mirrors the kind of subconscious thought processes that go through my mind in real life.

I did this intentionally. My model is completely symmetric and is not dependent on exact partner hand knowledge except that you hold ~8+ trumps for trump contract. Could be as low as 6 and as high as 11. Doesn't matter. That's simplicity. I agree with you that introducing non-symmetrical fit component may enhance it even better. That, unfortunately, complicated my calculation tenfold so I decided to leave it out for now. You or me are welcome to enhance it in future.

Concluding the above, I would say that my method is not 100% accurate. It is more accurate than any other I've seen so far, though. Plus it is still relatively easy to remember. A handy bidding tool that what it is. Its greatest benefits is the program. All things inside (the calculation goal, input data, algorithm) are clearly defined. The outcome is proven to be the best fit on a large statistical data. Later on I can add any variable/pattern to it to recalculate the whole table and to verify how significantly this variable impacts contract decision making. I've already researched 100 different pattern combinations. However, there could be those I missed. If you give me a hint at what I should look, I can check it out in a matter of weeks. Voila! And the question whether second fit benefits the no trumps contract is not a mystery anymore.

[akwoo]

A general criticism of your methodology:

You are trying to figure out how to predict the number of tricks your side can take in a contract.

However, the goal of bidding is to get to the optimum scoring contract, and that is NOT always the number of tricks you can take. There is a substantial percentage of hands where the optimal contract is a sacrifice.

In practice, I want my hand evaluation method to help me get to the right contract (while making it hard for opponents to get to theirs), and I don't care about the number of tricks I can take except insofar as it is related to the right contract.

So if you are taking a statistical approach, you should aim for best fit with par contract (which might be larger than the number of tricks you can take), not best fit with number of tricks.

Another way of thinking is this:

It's bad (though perhaps unavoidable) for a hand evaluation system to value the following pair of hands at 12 tricks:

♠AQxxx
♥Kxxx
♦KQx
♣x

♠KJxx
♥Axx
♦Axx
♣KJx

I don't really care though, whether my hand evaluation system values the following pair at 8, 9, 10, or 11 tricks, because being in 5H will probably be a good sacrifice:

♠x
♥AKxxx
♦Axxx
♣xxx

♠Qx
♥xxxxx
♦Kx
♣xxxx

[Stefan_O]

Hi tnevolin,

Interesting initiative!

Some examples would be helpful, though...

Under NT-evaluation, holding:

KTxxx, I get: 3 for the K, +1 for KT, +2 for 5-card suit => Total = 6 points.

KTx, I get: 3 for the K, +1 for KT, +1 for 3-card suit => Total = 5 points.

Since you still only require the usual 25 points for 3NT, this seems too optimistic(?) so I guess I misread your guidelines?

[Stefan_O]

tnevolin, on 2016-July-10, 12:19, said:

I've analyzed 400k deals on the computer to find all the parameters and adjusted values.

Were you comparing you evaluation-method to double-dummy-analysis par-contract?

Or using records from actual tournaments played?

Or how did you do this analysis?

[tnevolin]

akwoo, on 2016-July-11, 21:43, said:

A general criticism of your methodology:

You are trying to figure out how to predict the number of tricks your side can take in a contract.

However, the goal of bidding is to get to the optimum scoring contract, and that is NOT always the number of tricks you can take. There is a substantial percentage of hands where the optimal contract is a sacrifice.

In practice, I want my hand evaluation method to help me get to the right contract (while making it hard for opponents to get to theirs), and I don't care about the number of tricks I can take except insofar as it is related to the right contract.

So if you are taking a statistical approach, you should aim for best fit with par contract (which might be larger than the number of tricks you can take), not best fit with number of tricks.

Thanks for criticism, akwoo. Everything you say is perfectly valid. This applies to all evaluation and bidding system altogether, though. Not just mine. Everybody wants to know perfect sacrifice level just by looking at their hand. Unfortunately, this is not possible because you need to account for opponents strength and distribution as well. That's why other conventions and tools like law of total tricks are there. Perfecting hand strength evaluation never cancels them. Putting your hand on a heart, can you name at least one hand strength evaluation method that organically predicts all types of sacrifices (blocks, block raises, simple competition sacrifice)?

[tnevolin]

Stefan_O, on 2016-July-12, 04:31, said:

Hi tnevolin,

Interesting initiative!

Some examples would be helpful, though...

Under NT-evaluation, holding:

KTxxx, I get: 3 for the K, +1 for KT, +2 for 5-card suit => Total = 6 points.

KTx, I get: 3 for the K, +1 for KT, +1 for 3-card suit => Total = 5 points.

Since you still only require the usual 25 points for 3NT, this seems too optimistic(?) so I guess I misread your guidelines?

Sorry for little confusion. The section "High card combinations" counts points for high card combinations, not for length. So "Three cards", "Four cards", and "Five cards" parameters meant so many high cards out of 5 top in a suit. Next section "Long suit" is about number of cards in a suit. Let me update the document to make it clearer. So in your example it would be:
KTxxx, I get: 3 for the K, +1 for KT => Total = 4 points.
KTx, I get: 3 for the K, +1 for KT => Total = 4 points.
While this would account for:
AKQJTx, A = 4, K = 3, Q = 2, J = 1, five top cards out of five = +2, 6 cards suit = +1 => 13 points

Updated document. Please check if it is clearer now.

[Stefan_O]

tnevolin, on 2016-July-12, 07:02, said:

Updated document. Please check if it is clearer now.

Aah, got it -- yes, much clearer
Thanks!

[tnevolin]

Stefan_O, on 2016-July-12, 04:43, said:

Were you comparing you evaluation-method to double-dummy-analysis par-contract?

Or using records from actual tournaments played?

Or how did you do this analysis?

I've used real tournament boards. That was my initial goal - to get the average on number of tricks taken in real games. That would account for everything. Like strength of play, mistakes, deception, etc. In other words, my method produce a number that is an "average number of tricks other average pairs would get from this hand(s)".

[jogs]

akwoo, on 2016-July-11, 21:43, said:

A general criticism of your methodology:

You are trying to figure out how to predict the number of tricks your side can take in a contract.

However, the goal of bidding is to get to the optimum scoring contract, and that is NOT always the number of tricks you can take. There is a substantial percentage of hands where the optimal contract is a sacrifice.

In practice, I want my hand evaluation method to help me get to the right contract (while making it hard for opponents to get to theirs), and I don't care about the number of tricks I can take except insofar as it is related to the right contract.

That's totally unrealistic. We are decades away from being able to perform that level of analysis.
Find our expected tricks with its variance. Discover how each parameter affects the trick total.
As of today bridge experts don't even agree which parameters are most useful in generating tricks.

[m1cha]

Hi tnevolin,

great work! I only have some suggestions for cosmetic repairs.

First, I suggest you keep the HCP base values the same for all contracts and all suits. To compensate for this, you add an HCP adjustment for trump contracts:
Each honor card (top 4) in a side suit: -1

Mathematically it's the same, obviously, but I find it easier to remember and easier to use particularly if you are not sure which contract you will end up in during the bidding.

Similarly, I suggest you omit that large table "number of cards in trump suit" and replace it with an algorithm:
For each card more than 2 in the trump suit: +1
For each card less than 2 in the trump suit: -1

This will end up in exactly 2 points less than before, so you can change the contract requirements to
4 major: 27/26
5 minor: 30/29
Small slam, T: 33
Grand slam, T: 38
which is in closer agreement with traditional values and may be accepted and remembered more easily.

Actually I would prefer "For each card more/less than 4 in the trump suit" rather than "2" since we are aiming for an 8-card fit, not a 4-card fit. But I don't see a way of improving this without messing up more important things.

By the way, with the "critical contract requirement" I find two values for playing 4 in a major (or 5 in a minor), currently "29/28". Is that for discriminating between MPs and IMPs - because in IMPs you may want to play full games with as little as ~ 40 % success chance depending on vulnerability?

[tnevolin]

Again, thank you all for feedback. I've updated document to make it mode descriptive and converted to PDF for readability. Please have a look.
https://drive.google...RDlySlFJa09RNFU

[tnevolin]

Hi m1cha.
Answering to your suggestions below.

m1cha, on 2016-July-13, 19:01, said:

Hi tnevolin,

great work! I only have some suggestions for cosmetic repairs.

First, I suggest you keep the HCP base values the same for all contracts and all suits. To compensate for this, you add an HCP adjustment for trump contracts:
Each honor card (top 4) in a side suit: -1

Mathematically it's the same, obviously, but I find it easier to remember and easier to use particularly if you are not sure which contract you will end up in during the bidding.

You are very sharp! I actually added this exactly clarification in updated PDF document I just put on google drive. I agree that this wording is easier to remember/use. In fact, I mentally keep exactly same formula you proposed (and not the whole table) in head during the bidding.
This being said, I should mention that this document wasn't designed for learning purposes. It is rather pilot version where I tried to be as plain and clear as possible to avoid confusion. I see that converting it to learning manual format could be beneficial. Will work on that. Thanks for suggestion.
By the way, the same could be applied to NT model high cards and trump model high cards. They do not change their value so these rules can be completely thrown away. I just afraid ton of people would ask the same question in confusion: "What to do with high cards in NT model? Should it change? Your document doesn't say anything."

m1cha, on 2016-July-13, 19:01, said:

Similarly, I suggest you omit that large table "number of cards in trump suit" and replace it with an algorithm:
For each card more than 2 in the trump suit: +1
For each card less than 2 in the trump suit: -1

This will end up in exactly 2 points less than before, so you can change the contract requirements to
4 major: 27/26
5 minor: 30/29
Small slam, T: 33
Grand slam, T: 38
which is in closer agreement with traditional values and may be accepted and remembered more easily.

I agree with you that mathematically that will be the same. However, I also tried to match the exact trick value (1 point = 1/3 trick) so players would get better understanding of their potential not only in critical contracts but at other levels too. Shifting scale would forfeit this benefit. Besides, trick-wise my points are not same value as HCP or total points. So 26 total points are not even close to 26 my points. I assume one who decide to use my method would understand the difference. After all remembering few more numbers for different contract level requirements is a piece of cake comparing to three pages of preceding calculation rules. You play it 20 times and will memorize it easily. Yet, if you think nobody cares about it, then sure - I'll give up precision in favor of convenience.

m1cha, on 2016-July-13, 19:01, said:

By the way, with the "critical contract requirement" I find two values for playing 4 in a major (or 5 in a minor), currently "29/28". Is that for discriminating between MPs and IMPs - because in IMPs you may want to play full games with as little as ~ 40 % success chance depending on vulnerability?

These are for not vulnerable and vulnerable. I've added detailed description in last version.