[lycier]

Hello BBO Gib programmers :

You all are excellent creators who are creating the bridge robot history.Here hats off to you all.

It seems Gibs apply 8421 evaluation approach before Gib version 28, and then it is Total Point evaluation approach from version 28 to now.

Here, may I ask you guys two serious questions?

First question : Why is it very difficult to fix "BUG"? Where is DeBug?
Second question : What's the biggest secret of the bridge skills?

I am a layman on the programming, I think the hand evaluation is just a serious problem.So today my topic is TPs is the source of all evil !!!
That's to say
- DeBug is how to improve the hand evaluation approach effectively.
- The hand evaluation technology is just a biggest secret of the bridge skills.My words imply that it's impossible for you guys to creat expert robot if there is not a set of excellent hand evaluation technology.
IMO,the above are my answers.
Of course, you guys are really experts on the programming and the bridge game,but today I want to express what I see,what I understand. I would better take a example?
Please see this hand which I encountered today at BBO.
- Here is its traveller.





Now you would see there are 15 hands with 4s contract in the 16 hands traveller.My hand is a basic bidding sequence.

---

I strongly think this disaster is caused by rediculous TPs evaluationinstead of a unknown Bug.
At the initial stage of bidding, experts usually add length points instead of short suit points.
4♠ says " 7+hcp,strong rebiddable ♠,8-10TPs".
Obviously its description is a kind of evil. As we know that when with misfit between two hands, short-suit points are useless, on the contrary, short suit and voidness have led to a great depreciation, also is a important feature of reducing orginal value.
In a word, only when with suit fits, adding short-suit points are wise.


Every programmers, let's making a logic analysis at below.
Would you think that compared to 8421 evaluation, Total Point is a better improvement? If yes, why cannot continue to obtain its improvement? Why is it very difficult to reach expert's skill? Where is the root of the problem?
Here I have to say the hand evaluation approach is one of core issues.
Stephen Tu had ever said that " Gib is a kind of rigid...", his words just like a beacon, that's to say :
1- Gib is a kind of faithful mode. This is one of the biggest advantages of Gib.
2- Gib is a kind of rigidness mode.It's a devastating disadvantage of Gib.

IMO, I think the essence of bidding sequences are
1- Bidders are telling story.
2- Bidders are describing own hand dynamically instead of statically or rigidly.
3- A story about constantly correcting own evaluating for many times.

Now it would come out a serious issue.
If Gib can't make second evaluating,even third and more, this way will not be the most objective evaluation of the real.I meant that if you guys have great ability to handle with this key technology so well, other issues may be become easier to resolve.
Please think over.
As long as any people find a part of evaluation approach, it turns out it's sure to improve the skills greatly, so why not for Gib?

Best Wishes


Lycier

[hrothgar]

If I had my druthers and could get the GIB team to work on anything I wanted, improving bidding with 7-5 hands would not top the list...

[manudude03]

I must confess, I would have just bid 4S over 1H, but I do not like East's bidding on this hand at all. GIB does still use 8421 points on cuebidding, it is just effectively reworded to stuff like ♠A (8+ 8421 points), or ♣KQ (6+ 8421 points).

[Stephen Tu]

lycier, on 2016-May-17, 07:31, said:

Hello BBO Gib programmers :

As you may know, I am not a Gib programmer, having never worked for BBO. But I am a computer programmer, a decent bridge player, and have some knowledge of how GIB operates from following Gib discussion forum that Ginsberg the original creator posted in 15+ years ago, so I answer from that perspective.

Quote

It seems Gibs apply 8421 evaluation approach before Gib version 28, and then it is Total Point evaluation approach from version 28 to now.

No, it's always been total points. 8421 is only used in certain sequences mainly cue bidding sequences for specific honor identification.

Quote

First question : Why is it very difficult to fix "BUG"? Where is DeBug?
Second question : What's the biggest secret of the bridge skills?

The main reasons I think it is hard to improve Gib is that the sheer number of possible bidding sequences is impossibly huge, as is the number of possible hand types to deal with. It's hard to have enough rules to cover everything. It's easier for humans to extrapolate common sense to deal with new sequences; computers generally have to have everything written down explicitly.

Quote

I am a layman on the programming, I think the hand evaluation is just a serious problem.So today my topic is TPs is the source of all evil !!!

I don't think TP is a problem. It's good to take distribution into account as it contributes to trick taking power. Just in some situations it should have HCP minimums as well, and it would be nice if it could dynamically evaluate the hand a bit more as the bidding progresses (demote shortness in partner's suits, badly placed honors in opponent's suits, promote honors in partner's suits), which it doesn't really do now.

As for the example hand you gave:
- There should maybe be a hcp minimum for double then bid a suit, so it doesn't count so much on distribution to choose double over 1d overcall
- the bid suit after double to show good one suited hand should maybe be stronger, 19+ TP instead of 17+TP
- I don't see why the doubler has to jump to 3d, perhaps 2d is plenty and gives more room for advancer to bid spades again at a lower level and the misfit can be found lower.
- rebidding diamonds should deny 4 spades, jumping to 3 should probably deny 3 spades also
- don't see why it has to rebid 4s in preference to 3s
- the definition of 4s bid is probably OK. But just because a hand matches a bid definition doesn't mean that GIB should be forced to bid it, here a simulation could possibly capture that it's a bad idea to bid on ten high opposite possible stiff/doubleton, and that 4s is unlikely to make with so many clubs to get rid of.

Dynamic hand evaluation definitely would be an improvement. But using total points is OK, and a lot can be done just tweaking existing rules.

Quote

Stephen Tu had ever said that " Gib is a kind of rigid...", his words just like a beacon, that's to say :
1- Gib is a kind of faithful mode. This is one of the biggest advantages of Gib.
2- Gib is a kind of rigidness mode.It's a devastating disadvantage of Gib.

Please stop repeating this statement. I only said that Gib was rigid in not opening 1nt on semi-balanced hands. You are applying my statement to all kind of things where it really doesn't apply, I never meant it that way.

[lycier]

I know the above are decent expert players.
Thanks to everyone.

Stephen Tu, on 2016-May-17, 11:23, said:

I don't think TP is a problem. It's good to take distribution into account as it contributes to trick taking power. Just in some situations it should have HCP minimums as well, and it would be nice if it could dynamically evaluate the hand a bit more as the bidding progresses (demote shortness in partner's suits, badly placed honors in opponent's suits, promote honors in partner's suits), which it doesn't really do now.

Well and good.

Please see my hand, that is a strong evidence.

---

As you earlier said it should take distribution points into account.
From a modern perspective,some of your ideas may not be accurate and correct.

Let's make a analysis one by one.

1- How many TPs are there in the east hand? How many working points? How many value points?
My answer are :
- TPs=18
- Working points=13hcp (singleton Q is waste point for time being)
- Value points:
Generally accepted expert method is Only adding length points instead of short-suit points.
So its value points= 15cp-1(singleton)+3 length points=17
So this hand is not so strong enough to double first then bid its long suit to show its big hand.However in fact east has less than 17 TPs, such action is rediculous with no doubt.

The following disaster comes here immediately !
According to your TPs theory,even face to voidness in partner's decent suit, west think there is 8TPs in the hand, this is a classic misrepresenting descriptions !!!
We know that in fact only when with suit fit, voidness points are useful - that is a kind of dummy distribution, its values equal fitting-suit length in the dummy.If its length is zero, its value of voidness is zero.
Would you tell me it is not correct for what I said?

2- Please you would answer a very serious question.
How do the tricks come about?
- Would you think the tricks come from HCP? TPs? Working points? Value points?

The world experts think it is just suits fit to come about tricks, never be any kind of points.
Voidness in ♦ in the west hand,would you think it has ability to come about tricks?
Never ! Never and Never !
Now you see Gib is very very rediculous to take its voidness points into account even without fits in the hand,this is a greatly misrepresenting description.

So I would have to say Total Points evaluation approach is the source of all evil.

Any idea?
If you disagree with my opinions, would you offer your hand evidence?

[Stephen Tu]

lycier, on 2016-May-17, 15:46, said:

1- How many TPs are there in the east hand? How many working points? How many value points?
My answer are :
- TPs=18
- Working points=13hcp (singleton Q is waste point for time being)
- Value points:
Generally accepted expert method is Only adding length points instead of short-suit points.
So its value points= 15cp-1(singleton)+3 length points=17
So this hand is not so strong enough to double first then bid its long suit to show its big hand.However in fact east has less than 17 TPs, such action is rediculous with no doubt.

I agree that East should overcall 1♦ instead of doubling. But I disagree that "TP is the source of all evil". You don't have to abandon TP evaluation to make East overcall instead of doubling. Just raise the threshold, and install a HCP minimum. GIB apparently thinks 17TP is enough to double and bid a suit when short in a major. So just raise the TP minimum to say 19 or so, and also make say 17HCP minimum to trigger it to double first. That would force East to overcall instead. Make the 1d TP range wider.

I really don't understand your argument here that it's TP that's the problem.

Quote

According to your TPs theory,even face to voidness in partner's decent suit, west think there is 8TPs in the hand, this is a classic misrepresenting descriptions !!!

- I agree that west should devalue its hand as the auction progresses and no spade fit.
- but I disagree that TP is the root of the problem and that TP must be abandoned. West thinks it has 8 TP (which is normally justifiable given the extreme shape), and this matches its rule for bidding 4s. But just because a rule happens to be matched doesn't mean that it HAS TO BE be selected. And here it doesn't even fully match the rule, since the rule says "strong rebiddable spades", and I'd hope maybe 7 cds to the t9 doesn't count as strong rebiddable? This is a situation where simulations come to bear. West should deal out possible hands for East, with 2- spades, and maybe conclude that the weak trumps will doom 4s, and not bid it. Even though the TP total suggests that it should bid game, the simulation should reveal that lack of fit will likely cause game to fail, and thus overrule the general idea of bidding when TP suggests game. The simulation should reveal the wastage of values due to misfit.

Also if neither west or east jump the bidding, perhaps it is easier to stop lower. I see no reason why either has to jump.

Quote

2- Please you would answer a very serious question.
How do the tricks come about?
- Would you think the tricks come from HCP? TPs? Working points? Value points?

Now you see Gib is very very rediculous to take its voidness points into account even without fits in the hand,this is a greatly misrepresenting description.

Gib happens to count shortness instead of length. It really makes very little difference, because shortness somewhere creates length elsewhere. So what if Gib counts void as points? Counting shortness points, Gib counts west as 8 TP? Well if we force it to count length instead, then make it count 3 points for the 7 card suit and 1 pt for the 5 cd suit, we still get 7 points, not much difference! Often either method works out the same as the initial valuation. The trick is to use the auction to dynamically adjust for fit or lack of fit. It would be best if they dynamically adjusted point total, but that's not implemented. So simulation has to cover these gaps.

The problem is that choosing to bid game is supposed to take into account the previous auction and not just blind rule match. So it may be simply an issue of allowing/forcing simulation at the point where west decides to bid 4s. Basic bots I think have simulations disabled in a lot of these types of situations, so a strictly rule based system may cause it to do worse than advanced bots.

And keep the auction lower so west can show additional spades at a lower level without forcing to game, so it can reach decent games when East has like 3 spades or something useful like AK/AQ doubleton.

TP really isn't the issue IMO. Tweak the rules, use simulation, this hand can be fixed.

East can be forced to bid 1d. And if East doubles first, on a stronger hand, west can bid differently. Perhaps west should bid 3s as a preempt over 1h, showing this sort of hand in one shot. If starting with 1s, it can use simulations to better judge whether or not to bid spades again later, and at what level.

[lycier]

Very very difficult to understand why you wanted to deliberately conceal the weakness of TPs.
In fact, I have found that most of the reporting of supicious Gib issue on this forum are related to TPs - the source of all evil !!!
You of course know HCP for notrumph,TPs only for suit contract, this really needs Gibs have ability to switch freely. In fact, rigid Gib mode have wrongly taken TPs as a substitute of HCP. Obviously, this is not correct hand evaluating style.

Isn't true for what I said?
Who moved your cheese?

[Stephen Tu]

lycier, on 2016-May-17, 18:37, said:

Very very difficult to understand why you wanted to deliberately conceal the weakness of TPs.
In fact, I have found that most of the reporting of supicious Gib issue on this forum are related to TPs - the source of all evil !!!
You of course know HCP for notrumph,TPs only for suit contract, this really needs Gibs have ability to switch freely. In fact, rigid Gib mode have wrongly taken TPs as a substitute of HCP. Obviously, this is not correct hand evaluating style.

Isn't true for what I said?
Who moved your cheese?

I don't think you have demonstrated anything concrete. Distributional strength needs to be taken account somehow. Even in NT, there is some basis for using something other than just HCP with no adjustments for shape, because something like xx xx xxx AKxxxx opposite a 1nt opener is obviously likely to take more tricks than xxx xxx xxxx AKx. Otherwise what, you are going to count both as 7 points only and pass both hands?

What is your alternative to TP? How would you like GIB to calculate what to do. Basically all you have done is declare TP as evil based on your say so. You haven't demonstrated why it's really unworkable, and exactly how you would like GIB to operate instead.

So basically, I am saying what you have said doesn't really prove anything.

TP is a reasonable way to evaluate. I might quibble with the exact formula and personally evaluate certain hands a point or two more or less. But I don't think it's totally unworkable as you seem to be saying. Just some actions probably need a HCP floor as well to avoid GIB trying NT relying too much on distributional values only, or making understrength double vs. overcall decisions (or 2c openers). And the ranges for some actions need to be adjusted higher. Then it would bid this hand more reasonably.

You avoid addressing any of my arguments directly. For example, I said it can easily make East overcall 1d instead of double by make minimum TP for double with this shape higher, along with HCP minimum. What is your objection to that? Am I wrong about this, why do you think it is so?

How would you have it make the decision to overcall instead of double, what is your formula? And why would your proposed formula work substantially better than simply adjusting the point ranges for the existing TP formula?

Basically you should describe your preferred point counting formula, instead of just saying TP is garbage. The computer has to count points somehow.

And no matter what formula you use, when the hands fit badly the computer has to account for it somehow, there will be duplication of values. This is where simulations should kick in. So if you have 17 "schmoints" (whatever point counting technique you use) opposite 8 "schmoints", the computer shouldn't be thinking "I have 25 schmoints combined, therefore I must bid game". At that point it has to start dealing hands for partner consistent with the ranges and distributions for prior bids, and calculating if game makes often enough or not. Whether the partnership is double counting schmoints that don't work together. And if it doesn't, then pass or pull back a level and not bid the game.

If the early auction is managed better with reasonable ranges, it should be able to get these decisions right most of the time, with any semi-reasonable schmoint counting system. Including its existing system. TP isn't the root of all evil. Bad ranges in the definitions, bad priorities for choosing between bids, not activating simulations at the appropriate points are where it is going wrong on your example hand.

[lycier]

Excuses and denial are futile, you've done your best. Let BBO decide eventually.
Actually this type of thread is easy to offend people. Now you see everyone only watch,don't reply, so let's forget. And many thanks to you for your good reply.

In my eyes, skills especially hand evaluation approach are the only tools to express developers' mind.
I know the robot bridge history will prove my point of view is correct.

[Stephen Tu]

lycier, on 2016-May-18, 04:34, said:

Excuses and denial are futile, you've done your best. Let BBO decide eventually.
Actually this type of thread is easy to offend people. Now you see everyone only watch,don't reply, so let's forget. And many thanks to you for your good reply.

In my eyes, skills especially hand evaluation approach are the only tools to express developers' mind.
I know the robot bridge history will prove my point of view is correct.

Man you are ridiculous. You completely dodge all of my questions!

You didn't even present a point of view! All you say is "TP is evil". You have to provide your alternative! Otherwise how is GIB supposed to calculate how strong its hands are and how high to bid?

TP is bad. OK if we accept that, replace it with what exactly? You have to describe what you are after, what you want to change it to. Otherwise your complaint is like totally nonsense.

[lycier]

Stephen Tu, on 2016-May-18, 09:22, said:

Man you are ridiculous. You completely dodge all of my questions!

You didn't even present a point of view! All you say is "TP is evil". You have to provide your alternative! Otherwise how is GIB supposed to calculate how strong its hands are and how high to bid?

TP is bad. OK if we accept that, replace it with what exactly? You have to describe what you are after, what you want to change it to. Otherwise your complaint is like totally nonsense.

I have shown my point of view in the "OP", and my hand is a best evidence to show your TPs is evil and very rediculous, but you are making many futile excuse and denial. TPs replace HCP for notrumph and suit contract, whether it is with/without suits fit, very very rediculous, what you comment are like totally nonsense, like totally true lies.

[manudude03]

lycier, on 2016-May-18, 09:53, said:

I have shown my point of view in the "OP", and my hand is a best evidence to show your TPs is evil and very rediculous, but you are making many futile excuse and denial. TPs replace HCP for notrumph and suit contract, whether it is with/without suits fit, very very rediculous, what you comment are like totally nonsense, like totally true lies.

What Stephen has been trying to say is if you think TP is bad, what do you replace it with? You can't just rip it out or you will have no way to distinguish between these 4 hands:

AKxx AKx AKx xxx
Ax Ax Ax AKQxxxx
xx xx xx AKQxxxx
Axxx xxxx xxx KQ

[Stephen Tu]

lycier, on 2016-May-18, 09:53, said:

I have shown my point of view in the "OP"

Your OP is very hard to understand. I know English is not your native language, but if you want to convince people your view is correct, you have to try again to make your point using different words so that people can understand exactly what you are trying to advocate for, instead of only be guessing what you mean, and try to directly address statements and questions from other posters. Just repeating "TP is evil", and saying that "you are making excuses", is completely worthless and doesn't convince anyone of anything. I asked many questions about specific things in the auction. You answered none of them. I made statements about specific things that could have been altered in the auction to have GIB make better choices. You completely ignored them.

To be an effective debater, you need to listen to your adversary, and directly rebut his individual statements, demonstrate why a particular statement is not true. You can't just say "it's all excuses and denial" and be convincing. Take an exact quote from one of my posts, and state exactly what you think is wrong with it. Don't make blanket condemnations of everything without engaging specific statements. That's not constructive debate.

If you are saying that GIB could be improved with more dynamic reevaluation over multiple rounds of bidding, changing its point count based on the previous auction, then I agree with you. If you are saying GIB should count HCP only, and ignore distribution, then I don't. When you say TP is the root of all evil, I disagree. Evaluation needs a starting point. TP isn't exactly the formula I would start with, but it's not so far off that it's totally unworkable and has to be thrown out, perhaps it could be tweaked. There is no perfect point counting scheme. And as the auction progresses, any starting point is going to have issues with duplication of values or super-fitting values. The point count requirements for game, slam, etc. all assume an average amount of duplication of values. We have all seen games and slams make with many fewer HCP than typically required because the hands mesh well. As well as seen contracts with lots of combined points fail because the hands fit poorly. That's where simulation is supposed to kick in, it is exactly as human experts do in later rounds of the auction, visualize probable hands for partner and estimate how many tricks are taken. Then adjust how much to bid from there. But throwing out distribution in favor of HCP only isn't the answer. Otherwise you are saying I should pass say 11 HCP 5-5-x-x because I should treat it same as 11 HCP 4333 because after all I don't know whether I have a fit or not when I open the bidding. Initial bids often have to be made on a TP basis because it is still unknown how well or not the hands fit. If you were to bid on HCP only, it would lead to far too conservative bidding, you'd miss too many good games. It's after several rounds of bidding that one can pull back because of misfitting values. So that's why it's important to have good definitions on the initial rounds of bidding, so there is room to find if misfit or not. The reason GIB gets into trouble on your example hand is because of the initial double vs. overcall. Which is not forced by using total points, but rather because the threshold for doubling first isn't set high enough.

Ignore the East hand, pretend that it is strong enough for off-shape double like say it has also DA. Facing 1c-x-1h-?, should west bid spades or pass? If we are to take your argument (at least how I am guessing it is), the hand should be only 3 hcp, and should not bid spades, because normally 3 points is not enough to bid freely. So you want to bid 7105 the same as 4333 because of the possibility partner made an offshape takeout double and doesn't have spade length? That's insane IMO.

Quote

notrumph and suit contract, whether it is with/without suits fit, very very rediculous, what you comment are like totally nonsense, like totally true lies.

English tips:
-there is no "e" in "ridiculous", nor is there an "h" at the end of "notrump"
-I've noticed you use the phrase "true lies" often. It's not really a phrase that means anything or is used in English. It's the title of a James Cameron movie starring Arnold Schwarzenegger as a spy fighting terrorists, but it doesn't really have any meaning in English and is not ever really used ordinarily. The two words are directly contradictory, so the phrase doesn't make any sense.

[Vampyr]

manudude03, on 2016-May-18, 10:00, said:

What Stephen has been trying to say is if you think TP is bad, what do you replace it with? You can't just rip it out or you will have no way to distinguish between these 4 hands:

Yes, I guess Total Points is the only way the computer can account for distribution, when a human player does not really add points as such, so frequent inaccuracy is inevitable. But if the Total Points do not recalculate during the auction, they are a pretty big problem.

What I have seen in some examples is that the Total Points do recalculate, but rather than being an adjustment of the previous evaluation, they are entirely different (ie the new explanation precludes the one given the round before). This can be easily mended, I think. Also I saw one that gave a player 15 cards. This should be trivial.

[Vampyr]

Stephen Tu, on 2016-May-18, 11:55, said:

English tips:
-there is no "e" in "ridiculous", nor is there an "h" at the end of "notrump"
-I've noticed you use the phrase "true lies" often. It's not really a phrase that means anything or is used in English. It's the title of a James Cameron movie starring Arnold Schwarzenegger as a spy fighting terrorists, but it doesn't really have any meaning in English and is not ever really used ordinarily. The two words are directly contradictory, so the phrase doesn't make any sense.

Super film though.

[diana_eva]

My favorite lycier quote so far is "who moved your cheese"

Lycier, cool off please. Read carefully what people say, and reply to that. Like Stephen says, statements like "TPs is evil and very rediculous, but you are making many futile excuse and denial." are not helpful. I don't think you realize just how rude your replies sound. Please be polite, of we'll have to take you off the GIB Forums if it tilts you so badly to discuss robots.

[lycier]

Thank you for your reminding.

[Bbradley62]

Damn! Still can't upvote the mods...

[lycier]

Maybe my wording on the title is extreme inappropriate, so I should say sorry at first.
Today I occasionally saw a classic hand about a common problem of invalid voidness.

---

Result : 4♥E-2

After 2♥, it is no problem for west Gib to bid up to 4♥ with 11-12TPs in fact.
But a serious problem is west Gib really holds valid voidness points? Does this voidness cause appreciation or depreciation ?
NO,NO,No.
I 100% believe your programmers should know this voidness points is invalid for sure, East-West only have 16 working points,♠Ace is wasteful.
However Gibs ignore this negative factor, it turns out it is very very easy to obtain big wrong TPs, such approach is unreasonable.
Did I say the wrong？

It is appreciated if would point out my wrong.

[Stephen Tu]

Well, we don't really know why West bids 4h rather than 3h. Really this is a simulation matter, see how many game you make/miss by bidding 4h instead of 3. I don't think this is a situation where Gib is or should be trying to tally total points and deciding on that basis how much to bid, I think 3h and 4h show only slightly different ranges, and it should use sim decide which successful more often with a given hand.

See this is a tough sequence for raise to 3h. Unlike 1h-1nt-2c, you don't have an impossible 2s bid to distinguish between courtesy raises and really strong invites. So your raise has to cover a lot of ground since partner might have 16-18 unbal. And because of this wide range of your bid and his 2h bid, it hard for partner to say accept with 14 just because he has 14 instead of 12, because your raise might be catering him to have 16. So there is supposed to be a tendency to overbid a little with the strong end of invites, realizing this will sometimes go down when partner is really min or the hands don't fit well, in order to get to the good games partner might not raise 3h to 4h with. It just a tough sequence for natural system, some hands just going to get wrong by being too high or too low because range too wide, not enough bids. That why people play strong club or things like Gazzilli, though all system have trade-offs.


Also, you don't really know whether or not the void is bad or not. After all maybe partner only has J high spades and the k of diamonds. Or maybe partner is 5-5 instead of 6-4 with these same honors and it makes, and wouldn't have raised 3h to 4. And maybe this is IMPs, where on some hands both 4h and 3h are down, which favors bidding game more often since the extra undertrick is small relative to game bonus. Especially vul.

So I really don't see why you put the blame on TP for this particular hand. It not clear that TP is the cause for GIB to bid 4 rather than 3, it might be sim support that action. And I'm really not sure that 4h is that bad a bid.