BBO Discussion Forums: Freaky strong - BBO Discussion Forums

Jump to content

  • 2 Pages +
  • 1
  • 2
  • You cannot start a new topic
  • You cannot reply to this topic

Freaky strong

#1 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2011-November-06, 07:58

Yes, "twice rebiddable spades; 16+ total points" is accurate, but...

1

#2 User is offline   vang 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 278
  • Joined: 2004-February-18
  • Gender:Male
  • Location:Romania
  • Interests:Linux

Posted 2011-November-06, 14:06

better than to play 3HX i'd say ;-)
0

#3 User is offline   BunnyGo 

  • Lamentable Bunny
  • PipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 1,505
  • Joined: 2008-March-01
  • Gender:Male
  • Location:Portland, ME

Posted 2011-November-06, 14:14

View Postvang, on 2011-November-06, 14:06, said:

better than to play 3HX i'd say ;-)


Really? I figure you can get 3HX down 3 by pumping declarer.
Bridge Personality: 44 44 43 34

Never tell the same lie twice. - Elim Garek on the real moral of "The boy who cried wolf"
0

#4 User is offline   manudude03 

  • - - A AKQJT9876543
  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 2,610
  • Joined: 2007-October-02
  • Gender:Male

Posted 2011-November-06, 14:35

View PostBunnyGo, on 2011-November-06, 14:14, said:

Really? I figure you can get 3HX down 3 by pumping declarer.

510 beats 500 :)
Wayne Somerville
0

#5 User is offline   calm01 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 137
  • Joined: 2010-July-08

Posted 2011-November-06, 16:51

An error of hand evaluation of one trick between partners could be put down to a matter of opinion.

A three trick error suggests that hand evaluation has just one wheel left with the stagecoach missing.

Total points can be a poor evaluation method in the hands of a reasonable human player and perhaps readily becomes almost meaningless in a computer program such as GIB. Most beginner (human) players would realise that AKQJx has better length points than 98653 but does GIB? If you ever experienced unilateral biddng when playing with a beginner, you can relate to this.

One definition (I feel sure it is not the only one) for total points that can be found on the internet is:

"Total Points

The sum of high card points (HCP) and one of the shape points (long suit or short suit point) is called total points (TP).
When you are opening a bid or bidding a new suit you must evaluate your hand with HCP & long suit points. So,

TOTAL POINTS= HIGHCARD POINTS + LONG SUIT POINTS"

This definition is incomplete (how are long suit or short suit points counted?) and is not self-consistent as short suit points suddenly get overlooked at the conclusion.

Almost any recognised method will be better than the existing GIBs use of total points. If you doubt this just look at when GIB raises a major from 4 to the 5 level and (ignore its description) its variations for the same total points (zero to 2 extra tricks range compared to prior promises and holding between 0 to 4 key cards).

Why not abandon total points which GIB cannot seem to handle effectively and begin to rely on a more accurate and self-consistent technique of hand evaluation. There are so many better techniques to select from such as:

high card points especially for opening, simple responses and NT contracts,
fit/misfit adjustments,
consideration of likely blockages or good communications between a possible dummy and declarer,
playing strength adjustments,
entry considerations on weak hands with one very long weak suit,
losing trick count for known 8+ card fits where a suit contract is anticipated,
adjustment for hands holding 3 or 4 aces,
...

We will all use one more and have our own favourites.

Simply give GIB one 'favourite' and apply it sensibly and consistently in the framework of prior promised values, steadily making improvements where appropriate, and slowly, GIB will bid better and better. Only then add a secondary evaluation string to its bow.

But please, if you wish GIB to ever bid well, abandon its misuse of total points.

Calm01
0

#6 User is offline   cloa513 

  • PipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 1,528
  • Joined: 2008-December-02

Posted 2011-November-06, 18:05

View Postcalm01, on 2011-November-06, 16:51, said:

An error of hand evaluation of one trick between partners could be put down to a matter of opinion.

A three trick error suggests that hand evaluation has just one wheel left with the stagecoach missing.

Total points can be a poor evaluation method in the hands of a reasonable human player and perhaps readily becomes almost meaningless in a computer program such as GIB. Most beginner (human) players would realise that AKQJx has better length points than 98653 but does GIB? If you ever experienced unilateral biddng when playing with a beginner, you can relate to this.

One definition (I feel sure it is not the only one) for total points that can be found on the internet is:

"Total Points

The sum of high card points (HCP) and one of the shape points (long suit or short suit point) is called total points (TP).
When you are opening a bid or bidding a new suit you must evaluate your hand with HCP & long suit points. So,

TOTAL POINTS= HIGHCARD POINTS + LONG SUIT POINTS"

This definition is incomplete (how are long suit or short suit points counted?) and is not self-consistent as short suit points suddenly get overlooked at the conclusion.

Almost any recognised method will be better than the existing GIBs use of total points. If you doubt this just look at when GIB raises a major from 4 to the 5 level and (ignore its description) its variations for the same total points (zero to 2 extra tricks range compared to prior promises and holding between 0 to 4 key cards).

Why not abandon total points which GIB cannot seem to handle effectively and begin to rely on a more accurate and self-consistent technique of hand evaluation. There are so many better techniques to select from such as:

high card points especially for opening, simple responses and NT contracts,
fit/misfit adjustments,
consideration of likely blockages or good communications between a possible dummy and declarer,
playing strength adjustments,
entry considerations on weak hands with one very long weak suit,
losing trick count for known 8+ card fits where a suit contract is anticipated,
adjustment for hands holding 3 or 4 aces,
...

We will all use one more and have our own favourites.

Simply give GIB one 'favourite' and apply it sensibly and consistently in the framework of prior promised values, steadily making improvements where appropriate, and slowly, GIB will bid better and better. Only then add a secondary evaluation string to its bow.

But please, if you wish GIB to ever bid well, abandon its misuse of total points.

Calm01



This is not a misuse of total points error- this is an incomplete bidding table error whereby there is no other bid than 4 over a preempt. Its clearly a 7 bid.

By the GIB's total points= HCP+length+shortness
It also uses HCP.

Its favourite is total points and also uses HCP.
0

#7 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,398
  • Joined: 2004-August-21
  • Gender:Male

Posted 2011-November-06, 23:26

What's actually happening is that there's a rule for hands that have enough to bid game, but we're not already in a game forcing auction. The rule also checks whether our points plus the minimum partner has shown is less than 30. If partner had shown at least 4 points, we would have skipped this rule and fallen into a rule that makes a slam invitation.

At some point, this rule was changed to disallow simulations. With simulations, it probably would have found enough hands for partner that make slam, so it would have bid stronger. This is the equivalent of the human thinking "My hand is so good, I hardly need anything from partner to make slam." But I think it was causing GIB to overbid on some hands that weren't quite as freaky as this one, so simulation was disabled here.

cloa513 is probably right, we just need to add a higher priority rule that matches super-strong hands like this. However, I also think it's low priority, since freaks come up so rarely. It's more important to fix the common auctions - how often do you get zero-loser hands?

#8 User is offline   georgi 

  • PipPipPipPipPipPip
  • Group: Yellows
  • Posts: 1,317
  • Joined: 2007-December-30
  • Gender:Male
  • Location:Bulgaria

Posted 2011-November-07, 04:54

View PostBbradley62, on 2011-November-06, 07:58, said:

Yes, "twice rebiddable spades; 16+ total points" is accurate, but...



For such hands GIB has on disposal cuebiding the preemptive openings which is strong two-suits X-losers.

Like 3-4 - 4-losers, 4-5 -> 3 - losers. etc.

This could be improved as obviously GIB didn't find it, but has bidded conservative 4.

3 - 5 has not the meaning of 3- losers. So that could be one step of improving.

#9 User is offline   BunnyGo 

  • Lamentable Bunny
  • PipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 1,505
  • Joined: 2008-March-01
  • Gender:Male
  • Location:Portland, ME

Posted 2011-November-07, 05:05

View Postmanudude03, on 2011-November-06, 14:35, said:

510 beats 500 :)


Ah right...all 13 tricks is more than 500.
Bridge Personality: 44 44 43 34

Never tell the same lie twice. - Elim Garek on the real moral of "The boy who cried wolf"
0

#10 User is offline   xxhong 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 328
  • Joined: 2010-November-11

Posted 2011-November-07, 12:43

This is still a evaluation problem. Gib has no method to count total tricks, which is being done by human in every hand.

For example, with AKQJxx, you should count it for 6 tricks in most situations. AKQ is 3 tricks. A is one trick. Kx is half trick, QJT is one trick. AQ is 1.5 trick. This kind of counting would allow gib to make sensible decisions when holding very strong hands or making slam decisions.

This is also why gib is so bad after 5NT King asking, because gib only relies on simulations, but not on trick taking. For grand slams, the basic rule for human to bid 7 is when you can count 13 tricks (or you are likely to get 13 tricks).

View Postbarmar, on 2011-November-06, 23:26, said:

What's actually happening is that there's a rule for hands that have enough to bid game, but we're not already in a game forcing auction. The rule also checks whether our points plus the minimum partner has shown is less than 30. If partner had shown at least 4 points, we would have skipped this rule and fallen into a rule that makes a slam invitation.

At some point, this rule was changed to disallow simulations. With simulations, it probably would have found enough hands for partner that make slam, so it would have bid stronger. This is the equivalent of the human thinking "My hand is so good, I hardly need anything from partner to make slam." But I think it was causing GIB to overbid on some hands that weren't quite as freaky as this one, so simulation was disabled here.

cloa513 is probably right, we just need to add a higher priority rule that matches super-strong hands like this. However, I also think it's low priority, since freaks come up so rarely. It's more important to fix the common auctions - how often do you get zero-loser hands?

0

#11 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,398
  • Joined: 2004-August-21
  • Gender:Male

Posted 2011-November-07, 16:06

View Postxxhong, on 2011-November-07, 12:43, said:

This is still a evaluation problem. Gib has no method to count total tricks, which is being done by human in every hand.

For example, with AKQJxx, you should count it for 6 tricks in most situations. AKQ is 3 tricks. A is one trick. Kx is half trick, QJT is one trick. AQ is 1.5 trick. This kind of counting would allow gib to make sensible decisions when holding very strong hands or making slam decisions.

GIB does have the notion of suit quality; these are represented in its descriptions as "biddable", "rebiddable", "twice rebiddable"; it has additional categories that don't have names (for hands like AKQxxxx) that can be used in bidding rules. But it's hard to write bidding rules that cover all these cases, especially for rare cases like this one.

The theory is that simulations should come to the same conclusion. If AKQJxx usually takes 6 tricks, this should be reflected in most of the simulation results.

#12 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2011-November-07, 16:25

View Postbarmar, on 2011-November-07, 16:06, said:

GIB does have the notion of suit quality; these are represented in its descriptons as "biddable", "rebiddable", "twice rebiddable"; it has additional categories that don't have names (for hands like AKQxxxx) that can be used in bidding rules. But it's hard to write bidding rules that cover all these cases, especially for rare cases like this one.

I'm pretty sure I've seen suits described as "solid" as well. Any chance of getting a description of what GIB means by each of those four designations?
0

#13 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,398
  • Joined: 2004-August-21
  • Gender:Male

Posted 2011-November-07, 17:12

The categories are:

0-2 cards
3-card
4-card = 4 with poor honors
biddable = 5 cards, or 4 with 4+ HCP or QJTx
rebiddable = biddable + another card
twice rebiddable = rebiddable + another card
strong rebiddable = twice rebiddable with 8+ HCP
solid 6-card = AKQTxx or better
solid 7-card = AKQxxxx or better
solid 8-card = AKJxxxxx or AKxxxxxxx (there are 7 x's)

#14 User is offline   cloa513 

  • PipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 1,528
  • Joined: 2008-December-02

Posted 2011-November-07, 23:08

Why not if the simulation says it should get 12 tricks then bid small slam, if GIB gets 13 tricks bid the grand slam- how difficult is that?
0

#15 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,398
  • Joined: 2004-August-21
  • Gender:Male

Posted 2011-November-07, 23:47

As I mentioned earlier, the rule that generates this 4 bid was changed a while ago to prohibit simulations.

#16 User is offline   calm01 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 137
  • Joined: 2010-July-08

Posted 2011-November-08, 08:13

cloa513 said:

"Its favourite is total points and also uses HCP."

If true, there is little hope for GIB's bidding being substantially improved until its favourite is changed to a better system. Any programming effort would be tinkering with a poor evaluation system until total points is dropped.

My experience teaching beginners is that they soon grasp and love HCP and get so attached it is difficult to get them to downgrade Quacky hands (Queen and Jacks with not Aces and Kings).

Onece HCP is understood and employed sensibly, losing trick count seems the simplest approach that leads to the fewest poor evaluation errors by beginners. Yes I know adjustment for no Aces or 3/4 Aces is ideally required but this is often too sophisticated for most beginners and only confuses them.

Please, unless you want to continue to waste much of the BBOL programming effort on GIB, replace its favourite total points system with another evaluation system.

Note that even a simplistic losing trick count would give (on the assumption of an average 2 spades in partners hand) a losing trick count of zero and 7 spades would be easy to find even by a software program.

Please stop flogging a dead horse (total points). Yes, I, know admit that BBOL would have to admit it has been wrong all these years, but the alternative of an expensive to maintain GIB (in terms of return on investment) is the alternate approach.

Losing trick count is not for Christmas (freaky hands) - it is an everyday tool for 8 card fit hands that works here too.

Calm01
0

#17 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,398
  • Joined: 2004-August-21
  • Gender:Male

Posted 2011-November-08, 12:02

View Postcalm01, on 2011-November-08, 08:13, said:

Please, unless you want to continue to waste much of the BBOL programming effort on GIB, replace its favourite total points system with another evaluation system.

Unfortunately, none of us are AI programmers. We're going to have to tread very carefully if we're thinking of changing basic design of GIB.

Also, there's an enormous rulebase that's tied to its current evaluation methods. If we make a radical change to the GIB engine, we'll have to go through thousands of rules, adjusting them to work with the new evaluation method.

#18 User is offline   xxhong 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 328
  • Joined: 2010-November-11

Posted 2011-November-08, 13:15

Simulation is not a sure thing. It's a matter of luck, especially when the sample size is only 10-20 hands (Also, gib program has bugs to simulate when it holds a strong hand, it often generates wrong results, that's why gib rarely bid correctly after 5NT king asking).

Trick counting is way much simpler and really it should be applied. For this example hand, a trick counting method would just say that this hand is likely to produce 13 tricks, so a 7S should be bid.

Also, the basic philosophy of trick counting is way more solid than simulations. Now the simulation is based on the bidding history. If the opp intends to provide the wrong bidding, simulations would usually fail. Trick taking is always sound no matter what opps bid. This is extremely important in human-gib vs. human-gib money games.

This is really not very difficult to implement a trick counting method, especially for suits like AKQJxx. All you need is a simple table of honor combination, suit length and their corresponding tricks. I think this simple evaluation would improve gib's bidding a lot.


View Postbarmar, on 2011-November-07, 16:06, said:

GIB does have the notion of suit quality; these are represented in its descriptions as "biddable", "rebiddable", "twice rebiddable"; it has additional categories that don't have names (for hands like AKQxxxx) that can be used in bidding rules. But it's hard to write bidding rules that cover all these cases, especially for rare cases like this one.

The theory is that simulations should come to the same conclusion. If AKQJxx usually takes 6 tricks, this should be reflected in most of the simulation results.

0

#19 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,398
  • Joined: 2004-August-21
  • Gender:Male

Posted 2011-November-08, 15:09

GIB uses more than 10-20 hands in its simulations. We can tell it to use up to 250 hands, I'm not sure offhand what we have it currently set to.

And while trick counting works extremely well for freak hands like this, it's not really applicable to the vast majority of hands. I could imagine this as a special module that's only used to check for slams and grands. Adding trick counting to GIB wouldn't be hard. It already has a routine to calculate the suit quality metric I mentioned above, and it can calculate the tricks for solid suits at the same time.

I'm just not convinced that there's much bang for the buck doing this. Hands like these are so rare, we've maybe spent more time discussing it than anyone will spend playing these types of hands in the next year (unless we decide to create a Robot Goulash tourney -- which would probably be a disaster).

#20 User is offline   xxhong 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 328
  • Joined: 2010-November-11

Posted 2011-November-08, 15:37

I think it can be applicable to many other situations too. Counting for slams is the first step.

For example, a sound 2C opening scheme should also involve total trick counts if the HCP is less than 20.

Also, for any high level competitive decisions, gib should have total trick counts as well. Now the scheme is very wrong. Generally, when gib thinks its side is likely to make a game, it would double no matter which hand type it holds. It can be based on a lot of tricks and offensive oriented hands, it can also be based on quite defensive oriented hands. The total trick count would offer a pretty good guideline to decide to bid or to double.

Also, for penalty double decisions, trick count is especially useful. Suppose you hold xxx xx AKQJxx Ax, you have only 14 HCPs, but 7 tricks, so you actually should double 1NT openings. This is also very true for high level penalty doubles. For example, if you hold SAKQJ and opps bid 4S after a 2/1 constructive sequence with 4-4 fit, you should double this 4S to take your sure money. Only when trick count is implemented can you really implement good penalty doubles at high level. Now gib simply misses many such kind of chances.

The number of total tricks you can take is a very important parameter in bridge bidding. Without this parameter, the bidding can be very difficult, and a lot of nonsense sequence would occur. If every bridge player would consider the total trick taken in every hand, there is just no point not to allow bridge program to have this ability IMO.


View Postbarmar, on 2011-November-08, 15:09, said:

GIB uses more than 10-20 hands in its simulations. We can tell it to use up to 250 hands, I'm not sure offhand what we have it currently set to.

And while trick counting works extremely well for freak hands like this, it's not really applicable to the vast majority of hands. I could imagine this as a special module that's only used to check for slams and grands. Adding trick counting to GIB wouldn't be hard. It already has a routine to calculate the suit quality metric I mentioned above, and it can calculate the tricks for solid suits at the same time.

I'm just not convinced that there's much bang for the buck doing this. Hands like these are so rare, we've maybe spent more time discussing it than anyone will spend playing these types of hands in the next year (unless we decide to create a Robot Goulash tourney -- which would probably be a disaster).

0

  • 2 Pages +
  • 1
  • 2
  • You cannot start a new topic
  • You cannot reply to this topic

1 User(s) are reading this topic
0 members, 1 guests, 0 anonymous users