Bidding Failure?

[calm01]

I enjoy playing against 3 robots - the service works well and the erratic behaviour of the robots gives them a sort of hunan rouch.

 

However sometimes their extreme behaviour in the bidding exceeds the capabilites of the Rueful Rabbit.

 

The bidding was:

 

pass pass pass 2NT

pass pass pass.

 

I bid 2NT with AK95, A86, K7, KQJ10.

 

My robot partner passed with an 11 point hand including 2 aces and a good 6 card suit. The robot hand was 1064, Q, AJ10942, A93.

 

Taking the kind view, this smells of bidding on the basis of an analysis from a portfolio of similar hands and the corresponding historic results.

 

Such an approach is doomed to failure as bidding on what partner has promised and known combined assets is probably the only sensible way forward unless millions of hands are available for analysis on each and every distinct bidding sequence and vulnerability.

 

Perhaps it is time to review the robot bidding appriach and start again with a more sensible crew/approach.

 

What is your view of the best way forward?

 

Regards, Ric (userod calm01)

[barmar]

Very strange, since my experience is that the robots are normally very aggressive in responding to 2NT, frequently raising to game with 4 HCP.

 

Was this in the MBC or a tourney? I think the robot settings are different (I don't usually play in the MBC, but when I do they seem to play much faster).

[junyi_zhu]

I enjoy playing against 3 robots - the service works well and the erratic behaviour of the robots gives them a sort of hunan rouch.

 

However sometimes their extreme behaviour in the bidding exceeds the capabilites of the Rueful Rabbit.

 

The bidding was:

 

pass pass pass 2NT

pass pass pass.

 

I bid 2NT with AK95, A86, K7, KQJ10.

 

My robot partner passed with an 11 point hand including 2 aces and a good 6 card suit. The robot hand was 1064, Q, AJ10942, A93.

 

Taking the kind view, this smells of bidding on the basis of an analysis from a portfolio of similar hands and the corresponding historic results.

 

Such an approach is doomed to failure as bidding on what partner has promised and known combined assets is probably the only sensible way forward unless millions of hands are available for analysis on each and every distinct bidding sequence and vulnerability.

 

Perhaps it is time to review the robot bidding appriach and start again with a more sensible crew/approach.

 

What is your view of the best way forward?

 

Regards, Ric (userod calm01)

I actually agree that the basic design of gib bidding is very flawed. Bidding should be based on rules and occasionally on simulations for high level competitive auctions. Currently, gib's bidding is very random and often breaks many rules as long as gib forms some unlucky situations in its very limited simulations. That also means that gib would give different bids if the simulation gives different results. This is very difficult to cope with IMO. When I open 2NT, I often see gib raise it to 3NT with very little value, like one Jack one Queen or 4 jacks. Also, when I show a hand in the range of 20-21 points in competitive auctions, I often see gib overbids to games with jacks. These are really very very bad IMO and many players I know just don't play with gib for such kind of reasons including my regular partner.

 

Also, it is really not too difficult to implement a feature to allow the human to declare when in dummy. It's just frustrating and pointless to watch gib goes down in many cold slams and games because some intrinsic problems. Gib's declarer play is about an average club player, and there is really no point to see gib butcher so many contracts playing one's own money. Also, this implementation would make playing with gib way more enjoyable, cause you only need to tolerate the bad bids and bad defense by gib, not the bad declaring plays. This implementation should draw a lot of players into MB IMO.

 

My last point is that a counting signal should really be implemented into gib's defense. It's really trivial to do such kind of programming. An attitude signal is not that easy to implement, but count is really not that difficult. With the simple counting defensive signals, many defensive situations can be solved and defense can be much more enjoyable.

[arigreen]

Barmar is correct that GIB behaves differently depending on the type of game. In tournament play and money bridge GIB is given more time to think and it often simulates during the bidding. The GIBs in the main bridge club do not simulate.

 

GIB does use a database of rules when bidding. Every now and then there is a gap in the rules and no rules match. If simulations are enabled, GIB will usually simulate among reasonable choices and make some bid to keep the auction open if pass would be egregious. When simulations are disabled, however, GIB has no choice but to pass.

 

The hand in question demonstrates one of these gaps. In the main bridge club, GIB would pass 2NT, but in tournaments or at money bridge GIB would simulate and usually bid 6N.

[cloa513]

Any chance that GIB would simulate the 6D which is solid.

[arigreen]

Not this time, unfortunately.

[junyi_zhu]

Barmar is correct that GIB behaves differently depending on the type of game. In tournament play and money bridge GIB is given more time to think and it often simulates during the bidding. The GIBs in the main bridge club do not simulate.

 

GIB does use a database of rules when bidding. Every now and then there is a gap in the rules and no rules match. If simulations are enabled, GIB will usually simulate among reasonable choices and make some bid to keep the auction open if pass would be egregious. When simulations are disabled, however, GIB has no choice but to pass.

 

The hand in question demonstrates one of these gaps. In the main bridge club, GIB would pass 2NT, but in tournaments or at money bridge GIB would simulate and usually bid 6N.

it appears that gib's rules are not very consistent and there are many holes in the rule, which may lead to the pass in many situations I guess. Still, it's a large amount of work to make the rules consistent IMO.

[calm01]

Could a simple check reduce the size of these occasional bidding failures?

 

For example:

 

When a combined fit is available (at least 8 diamonds are implied) and over 28 points are held (here at lest 29 are implied) the robot must bid to game in NT or a suit.

 

This would at least get the partnership to game when a slam is on and avoid the programming embarrassment of ending in a part score in a slam hand.

[junyi_zhu]

Could a simple check reduce the size of these occasional bidding failures?

 

For example:

 

When a combined fit is available (at least 8 diamonds are implied) and over 28 points are held (here at lest 29 are implied) the robot must bid to game in NT or a suit.

 

This would at least get the partnership to game when a slam is on and avoid the programming embarrassment of ending in a part score in a slam hand.

Good idea. Now the program doesn't take rules very seriously. I think a few rules should be reinforced here:

1 With combined 25 HCP or more, game has to be bid.

2 With combined 33 HCP or more, slams has to be bid (except that bidding shows no controls in one suit)

3 with combined 37 HCP or more, grand slams has to be bid (unless miss trump K).

I think any simulations should never override this 3 robots' fundamental laws in constructive bidding.

 

Simulations just override way too many rules. That's why we often see gib bids 6NT with only 30-31 HCP no fit.