gib's bidding seems improved a lot

[junyi_zhu]

I just downloaded the new version of bbo and tested gib's bidding in bidding practice room. The most tests I did were 1NT openings. Now gib's bidding is much "saner" than before. One problem I found is that after a RKC sequence and 5NT by gib, it tends to overbid qutite a lot.

example 1:

AKQxxx ATxx x xx

1NT 4H(by gib)

4S 4N

5H 5N

6D 7S

I had xxx KJ9 AKT Axx,

so it's a 50% grand slam. luckily I didn't hold something like:

Jxx Qxx AKQ Axx

 

Overally, it was a good job.

Well done Fred.

[fred]

Overally, it was a good job.

Well done Fred.

Thanks Junyi, but Uday deserves most of the credit.

 

He is the one that made all of the actual programming changes. I was mostly involved in providing additional bridge expertise (Uday won't admit it but he has plenty of his own) and helping with testing.

 

The 5NT thing is problematic. We were not entirely happy with our "solution" and likely we will revisit this at some point. In the interest of avoiding further delays to today's release we decided to stop trying (for now at least) to get this right.

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

[H_KARLUK]

I got a msg posted by BBO when i logged. Downloaded and upgraded. Then i logged off. After a minute i logged in and signed.

 

We were 7 players. Finished as 2nd. It was "random" category. So far I did not see anything wrong "e x c e p t" tho my sys is really fast there were some lags.

 

AKQxxx ATxx x xx

1NT 4H(by gib)

 

 

I am surprised. If I got such collection I wld bid 2♣

 

Say 2 ♦ received.

 

4 ♦ is obvious ;-) describes 6♠/4 ♥forcing (smolen)

4 ♣ describes 6♥ / 4♠ !

[junyi_zhu]

Overally, it was a good job.

Well done Fred.

Thanks Junyi, but Uday deserves most of the credit.

 

He is the one that made all of the actual programming changes. I was mostly involved in providing additional bridge expertise (Uday won't admit it but he has plenty of his own) and helping with testing.

 

The 5NT thing is problematic. We were not entirely happy with our "solution" and likely we will revisit this at some point. In the interest of avoiding further delays to today's release we decided to stop trying (for now at least) to get this right.

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

Another old problem is the range of raising to 6NT.

Here comes this hand:

KTxxx QTx Jxx Tx

my hand is

Ax AKx AKQxx KQx, certainly it has some plays, but my hand can certainly be

weaker.

Gib still thinks his hand is strong enough to push to 6NT after my 3NT opening, which shows 25-27. That means a lot of bad 6NT could be reached. In old versions, basically gib thinks 31 HCP is enough to push to 6NT it seems.

[1eyedjack]

I downloaded and installed the GIB upgrade, and my version number of the software as displayed in the login screen remains 5.0.3. Not even changed to (say) 5.0.3a

 

Is that intended? (Once you get past the splash screen I cannot find an option to display the version number, such has Help/About)

[H_KARLUK]

I downloaded "Upgraded GIB", saved it in my hard drive.

 

When I executed setup engine worked well in to correct path I navigated.

 

I tested immediately in online random robot rewards.

 

No aberration occurred yet. Card reading and defence was over average.

 

Welcome again partner. You are silent as usual

 

Good work. Really. I am happy to see th BBO cares member satisfaction.

[fred]

I downloaded and installed the GIB upgrade, and my version number of the software as displayed in the login screen remains 5.0.3. Not even changed to (say) 5.0.3a

 

Is that intended? (Once you get past the splash screen I cannot find an option to display the version number, such has Help/About)

Yes, that's intended.

 

The Windows client (to which the version number applies) has not changed. The GIB upgrade changes only files that are directly related to GIB.

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

[pigpenz]

I downloaded and installed the GIB upgrade, and my version number of the software as displayed in the login screen remains 5.0.3. Not even changed to (say) 5.0.3a

 

Is that intended? (Once you get past the splash screen I cannot find an option to display the version number, such has Help/About)

i think you have to go to

bridge base online\robots\gib\bridge.exe

 

from the dos prompt i think you can find out what version gib is

....bridge /? possibly

 

mine says GIB version 6.1.3, 1/4/03 still

[fred]

I downloaded and installed the GIB upgrade, and my version number of the software as displayed in the login screen remains 5.0.3.  Not even changed to (say) 5.0.3a

 

Is that intended?  (Once you get past the splash screen I cannot find an option to display the version number, such has Help/About)

i think you have to go to

bridge base online\robots\gib\bridge.exe

 

from the dos prompt i think you can find out what version gib is

....bridge /? possibly

 

mine says GIB version 6.1.3, 1/4/03 still

We have not bumped up the version number on bridge.exe either. Probably we should have done this...

 

One way to make sure the latest version is properly installed is to check the modification dates of the files in in the directory that Pigpenz refers to. If the dates are from November 2008 then the upgrade was successful.

 

Another way to make sure is to play some with GIB. If you see GIB transfer to a 4-card suit then something went wrong with the upgrade :rolleyes:

 

Fred Gitelman

Bridge Base Inc.

www.bridgebase.com

[Walddk]

I agree that GIB has improved a lot, but it surprises me that fairly simple card combinations are not handled correctly. An example from today where GIB is in 4♠ with this trump suit ...

 

Q1072

 

AK954

 

GIB played a low spade to the queen. It did not matter since the suit broke 2-2, but one would assume that it had been programmed to cater for a 4-0 break by cashing a top honour from hand.

 

Roland

[hotShot]

I agree that GIB has improved a lot, but it surprises me that fairly simple card combinations are not handled correctly.

....

It did not matter since the suit broke 2-2, but one would assume that it had been programmed to cater for a 4-0 break by cashing a top honour from hand.

 

Roland

I doubt that GIB handles suit combinations at all, it generates deals that fit the 2 known hands and the bidding. Without other informations available, the change of a 4-0 split is 9.6%. If you assume that GIB can handle about 10 generated deals in the given time, there might be no 4-0 split in the generated sample.

GIB will play the card that is best in most of the generated deals. If there was no 4-0 split in the sample, GIB won't cater for it.

 

If there had been some bidding from opponents or more informations are available from previously played cards, GIBs generated hands will be close to the actual hand. With enough information available it will do better than a human player, because it uses probabilities that are adapted using the knowledge.

If no informations are available, GIB can't apply restrictions to the unknown hands and will generate and analyze deals that deviate a lot from the actual deal. The resulting analysis might fail for the actual deal.

 

The card played by GIB may seem strange to a human player, even if it is a good move.

[pigpenz]

I agree that GIB has improved a lot, but it surprises me that fairly simple card combinations are not handled correctly. An example from today where GIB is in 4♠ with this trump suit ...

 

Q1072

 

AK954

 

GIB played a low spade to the queen. It did not matter since the suit broke 2-2, but one would assume that it had been programmed to cater for a 4-0 break by cashing a top honour from hand.

 

Roland

yes I have noticed that GIB seems to ignore what would be normal safety plays

[barmar]

Note the subject line of the thread, "gib's bidding has seems improved". AFAIK, the work the BBO people have done recently on GIB has only been on its bidding, they haven't made any changes to the way it plays the hands.