I really think GIB is getting worse

[pretender]

Bidding is one thing, but you would think GIB should excel in the play of the hand, even if we're not getting the best GIB for the robot races. Here are a few hands illustrating play/defense errors that happen way more often to GIB than they should.

 

The "I completely trust your bidding when defending the hand" error:

http://tinyurl.com/lm35n3r

I've often made 6 of something with GIB not cashing its two defensive tricks, but this is the first time I've seen it not cash the second trick so late in the hand (6th trick)

 

The "It's so unlikely this irrelevant play will hurt me so I'll make this irrelevant play anyway instead of the highest % play" error:

http://tinyurl.com/m4qqh8y

Sure the play is irrelevant if the weak 2 opener had 6 diamonds, but what if it had 7? And it's not like the bidding description specifically says only 6 diamonds.

 

The "Isn't the whole point of a robot player so that it can calculate the most basic safety plays?" error:

http://tinyurl.com/lnjsg6n

 

The "WTF I'd love to hear any reason anybody can come up with for this play" error:

http://tinyurl.com/mx4guo8

This hand occurred today. Key point happens at trick 7.

[cloa513]

Bidding is one thing, but you would think GIB should excel in the play of the hand, even if we're not getting the best GIB for the robot races. Here are a few hands illustrating play/defense errors that happen way more often to GIB than they should.

 

The "I completely trust your bidding when defending the hand" error:

http://tinyurl.com/lm35n3r

I've often made 6 of something with GIB not cashing its two defensive tricks, but this is the first time I've seen it not cash the second trick so late in the hand (6th trick)

 

The "It's so unlikely this irrelevant play will hurt me so I'll make this irrelevant play anyway instead of the highest % play" error:

http://tinyurl.com/m4qqh8y

Sure the play is irrelevant if the weak 2 opener had 6 diamonds, but what if it had 7? And it's not like the bidding description specifically says only 6 diamonds.

 

The "Isn't the whole point of a robot player so that it can calculate the most basic safety plays?" error:

http://tinyurl.com/lnjsg6n

 

The "WTF I'd love to hear any reason anybody can come up with for this play" error:

http://tinyurl.com/mx4guo8

This hand occurred today. Key point happens at trick 7.

 

For Hand 3- you mean why doesn't it cash the A♠ first.

 

Hand 4 is like the classic discard on card to be ruffed to guarantee down on contact.

 

 

Which GIB are you using weekly or daily rental?

Honestly a GIB surprised me when playing free first of the month- killing an otherwise good contract from Vgraph (no defenders found the play). Unfortunately due to the crap system BBO has- that is lost forever.

And the other plays were surprisingly good too.

[rwbarton]

The first three are not so surprising if you know how GIB works, considering how little time it is given to calculate per play in the robot races. If you think you can write a better bridge-playing robot, give it a try B-)

 

The fourth hand is truly odd, though. The play on trick 5 is not so good either. This one looks like an honest bug as far as I can see.

[pretender]

The first three are not so surprising if you know how GIB works, considering how little time it is given to calculate per play in the robot races. If you think you can write a better bridge-playing robot, give it a try B-)

 

The fourth hand is truly odd, though. The play on trick 5 is not so good either. This one looks like an honest bug as far as I can see.

 

The fourth hand is the truly crazy one, but not the first time I've seen GIB make a similar mistake.

 

The first two are not surprising, but the third one still seems like something GIB should get right, even if it is sped up for the robot races.

[rwbarton]

The first two are not surprising, but the third one still seems like something GIB should get right, even if it is sped up for the robot races.

 

GIB (specifically, the version of GIB on BBO) has no notion of considering suits in isolation. All it does is generate entire layouts that are consistent with the information it has so far, compute the double-dummy value of each valid play in each layout, and pick the play (or one of the plays) that has the highest average double-dummy value. Double-dummy analysis of a specific 12-card position takes a fairly long time, on the computer time scale. Based on the reports of GIB's play here, I would guess that the robots typically are not given time to analyze much more than 10 layouts on average, and perhaps at early stages as low as 1 layout. Maybe barmar knows the actual number?

 

If no layout among the (say) 10 that were analyzed has the specific JT63 - trump layout, then GIB's choice at trick 2 between South's spades will be random. This is not an easy problem to solve, given that BBO's resources for running GIB are finite.

[1eyedjack]

The "It's so unlikely this irrelevant play will hurt me so I'll make this irrelevant play anyway instead of the highest % play" error:

http://tinyurl.com/m4qqh8y

Sure the play is irrelevant if the weak 2 opener had 6 diamonds, but what if it had 7? And it's not like the bidding description specifically says only 6 diamonds.

 

Somewhat drifiting from the purpose of the thread but I thought that the bidding was rather bizarre, although to be fair the human element may have contributed.

 

I thought that South was a bit strong for 4H, although perhaps the "get in and out fast" maxim of the race medium takes precedence.

 

I thought that North was far too weak to try 4N. That was until I moused over the 4H bid and the description shows "25+ total points". With no upper limit on strength then presumably the bid is forcing (which perhaps makes it a worse bid than I thought before?). However I take the "no upper limit" with a pinch of salt. I cannot believe that the bid can't be passed, so I put that down to lazyness in entering the description. Also 25 total points opposite a possibly yarborough makes the 5 level (in my view) an unacceptable risk.

 

If forced to bid I would expect 5H to be weaker than 4N and would therefore by my choice. But I don't now get the option of mousing over the option to determine how GIB would mean it.

If not forced to bid I think it is a clear pass, despite the 25+ description of 4H. Sure, you would miss a good slam on this hand, but that I think is more down to the 4H bid.

 

And then we get to the 5N bid!! Looking for a grand? Seriously? The entire table already knows that all key cards are held with the 5S bid.

[1eyedjack]

I don't play in race tourneys. Just not my thing. But I do observe that in NON-race robot tourneys GIB bids and plays blindingly fast. Every so often it hitches for maybe 5 seconds, but not for longer and rarely at that. I cannot imagine that the extra speed of play achieved by using a lower standard of GIB in a race tourney compensates sufficiently for the drop off in standard.

 

If anything, in non-race robot tourneys, I for one would be happier if GIB played a bit slower than it does currently, if its standard were noticeably to improve thereby.

[rwbarton]

I assume the time GIB is given per play is limited by how much BBO spends on running all the GIBs on their servers, not by how long the typical user is willing to wait for GIB to play. The advanced bots play slower, right? That extra thinking time is what you're paying extra for.

[1eyedjack]

I assume the time GIB is given per play is limited by how much BBO spends on running all the GIBs on their servers, not by how long the typical user is willing to wait for GIB to play. The advanced bots play slower, right? That extra thinking time is what you're paying extra for.

I only ever play against advanced bots (in robot non-race tourneys) and seldom encounter a noticeable delay while it thinks. What you say about how their resources are calculated may be right, I wouldn’t know. But I have my doubts, given that all bots on all hands in a tourney, which are not played simultaneously, must be assigned identical resources.

[uday]

It isn't uncomplicated.

 

There are two things that distinguish the 'basic' bots from the 'advanced' bots.

 

1. When GIB needs to make a call, should it simply use the rules it has at hand, and make the bid that the rules tell it to make, or should it be allowed to exercise 'judgment' by simulating what might happen if it used a bid that was 'close' to the recommended answer, and then pick a slightly off color bid if it judges that it will work better?

 

2. How much time should GIB take?

 

Both matter. The simulations in the bidding are definitely a double-edged sword. They're great at deciding whether to move after 1n p 4n p. In some more complicated situations, not so much.

 

The time that we allot to GIB affects both think time and the number of deals used in the samples (which in turn drive what it thinks will work ).

 

The basic bots get a lot less time. In addition, they're told to avoid simulations during the bidding. This makes them a lot (a lot) faster and cheaper.

 

The tourney bots are assigned identical resources. There is a case to be made that this should not be the case but...

[chasetb]

This one escapes any logic. Here is the play at my table, basic robots and dial-up - http://tinyurl.com/kgx78lg

 

Here's the same bloody hand, thrice - http://tinyurl.com/lnzee2s / http://tinyurl.com/lxsa78e / http://tinyurl.com/k4qs2lj

 

Skip to a different hand. First off, my description should be only 5+ ♠, and second, why is GIB jumping to 4 when it doesn't have a hand worth it?