What lead tables, if any does GIB employ?

[calm01]

http://tinyurl.com/8ydvaf2

[Bbradley62]

Third from your thumb.

[barmar]

GIB has a table of opening leads, but I don't think it's used in the middle of the hand except for honor leads.

[calm01]

Barmar said:

 

"GIB has a table of opening leads, but I don't think it's used in the middle of the hand except for honor leads."

 

Perhaps the opening lead approach could be extended to the middle of hands. Then GIB will begin to get past the palooka stage as a defender.

 

Fixing leads is much more important than fixing an aspect of bidding because leads impact multiple times in each and every hand.

 

Being able to trust partners leads informs what to do and what not to do, where high cards are and where high cards are not, what declarer needs to do and how to disrupt declarer. So leads provide critical insight in the defence. Knowing when to lie to partner is a skill that can only be built on a basis of trust.

 

Fixing leads for GIB must be a mission critical issue.

 

Go to it wit gusto and pleasure.

[barmar]

Fixing leads for GIB must be a mission critical issue.

Fixing GIB play is very hard. How do we know when to override what the simulations choose?

 

Maybe it had a good reason for leading the 9, perhaps it was unblocking to keep off an end-play. On this layout it didn't work, but maybe other layouts would. It's very hard to tell why GIB makes some plays. There's no planning, just simulations with double dummy analysis.

 

It's all well and good to say that it's mission critical, but we don't have any AI programmers on staff to redesign this thing. Unless we replace GIB with another bridge program, we're stuck with its general design.

[manudude03]

Fixing GIB play is very hard. How do we know when to override what the simulations choose?

 

Maybe it had a good reason for leading the 9, perhaps it was unblocking to keep off an end-play. On this layout it didn't work, but maybe other layouts would. It's very hard to tell why GIB makes some plays. There's no planning, just simulations with double dummy analysis.

 

It's all well and good to say that it's mission critical, but we don't have any AI programmers on staff to redesign this thing. Unless we replace GIB with another bridge program, we're stuck with its general design.

 

It might not be possible, but is there any way of getting it to show you the hands it simulated? That might give you an idea if it's a problem with its criteria.

[xxhong]

I think playing K to deny A, Q to deny K, J to deny Q, T to deny J and 9 to deny T is a must in switching when holding sequences. It's easy to implement and may improve the performance a lot. That's actually what wbridge5 is doing in its defense I think.

 

Fixing GIB play is very hard. How do we know when to override what the simulations choose?

 

Maybe it had a good reason for leading the 9, perhaps it was unblocking to keep off an end-play. On this layout it didn't work, but maybe other layouts would. It's very hard to tell why GIB makes some plays. There's no planning, just simulations with double dummy analysis.

 

It's all well and good to say that it's mission critical, but we don't have any AI programmers on staff to redesign this thing. Unless we replace GIB with another bridge program, we're stuck with its general design.

[barmar]

It might not be possible, but is there any way of getting it to show you the hands it simulated? That might give you an idea if it's a problem with its criteria.

Yes, we have tools to see the hands it's using in the simulations. But there are 250 of them, so trying to get a sense of what they imply is hard. Mostly what I can tell with this is when it hasn't made an inference that a good human player would have, such as a lead being a singleton.