[onoway]

I see a lot of info to explain what GIB is doing with the bidding but where oh where is there some sort of explanation about GIB leads?

It seems that bidding a suit is the surest way to make sure that a GIB partner NEVER leads it, often preferring to lead the trump suit if not the first lead then the second, no matter what it has or hasn't got in trumps. In the last month or so I have played with them, my GIB p has led my bid suit exactly ONCE. Even when getting the lead after the first round it will dodge around all the other suits before it will lead the one I bid. I have never been able to figure out how to signal because it seems no matter what I discard GIB goes its merry way regardless.

Is there some sort of method to the GIB play that I just haven't found yet and if so, PLEASE point me to it as this is extremely frustrating, and now fairly often leads to me just snarling at it and going off to do something else.

[Stephen Tu]

GIB doesn't lead using human heuristics. It leads much like it does on every play -- it deals a bunch of random hands that fit the auction, then picks the card that tends to do best on average. In theory actually GIB can outperform humans on opening lead; humans blow a ton of tricks on lead and maybe GIB is right that more passive leads are better. The best we can do is to improve the bidding database so that its estimate of what declarer and dummy hold are more accurate, then in theory it should pick better leads.

If it is not leading overcalled suits as often as it is supposed to, maybe it is supposed to bias the deal generator a bit more on the assumption of better quality suits in partner's hand. Also vs. NT on lead of a low card it should probably bias it more towards 5+ cd suits.

[barmar]

I find that GIB will lead my suit unless it has something else on its mind. It tends to be very eager to lead singletons and doubletons, even in suits the opponents have bid.

As Stephen said, its leads are made based on simulations, not rules. The only rules it uses are when it's leading from honor sequences, it knows the standard honor leads (but v21 had a bug in one of them -- I think we fixed it in v22).

GIB doesn't generally pay attention to your carding, except that it can recognize the standard honor leads.

[Bbradley62]

barmar, on 2012-April-30, 15:24, said:

The only rules it uses are when it's leading from honor sequences, it knows the standard honor leads (but v21 had a bug in one of them -- I think we fixed it in v22).

GIB still doesn't lead Q from KQT9(x) against NT...

---

and he still leads low from xxx against NT:

---

[Bbradley62]

barmar, on 2012-April-30, 15:24, said:

The only rules it uses are when it's leading from honor sequences, it knows the standard honor leads (but v21 had a bug in one of them -- I think we fixed it in v22).

Basic bot on opening lead: http://tinyurl.com/79yz5l9

[wynsten]

I just started a topic "Robot Defence" which wasn't particularly about leads, but if you look at the hand I posted, the Robot lead is inexplicable (bidding is 2nt - 3nt; With KJxx spades and Axxxx diamonds robot leads top of xx hearts).

[pigpenz]

as a declarer i find the bot leads challeging
they usually dont give up much and takes awhile to realize that the suit is actually
alot of times with the other defender

now when I lead like GIB i get killed
I stick to the old and true lead what you got unless there is something in the auction to
say other wise.

[wynsten]

That short suit lead from nothing stuff works OK against trump contracts, but not so much at notrump and you're right - I've seen that "suit with the other defender" phenomenon, but it didn't register with me that it was systemic. Makes sense though. Thanks for the tip. If robot is reluctant to lead from an honour, then since long suits tend to have more honours than short suits, he is going to avoid leading long suits. That's OK for trump contracts, and notrump slams, but good old 3NT needs to find a long suit and statistically your best shot is the one you are looking at.

[pigpenz]

wynsten, on 2012-May-01, 20:48, said:

That short suit lead from nothing stuff works OK against trump contracts, but not so much at notrump and you're right - I've seen that "suit with the other defender" phenomenon, but it didn't register with me that it was systemic. Makes sense though. Thanks for the tip. If robot is reluctant to lead from an honour, then since long suits tend to have more honours than short suits, he is going to avoid leading long suits. That's OK for trump contracts, and notrump slams, but good old 3NT needs to find a long suit and statistically your best shot is the one you are looking at.

problem is that GIB plays one GIB for something
then the other GIB plays the first one for what he lead
they keep switching back and forth alot

thats why so many hopeless 3NT contracts make

[barmar]

The basic issue is that GIB doesn't generally try to figure out what a lead implies about the player's hand. So just as it doesn't automatically return the suit that was led (on the assumption that the lead was the player's best suit), neither does it automatically switch (on the assumption that it must be a passive lead).

While its strategy (or lack of strategy, if you will) sometimes lets you make unmakeable contracts, it also seems to set many contracts that the more natural defense would have let make. I've been in plenty of 3NT contracts where a human would have led away from their King into my AQ, giving me my 9th trick. This is totally anecdotal, I haven't tried to keep statistics of which approach is better.

[pigpenz]

barmar, on 2012-May-04, 12:29, said:

The basic issue is that GIB doesn't generally try to figure out what a lead implies about the player's hand. So just as it doesn't automatically return the suit that was led (on the assumption that the lead was the player's best suit), neither does it automatically switch (on the assumption that it must be a passive lead).

While its strategy (or lack of strategy, if you will) sometimes lets you make unmakeable contracts, it also seems to set many contracts that the more natural defense would have let make. I've been in plenty of 3NT contracts where a human would have led away from their King into my AQ, giving me my 9th trick. This is totally anecdotal, I haven't tried to keep statistics of which approach is better.

yes at times their leads are diabolical
but at times its like watching a ping pong ball go back and forth
even after they had already found the killing lead or switch.

I wish we could get GIB data for use with bridgebrowser to see what the
results are of GIB's leads vs human leads....but just watching the vugraphs the
last week you dont see too many GIB type of leads being made