[steve2005]

advanced robots challenge match
another bad lead: trump. here K♠ or ♦ should be safe.
obvious south has 1♥ so need ten or Q to be ok.

South should double diamonds but still safer to lead ♦

hand

[Stephen Tu]

Maybe they just aren't giving the bots enough time/resources to run large enough samples to get it better. My home PC GIB likes to lead a spade on this hand, also on the other thread you gave earlier.

That said, robot leads tend to look different from human leads in general, which are more based on rules of thumb than rigorous statistical evaluation. I don't think that is necessarily a bad thing, I think humans are kind of terrible on opening lead themselves. It may well be that the best lead on some hands in general is counter-human-intuition. I tend to trust computer's gut feeling based on its samples (when it's idea of what people have based on the auction isn't screwed up by faulty bidding definitions) more than my own, usually.

The bigger problems IMO, which are hard to solve, are:
- inability to signal/interpret signals
- too much credit given to opp's bidding so they "must" have certain # of HCP or certain # of cards in a particular suit, not sampling with looser constraints to improve tie-breaking
- not looking at things from single dummy perspective, assuming declarer is getting things right so then unnecessarily giving away certain card positions.

[steve2005]

This was with advanced robots.
But if there is a bias against leading bid suit or if there is a bug giving Gib all the time in the world wont change outcome.
That your PC Gib found a spade lead makes me suspect BBO has made a change in the program accidently or on purpose and screwed it up!

[virgosrock]

Stephen Tu, on 2018-April-29, 07:38, said:

Maybe they just aren't giving the bots enough time/resources to run large enough samples to get it it better. My home PC GIB likes to lead a spade on this hand, also on the other thread you gave earlier.

That said, robot leads tend to look different from human leads in general, which are more based on rules of thumb than rigorous statistical evaluation. I don't think that is necessarily a bad thing, I think humans are kind of terrible on opening lead themselves. It may well be that the best lead on some hands in general is counter-human-intuition. I tend to trust computer's gut feeling based on its samples (when it's idea of what people have based on the auction isn't screwed up by faulty bidding definitions) more than my own, usually.

The bigger problems IMO, which are hard to solve, are:
- inability to signal/interpret signals
- too much credit given to opp's bidding so they "must" have certain # of HCP or certain # of cards in a particular suit, not sampling with looser constraints to improve tie-breaking
- not looking at things from single dummy perspective, assuming declarer is getting things right so then unnecessarily giving away certain card positions.

Well said Stephen. I am ok with GIBBO's leads except when failure to lead pd's suit lets them make the contract
The ones I hate are when it will lead A or AK and regardless of what you play will shift. But then even those are rare I must concede.
Also, I don't think form of scoring - MP/IMP/rubber/total points - matters for GIBBO.

vrock

[steve2005]

virgosrock, on 2018-April-30, 07:46, said:

Well said Stephen. I am ok with GIBBO's leads except when failure to lead pd's suit lets them make the contract
The ones I hate are when it will lead A or AK and regardless of what you play will shift. But then even those are rare I must concede.
Also, I don't think form of scoring - MP/IMP/rubber/total points - matters for GIBBO.

vrock

Gib switches when it gets a favourable result on opening lead way too often. There has to be a bug.

Form of scoring does matters but not much, i think is just how result is scored. But is just trying to maximize average score.

[virgosrock]

steve2005, on 2018-April-30, 08:42, said:

Gib switches when it gets a favourable result on opening lead way too often. There has to be a bug.

Form of scoring does matters but not much, i think is just how result is scored. But is just trying to maximize average score.

yes it shifts too often. i think it assumes you have led from nothing like it does. i have tried leading 5's or higher because 4's or lower almost guarantees a shift. Does not really help.
i don't think it even knows the rule of 11. However, when IT leads the 5 or 6 i am fairly successful leading the same suit back. Weird!!

vrock

[steve2005]

virgosrock, on 2018-April-30, 11:12, said:

yes it shifts too often. i think it assumes you have led from nothing like it does

vrock

Sometimes declaring Gib has a choice of two finesses to take.
Gib picks one and it works,
Comes back to hand and it doesn't do the one that worked. Yes they could be holding up, but maybe it worked!!

[virgosrock]

steve2005, on 2018-April-30, 12:07, said:

Sometimes declaring Gib has a choice of two finesses to take.
Gib picks one and it works,
Comes back to hand and it doesn't do the one that worked. Yes they could be holding up, but maybe it worked!!

LOL yes. you duck to throw it off. never takes the finesse again. well, "never" is too strong - rarely takes same hook again.

vrock