BBO Discussion Forums: New bots? - BBO Discussion Forums

Jump to content

Page 1 of 1
  • You cannot start a new topic
  • You cannot reply to this topic

New bots? Bidding problems

#1 User is offline   wuudturner 

  • PipPipPip
  • Group: Full Members
  • Posts: 81
  • Joined: 2011-November-07

Posted 2012-April-07, 08:35

Two hands out of 12 boards:

Vulnerable versus not, holding

AJ864
QT85
972
2

You (a bot) pass as dealer. LHO opens 1!C, partner bids 2NT, of course unusual showing the red suits.

With this hand, does 3!S come to mind here, announced as 10-12 total points, but 2- hearts and 3- diamonds? Will you easily get to your cold game in hearts with that bid? Should partner know that you have such nice hearts?

Later on, with nobody vulnerable, you as a bot pick up

8743
A74
2
KQT98

The bidding has gone

RHO you LHO Partner
2!S - pass - pass - 3!D
pass - pass - Dbl - pass
pass - ???

Why take the contract out into 3!H here, showing 4+ hearts and 9-11 points? Should partner know that you have a nice club suit? And, should partner, with a decent diamond suit and T32 in hearts, know that the poor 3-3 fit will not be better than some other place? For a heart bid to be any good here, partner must have a heart suit, coupled with a diamond overcall in the balancing seat. In that case of course, partner might have found some other initial bid.

It appears in both cases the bot programming has a serious glitch in it. The bots seem to be making pure point count responses to my bids. They find a bid that fails to describe anything about the hand but the point count value of the hand, whereas shape is completely ignored. Walrus bidding. While i accept that bots are bots, and judgement is a difficult thing to teach to a bot, but these seem to be recent changes, and downhill in terms of quality.

John
0

#2 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2012-April-07, 14:41


0

#3 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2012-April-07, 14:44


0

#4 User is offline   pigpenz 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 2,553
  • Joined: 2005-April-25

Posted 2012-April-07, 17:36

looks like 4on first should go down
and on second hand best chance for a contract is 3NT.
0

#5 User is offline   cloa513 

  • PipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 1,529
  • Joined: 2008-December-02

Posted 2012-April-07, 17:58

View Postpigpenz, on 2012-April-07, 17:36, said:

looks like 4on first should go down
and on second hand best chance for a contract is 3NT.

True but you'd have to agree 4 is a reasonable guess hopefully partner is short is spades rather than clubs.
0

#6 User is offline   pigpenz 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 2,553
  • Joined: 2005-April-25

Posted 2012-April-07, 19:22

View Postcloa513, on 2012-April-07, 17:58, said:

True but you'd have to agree 4 is a reasonable guess hopefully partner is short is spades rather than clubs.

I was just referring to his comment that 4 is cold....it can be set but requires the lead continuation and ruff...which usually doesnt always happen with GIB as your partner.
0

#7 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,594
  • Joined: 2004-August-21
  • Gender:Male

Posted 2012-April-08, 10:33

Hand 1. Taking account of shape is hard to do in bidding rules, since there are too many possibilities to list them explicitly. That's why we have simulations -- they try to emulate human judgement. If West had passed, the simulations would have gotten it to bid 4. However, the 3 bid changed the hands in the simulations, and in some cases it decided that 3 was better; in my 11 tests, it chose 3 2 times, 3 5 times, and 4 4 times. In its simulations, it didn't expect 3 to be passed out.

Hand 2. I haven't been able to reproduce this in my tests (it's hard because the random number sequences are different depending on whether you're testing one bot or 3). The book bid is Pass, so the blame falls again on simulations: it presumably guessed that your hand was likely to have some good hearts because you can't have many spades and your diamonds are not robust. In fact, in 2 of my test runs, the bot bid 3 even BEFORE getting doubled.

Simulations aren't perfect -- we don't simulate enough hands to avoid occasional glitches due to skewed deals. But in the absence of real judgement, it's the best we can do.

P.S. It's easier if you start different threads for unrelated issues. It gets confusing when discussions of different issues are interleaved.

#8 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2012-April-08, 10:55

Same hand, somewhat different auction:
Why should the 3 bid show exactly 12 total points, as the description indicates?
0

#9 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2012-April-08, 11:01

View Postbarmar, on 2012-April-08, 10:33, said:

In its simulations, it didn't expect 3 to be passed out.

At the (few) other tables where South bid 2NT and North responded 3, South raised to 4. Is this really what GIB's simulations "wanted" to happen? When he bids 3, does GIB think he's investigating slam? If not, why doesn't he just bid 4?
0

#10 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,594
  • Joined: 2004-August-21
  • Gender:Male

Posted 2012-April-08, 11:47

In its simulations, the only time it ended up in 4 was when South had 3=5=5=0 distribution; in those cases, East bid 4 and South supported . Most of the auctions in the simulations ended in 4.

When a bid comes from simulations, there's no "reason" for it -- it's not trying to do anything. It just finds that when it makes the bid, something better happens as a result.

#11 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2012-April-08, 12:13

View Postbarmar, on 2012-April-08, 11:47, said:

When a bid comes from simulations, there's no "reason" for it -- it's not trying to do anything. It just finds that when it makes the bid, something better happens as a result.

Which makes GIB a successful "masterminder" but a difficult partner to bid with. KISS when possible...
0

#12 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,594
  • Joined: 2004-August-21
  • Gender:Male

Posted 2012-April-08, 13:15

View PostBbradley62, on 2012-April-08, 12:13, said:

Which makes GIB a successful "masterminder" but a difficult partner to bid with. KISS when possible...

Understood. But we're between a rock and a hard place. We can't write rules for everything, so we need simulations. But sometimes the simulations do weird things (just like human partners).

In cases where we see simulations often screwing up, we'll fill in with more rules. But the more complicated the auction gets, the harder it is to write rules to match it -- there are simply too many possibilities.

#13 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2012-April-08, 13:46

View Postbarmar, on 2012-April-08, 11:47, said:

View PostBbradley62, on 2012-April-08, 11:01, said:

At the (few) other tables where South bid 2NT and North responded 3, South raised to 4. Is this really what GIB's simulations "wanted" to happen? When he bids 3, does GIB think he's investigating slam? If not, why doesn't he just bid 4?
In its simulations, the only time it ended up in 4 was when South had 3=5=5=0 distribution; in those cases, East bid 4 and South supported . Most of the auctions in the simulations ended in 4.

When a bid comes from simulations, there's no "reason" for it -- it's not trying to do anything. It just finds that when it makes the bid, something better happens as a result.



View Postbarmar, on 2012-April-08, 12:01, said:

Remember the old rule: support with support.
Seems right here...
0

#14 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,594
  • Joined: 2004-August-21
  • Gender:Male

Posted 2012-April-08, 14:05

As I said, when simulations are involved, it sometimes does weird things. GIB doesn't know about general rules -- they're used by the people writing the bidding DB.

And I don't think you want us disabling simulations here. The book bid is 3, it takes simulations to figure out that forcing to 4 is better.

#15 User is offline   Bbradley62 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 6,542
  • Joined: 2010-February-01
  • Gender:Male
  • Location:Brooklyn, NY, USA

Posted 2012-April-08, 18:21

View Postbarmar, on 2012-April-08, 14:05, said:

And I don't think you want us disabling simulations here. The book bid is 3, it takes simulations to figure out that forcing to 4 is better.

What does that 3 show?
0

#16 User is offline   pigpenz 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 2,553
  • Joined: 2005-April-25

Posted 2012-April-08, 18:57

so when GIB does simulations....is it changing the possibities of what the other 3 hands are?

for south I would make it 5-5 reds in the south, with possibility of 6-5
and from bidding it seems E-W are 5-5 from bidding in clubs.
question would be how strong can south be for 2NT call?

I'll probably plug some numbers in DMPro just to see
0

#17 User is offline   barmar 

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • Group: Admin
  • Posts: 21,594
  • Joined: 2004-August-21
  • Gender:Male

Posted 2012-April-09, 15:56

View Postpigpenz, on 2012-April-08, 18:57, said:

so when GIB does simulations....is it changing the possibities of what the other 3 hands are?

No. Here's roughly how simulations work during the bidding:

1. Find the book bid for the hand it has.
2. Adjust the hand in various ways (adding/subtracting a few points, swapping cards between suits), and for each adjusted hand find the book bid.
3. Deal a bunch of hands, order them by how well they match the bidding so far, use the first N.
4. For each of those hands, simulate what will happen with each of the bids from steps 1 and 2. This simulation uses the book bids for the rest of the auction (allowing further simulations results in exponential explosion), and then does double-dummy analysis of the play.
5. It estimates the MP or IMP score of each result (I'm not sure how it figures this out). The book bid is given a small bonus, so it should be preferred in close cases.
6. It chooses the bid with the best average.

#18 User is offline   pigpenz 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 2,553
  • Joined: 2005-April-25

Posted 2012-April-09, 16:21

View Postbarmar, on 2012-April-09, 15:56, said:

No. Here's roughly how simulations work during the bidding:

1. Find the book bid for the hand it has.
2. Adjust the hand in various ways (adding/subtracting a few points, swapping cards between suits), and for each adjusted hand find the book bid.
3. Deal a bunch of hands, order them by how well they match the bidding so far, use the first N.
4. For each of those hands, simulate what will happen with each of the bids from steps 1 and 2. This simulation uses the book bids for the rest of the auction (allowing further simulations results in exponential explosion), and then does double-dummy analysis of the play.
5. It estimates the MP or IMP score of each result (I'm not sure how it figures this out). The book bid is given a small bonus, so it should be preferred in close cases.
6. It chooses the bid with the best average.

I was curious...looking at the hands I generated I couldnt see what benefit bidding 3
has......playing 4 from N or S was pretty equal 64%N 66%S....so bidding 3 instead of 4 seems pretty pointless.
1

#19 User is offline   wuudturner 

  • PipPipPip
  • Group: Full Members
  • Posts: 81
  • Joined: 2011-November-07

Posted 2012-April-09, 17:25

View PostBbradley62, on 2012-April-08, 11:01, said:

At the (few) other tables where South bid 2NT and North responded 3, South raised to 4. Is this really what GIB's simulations "wanted" to happen? When he bids 3, does GIB think he's investigating slam? If not, why doesn't he just bid 4?


I was thinking this, that GIB expected more bidding, that 3!S would not be passed, but given the description - that 3!S shows no fit in a red suit, it will surely not lead to a constructive auction. So my point is, the notes are poor - that 3!S explicitly denies a red suit fit. Had the note stated that 3!S is merely forward going, showing say 5+ spades and 10+ total points, then partner will logically bid again. But when you are known NOT to have a fit in a red suit, logic says to pass, to stop in what appears to be a misfit.

I'll acknowledge that 4!H need not be cold, even so, 4!H is a contract I would very much like to be in, and I see no reason why GIB would search for a 5-3 spade fit when a 5-4 heart fit is known!
0

#20 User is offline   wuudturner 

  • PipPipPip
  • Group: Full Members
  • Posts: 81
  • Joined: 2011-November-07

Posted 2012-April-09, 18:36

View Postpigpenz, on 2012-April-09, 16:21, said:

I was curious...looking at the hands I generated I couldnt see what benefit bidding 3
has......playing 4 from N or S was pretty equal 64%N 66%S....so bidding 3 instead of 4 seems pretty pointless.


I suppose that 3!S caters to the hands where South has 5550 shape, or where South has bid 2NT on 5450 shape.

Yes, admitttedly, any of these shapes will be rare, but a careful bidder does best to look for those cases.

I guess to be serious, my point is there are NO circumstances where bidding 3!S gets us to a good contract, as we can never have a better spade fit than we do the known heart fit. I DO understand the concept of simulation, having written a bridge simulator myself. This is what confuses me, since 3!S will not improve the contract, especially since 3!S lies about the bot's holdings in the red suits!
0

Page 1 of 1
  • You cannot start a new topic
  • You cannot reply to this topic

2 User(s) are reading this topic
0 members, 2 guests, 0 anonymous users