BBO Discussion Forums: what went wrong? - BBO Discussion Forums

Jump to content

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

what went wrong? bot loses its tiny mind

#1 User is offline   jammen 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 119
  • Joined: 2007-November-16

Posted 2023-January-13, 12:12



2 spades was forcing showing 4 spades and 5 hearts. All other bids were natural except for 4 diamonds which showed 4-5 clubs, 4 diamonds, 2 hearts, 2-3 spades. I corrected to 5 clubs and dummy tabled a 3-3-3-4 12 count.

Why didn't the bot pass 3nt? Why didn't the bot support hearts when I showed 4-5 in the majors? Who writes these bid descriptions?
0

#2 User is online   smerriman 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 3,721
  • Joined: 2014-March-15
  • Gender:Male

Posted 2023-January-13, 13:37

Wrong forum.

But to answer your questions:

- the bidding database tells North to bid 4 over 2. It's completely understandable, however, that a simulation of possible hands you might hold may tell it that 2NT may work out better, and in this case it did conclude that.

- having chosen 2NT, the bidding database tells North to pass 3NT. Also understandable, and was the point of bidding 2NT in the first place.

However, when GIB ran a second simulation over hands where you'd bid 3NT, it's also understandable that it might now conclude 4 is the correct contract, given this uses a completely different set of hands than the first simulation.

The twist is that the bidding database also says that if it bids 4, you'll respond with 4 on every single possible hand you might hold. So it just treats this as another way of getting to 4, and it never sees it costing anything.

In both cases, therefore, there was nothing wrong with the bid descriptions; GIB is allowed to pretend it has something that it doesn't if it concludes that following the bid descriptions from that point onwards will lead to a better final outcome.

But obviously still a silly bid, just not one the robot can see could ever go wrong.
0

#3 User is offline   jammen 

  • PipPipPipPip
  • Group: Full Members
  • Posts: 119
  • Joined: 2007-November-16

Posted 2023-January-13, 14:00

View Postsmerriman, on 2023-January-13, 13:37, said:

Wrong forum.

But to answer your questions:

- the bidding database tells North to bid 4 over 2. It's completely understandable, however, that a simulation of possible hands you might hold may tell it that 2NT may work out better, and in this case it did conclude that.

- having chosen 2NT, the bidding database tells North to pass 3NT. Also understandable, and was the point of bidding 2NT in the first place.

However, when GIB ran a second simulation over hands where you'd bid 3NT, it's also understandable that it might now conclude 4 is the correct contract, given this uses a completely different set of hands than the first simulation.

The twist is that the bidding database also says that if it bids 4, you'll respond with 4 on every single possible hand you might hold. So it just treats this as another way of getting to 4, and it never sees it costing anything.

In both cases, therefore, there was nothing wrong with the bid descriptions; GIB is allowed to pretend it has something that it doesn't if it concludes that following the bid descriptions from that point onwards will lead to a better final outcome.

But obviously still a silly bid, just not one the robot can see could ever go wrong.


That makes as little sense as the bots bidding.
0

#4 User is offline   harikannan 

  • PipPipPip
  • Group: Full Members
  • Posts: 54
  • Joined: 2021-June-19

Posted 2023-January-14, 00:18

NMF is played by gib.
0

#5 User is offline   harikannan 

  • PipPipPip
  • Group: Full Members
  • Posts: 54
  • Joined: 2021-June-19

Posted 2023-January-14, 00:21

Is it possible that your not choosing new minor forcing would have led the bot to believe you had just 4 hearts.
0

#6 User is online   smerriman 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 3,721
  • Joined: 2014-March-15
  • Gender:Male

Posted 2023-January-14, 00:59

No, this sequence specifically shows 4-5, and GIB was aware of the 5-3 fit (see its logic above).

Agree it makes no *bridge* sense, but it makes 100% logical sense based on the algorithm GIB uses.
0

#7 User is offline   pescetom 

  • PipPipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 7,204
  • Joined: 2014-February-18
  • Gender:Male
  • Location:Italy

Posted 2023-January-15, 10:17

View Postsmerriman, on 2023-January-13, 13:37, said:

The twist is that the bidding database also says that if it bids 4, you'll respond with 4 on every single possible hand you might hold. So it just treats this as another way of getting to 4, and it never sees it costing anything.

In both cases, therefore, there was nothing wrong with the bid descriptions; GIB is allowed to pretend it has something that it doesn't if it concludes that following the bid descriptions from that point onwards will lead to a better final outcome.

But obviously still a silly bid, just not one the robot can see could ever go wrong.

I'm still struggling to see why it set itself this last problem rather than bid 4 when the simulation suggested that contract.
The bid description of 4 looks ok, but the hand does not match, so why consider it or at least consider it equal to 4-pass at equal probability of success?
Did it also consider 4 and exclude that only because he will not always respond 4?
0

#8 User is online   smerriman 

  • PipPipPipPipPipPipPip
  • Group: Advanced Members
  • Posts: 3,721
  • Joined: 2014-March-15
  • Gender:Male

Posted 2023-January-15, 18:47

View Postpescetom, on 2023-January-15, 10:17, said:

I'm still struggling to see why it set itself this last problem rather than bid 4 when the simulation suggested that contract.
The bid description of 4 looks ok, but the hand does not match, so why consider it or at least consider it equal to 4-pass at equal probability of success?

How do you know it didn't consider it equal :) The description of 4 doesn't match the hand either, so it doesn't know this is "closer to real bridge" than 4. It just generates a list of candidate bids for the next bid, and then calculates a score for each of them by extrapolating with the database. If two options give identical scores, I think it literally just picks the one that happened to be tested first.

If you wanted logic like "why didn't it bid 4 when the simulation suggested that contract", you could perhaps break ties by "whichever bid resulted in a smaller number of extrapolated bids before the end of the auction". But if anything, in most cases you'd want the exact opposite; e.g. GIB typically blasts slam because its database can't extrapolate to grand, whereas if it had bid slowly you'd get there when it gets around to simulating with more information later.

(Incidentally, there is a very weird issue which *did* result in it marginally preferring 4, so they weren't strictly tied, even though all led to a 4 contract. I noticed this months ago and have been trying to reverse engineer the logic, but haven't figured it out yet. Might be a straight out bug, but I'm not sure yet. But even if fixed it wouldn't impact this situation, since they'd still be tied.)
0

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

1 User(s) are reading this topic
0 members, 1 guests, 0 anonymous users