[pilowsky]

lorserker, on 2024-January-26, 03:32, said:

4S is a bit creative, but I don't consider it a blunder.
5C does have equity, and you may give some info on the way.
would you prefer if N passed 4C? or if N bid 5C directly?

Fair point.
I guess that means that 4C indicated more than the South hand was worth and I should have left 3D alone?

[smerriman]

pilowsky, on 2024-January-26, 14:32, said:

Fair point.
I guess that means that 4C indicated more than the South hand was worth and I should have left 3D alone?

The bidding seems fine; the descriptions are very poor but that's just inherited from GIB. After trick 4, cross to the king of trumps, ruff another spade, and then claim 11 tricks.

[0 carbon]

I don't see how Ben could bid using the average of 10m deals, with many different systems around the world.
Ben should use the announcements and alerts of systems and see what worked within each system and for each convention.

Then Ben should announce a system, or agree a system, and have alerts visible to P for all P's bids.
Develop a BBOalert file for the alerts.

[lorserker]

0 carbon, on 2024-January-26, 23:12, said:

I don't see how Ben could bid using the average of 10m deals, with many different systems around the world.
Ben should use the announcements and alerts of systems and see what worked within each system and for each convention.

Then Ben should announce a system, or agree a system, and have alerts visible to P for all P's bids.
Develop a BBOalert file for the alerts.

For consistency, we have trained Ben on the GIB system using deals bid by GIB.
The human deals from ACBL where only used to train card play.

Training bidding on human deals (with diverse bidding systems occurring in the data) is also interesting, and we have tried that too. What it converged to was a basic 5-card major system with strong NT (something like sayc).

[etha]

so Ben failed to learn anything because the stupid human had changed the wrong file. Just watched about 100 hands quickly and only saw one RKCB error I think. So indeed it seems to be learning. I'll actually play some hands slowly with it later and check my visual watching of very fast hands stupid human may be wrong again.

Yes looks good not seen anymore errors but hands are going past very fast.

I did wonder why it didn't seem capable of learning this.


Added bonus is it is now 0.75 imps/bd better on RKCB hands :-).


There is a downside thought to throwing a load of RKCB hands into the training of a normal net, it will like to bid 4NT more than it should I think as a result.

[smerriman]

---

West Ben's hand would have been a good example to post in the recent thread about upgrading. Quad tens and trip aces is a clear upgrade (18.4 K&R), so you could say Ben is good at evaluating its hand.

At least, until you see Ben East's decision to downgrade its hand by goodness knows how many points, when you realise it must have just been a fluke.

[pilowsky]

Ben makes a 5th down lead into 3NT.
I think GIB is programmed to make 4th down leads from 4+.
Does it have any lead rules we should know about?

[pescetom]

pilowsky, on 2024-February-04, 18:56, said:

Ben makes a 5th down lead into 3NT.
I think GIB is programmed to make 4th down leads from 4+.
Does it have any lead rules we should know about?

I imagine the problem is just that Ben was fed hands where players do not lead 4th down, either because they play 3/5 and did not read or care about the GiB agreements or because they had no idea they should agree such things in the first place.

[pilowsky]

pescetom, on 2024-February-06, 13:36, said:

I imagine the problem is just that Ben was fed hands where players do not lead 4th down, either because they play 3/5 and did not read or care about the GiB agreements or because they had no idea they should agree such things in the first place.

I don't think 'unprogrammed' GIB cares either - I thought I read in an earlier thread that it was something that had to be hard-wired.
My expert knowledge of neural network programming comes from watching codebullet videos where it takes hundreds of thousands of trials to get a good model.
So I'm wondering what the 'targets' set for Ben are.

[smerriman]

pilowsky, on 2024-February-06, 15:25, said:

I don't think 'unprogrammed' GIB cares either - I thought I read in an earlier thread that it was something that had to be hard-wired.

GIB gives a small bonus to standard lead conventions after calculating double dummy averages, so if simulations show everything else is equal between 4th and 5th (or others), it will prefer 4th. Except where it's programmed to do random things for no reason a small percentage of the time.

[helene_t]

Ben passes a reverse after a freebid https://www.bridgeba...HJ|pc|DJ|pc|S9|

[tsankaR]

https://www.bridgeba...DJ%7Cmc%7C13%7C

2nt passed!

[lorserker]

tsankaR, on 2024-February-12, 10:17, said:

https://www.bridgeba...DJ%7Cmc%7C13%7C

2nt passed!

thanks for reporting. i have answered you in the other thread

[helene_t]

https://www.bridgeba...CQ|pc|ST|pc|SJ|

4c was explained as natural but i think it was meant as gerber. By the way, why is tthere no transfers to minors after 2nt? Bu ok thars a gib issue.

Either way, the 6n bid is weird

[pescetom]

Quote

4c was explained as natural but i think it was meant as gerber.
.....

Either way, the 6n bid is weird

The auction is terrible even if it was Gerber (surely a rebid of 3♥ would be forcing with slam interest?).


Does Ben know anything about lead conventions? Finessing for the 9 looks reasonable to me (unless the lead came from T4 and N made a genial falsecard I guess).

[Gunnar60]

Hi,

I have started to make videos of Ben & Friends Daily. The video shows the winners play.

The intention is to make it possible for interested that have not access to Ben on BBO to study Ben's play.

I have page on my Site Computerbridge.se with info about Ben. The videos will accessable there. The two first videos: youtu.be/eAO35KfRIeE - youtu.be/eAO35KfRIeE

Link to Ben's page: computerbridge.se/ben/

[Gunnar60]

I will continue to make videos of Ben & Friends every day for time to come... When they are ready I will upload them to my YouTube channel computerbridge.se.

If you are interested in more videos, subscribe to computerbridge.se in YouTube and you will see the new one's when you go into YouTube.
I will not publish more links to Ben & Friends plays here and the Ben's Page will not be updated every day

Anyway here is the link to the 02/19 play, with develin as winner: https://youtu.be/7Ltowcxizgo

Have you any suggestion for improvements, please let me know...

I am working on following improvements:

- Get the score shown in play view, which now is buggy.
- Show the explanations of the bids

Regards
Bjorn

[Gunnar60]

I have updated https://www.computerbridge.se/ben/ with the threes latest Ben & Friends Daily.
I will make three more.

[smerriman]

I guess the workaround of Ben being bypassed when responding to Blackwood means it still doesn't have a clue what it's actually for..

---

5♠ was a sign-off of course, with 6♥: Cue bid -- 1+ ♣; 1+ ♦; 3- ♥; 21- HCP; twice rebiddable ♠; no ♣A; no ♦A; ♥A; 15-22 total points; forcing.

[pilowsky]

Ben breaking bad?
My board.

39 other boards!.