Now for GIB 4♣ is natural, but as it does not deny the 5-card ♥, I really don't get the Pass, but I assume it is because BEN never had been trained with that sequence.
Ben on BBO - Feedback thread Discussions about Ben models trained on BBO data
#161
Posted 2024-August-07, 15:53
Now for GIB 4♣ is natural, but as it does not deny the 5-card ♥, I really don't get the Pass, but I assume it is because BEN never had been trained with that sequence.
#162
Posted 2024-August-13, 07:56
23 players saw BEN bid the very strange 2♥
One lucky player got the more natural 2♠ and scored 100%.
Also worth noting the failure to cover ♠9 at trick 7.
#163
Posted 2024-August-15, 09:25
Just to clarify: I am using the Open Source version of BEN as participant in the BEN and Freinds daily, so there are 4 BEN's but one (south) is not controlled by BBO
e
Down 1 and a fine score of 64%
Using the new bidding model all was the same until trick 6, where BEN suddenly found a much better plan.
Making an overtrick and a score of 1.2%
Please remove the randomness in the BBO-BEN
#164
Posted 2024-August-20, 10:16
But this is interesting, because BEN haven't been trained in the default GIB system
Explanation is correct, but BBO-BEN just passed.
And yes, the south hand is not good enough for 4♥, but I will discuss that with the open source version of BEN.
#167
Posted 2024-October-02, 01:43
#168
Posted 2024-October-04, 14:20
mstockyard, on 2024-September-22, 08:30, said:
If you are interested in playing with the Open Source version of BEN (Not the BBO version) I might be able to set it up.
As a start feel free to challenge the Robot on BBO (Name is TwoOneBEN). In a challenge you will play 3 GIB-robots, and the compare with your opponent. Then you can get an impression of the strength and especially weaknesses of the robot.
#169
Posted 2024-October-09, 23:48
#170
Posted 2024-October-25, 00:38
#171
Posted 2024-October-25, 01:47
pilowsky, on 2024-October-25, 00:38, said:
Very unlucky hand where ♣A was in the wrong hand. And very unsporting of you to double.
#172
Posted 2024-October-25, 02:03
(I'm still trying to figure out how I scored more than 50% on this hand, though )
#173
Posted 2024-November-05, 15:26
pilowsky, on 2024-March-02, 20:37, said:
thorvald, on 2024-April-21, 16:34, said:
TwoOneBEN, on 2024-May-31, 09:31, said:
TwoOneBEN, on 2024-June-04, 05:43, said:
TwoOneBEN, on 2024-June-26, 06:49, said:
TwoOneBEN, on 2024-August-13, 07:56, said:
And a couple of days ago, a 1.7% board for me because the defense differed at my table only.
It seems Ben and lorserker have met the same fate as GIB and jdonn
#175
Posted 2024-November-19, 05:58
If you run into oddities, please raise them.
#176
Posted 2024-November-22, 02:01
I've bid to what I assume will be the final contract, but instead of passing out the Ben sitting East hangs. After a bit the table closes with an "An error has occurred...GIB Ask Failed: ERROR(2): Robot Error" message. Reopening the table leads to the same issue.
#177
Posted 2024-November-22, 06:16
kcostell, on 2024-November-22, 02:01, said:
I've bid to what I assume will be the final contract, but instead of passing out the Ben sitting East hangs. After a bit the table closes with an "An error has occurred...GIB Ask Failed: ERROR(2): Robot Error" message. Reopening the table leads to the same issue.
The goulash game was mis-configured, sorry about that. We removed it and we'll post it again tomorrow.
Thanks for letting us know.
#179
Posted 2024-November-27, 22:00
That is, as I believe I touched on in an early page of this thread, the most crucial aspect of a human playing with a robot is that it's consistent and understandable. Given it has learnt everything from neural networks and differs from GIB in some areas, I'm still no closer to figuring out when a bid for Ben means exactly what it does for GIB and when it doesn't, implications from when it leads a particular card (e.g. its choice from AKx), whether it signals, how it will interpret cards I play, etc.
#180
Posted 2024-November-28, 11:47
smerriman, on 2024-November-27, 22:00, said:
Just a little. The only thing i changed is that Ben will now consistently make the same move in identical situations. Because many people gave the feedback that it should do that.
I will try to do more in the future.
Defensive carding is an area which wasn't addressed at all yet.
Bidding system - it is identical to GIB, but different bids can be made in unclear situations (decisions). Rarely, a sequence can come up where it really doesn't know how to bid and makes up something (there are examples in this thread). I think that in the future Ben will have its own explanations instead of leaning on GIB to explain its bids.
About predictability and consistency I am a bit biased. I like a little randomness, because even when I am playing with my regular partner I am not 100% sure that he has what I think he has and that he would always do the same action. I think it is normal to take some reserve when you play. The robotic consistency which allows you to make inferences which are totally certain, are a bit unappealing to me.
Although I have some starting points how I may optimize Ben for humans, future directions will depend on feedback, because the whole topic is subjective and unclear what exactly the goal is.