Another flaw in the bidding database

[thorvald]

[hv=pc=n&n=s7hj52daj64ck8432&d=e&v=e&b=6&a=1s2n(5-5%20minors)4s?]133|200[/hv]

 

A simulation would probably solve this, but for the not advanced bots, I think there is an error in the bidding database, when passing on this hand

[thorvald]

My opponent did handle the error :-)

 

[hv=url=https://www.bridgebase.com/tools/handviewer.html?bbo=y&lin=pn%7CHuman%2CRobot%2CRobot%2CRobot%7Cst%7C%7Cmd%7C4SK8H6DKQT83CAJ975%2CSAQT62HT843D95CQT%2CS7HJ52DAJ64CK8432%2CSJ9543HAKQ97D72C6%7Csv%7CE%7Cah%7CBoard%206%7Cmb%7C1S%7Can%7CMajor%20suit%20opening%20--%205%2B%20%21S%3B%2011-21%20HCP%3B%2012-22%20total%20points%7Cmb%7C2N%21%7Can%7CUnusual%20notrump%20overcall%20--%205%2B%20%21C%3B%205%2B%20%21D%3B%2012%2B%20total%20points%3B%20forcing%7Cmb%7C4S%7Can%7CPreemptive%20raise%20--%204%2B%20%21S%3B%209-%20HCP%3B%206%2B%20total%20points%7Cmb%7CP%7Cmb%7CP%7Cmb%7C4N%7Can%7C5%2B%20%21C%3B%205%2B%20%21D%3B%2018%2B%20total%20points%7Cmb%7CP%7Cmb%7C5C%7Can%7C3%2B%20%21C%7Cmb%7CP%7Cmb%7CP%7Cmb%7CP%7Cpc%7CHK%7Cpc%7CH6%7Cpc%7CH3%7Cpc%7CH2%7Cpc%7CS9%7Cpc%7CSK%7Cpc%7CSA%7Cpc%7CS7%7Cpc%7CH4%7Cpc%7CHJ%7Cpc%7CHA%7Cpc%7CC5%7Cpc%7CC7%7Cpc%7CCT%7Cpc%7CCK%7Cpc%7CC6%7Cmc%7C11%7C]399|300[/hv]

[smerriman]

It's not a flaw in the bidding database. The bidding database needs game-level points to bid game. Other than a few basic situations like 1M-4M or raise of a preempt which use the law, sacrifices are simulations only. Which makes total sense, since you can't judge whether to sacrifice based on fixed rules about combined HCP, etc.

[thorvald]

It's not a flaw in the bidding database. The bidding database needs game-level points to bid game. Other than a few basic situations like 1M-4M or raise of a preempt which use the law, sacrifices are simulations only. Which makes total sense, since you can't judge whether to sacrifice based on fixed rules about combined HCP, etc.

 

I think you are wrong here. It is possible to define hands, where you should bid 4N or 5Mi based on hcp and length in the minor. Then you can add simulations for the advanced bots on the more freakish hands.

 

Just like you can define hands, that would double for penalty here.

 

Principle for needing game-level points is fine for generic rules.

 

Just to clarify: There should never be a need for simulation for the first bid of a hand (like here)

[thorvald]

Try giving the bot all 13 ♠, and the rules will make the book bid 4♠ in any position

[smerriman]

It's certainly *possible*. But what would be the benefit of doing so? I can't see how it'd improve the advanced robot at all, which is the only important one. Maybe I'm missing something?

 

Edit - in fact, looking at this again, I can't even see how it would *affect* the advanced bot, let alone improve it. There aren't earlier simulations involved, so it's not going to influence auction projections at earlier stages. The only tiny thing I can see is that it may change what happens if the opponents compete further to 5♠, but you really shouldn't be thinking of walking the dog and competing to 5 then 6 on two separate bids.

 

I think we're on completely different wavelengths here, so it would be interesting to hear how you can see redefining a 5♣ bid would actually change anything.

[thorvald]

In the free daylong today I got this board

 

[hv=url=https://www.bridgebase.com/tools/handviewer.html?lin=st||pn|Thorvald,~~v3fakebot,~~v3fakebot,~~v3fakebot|md|4SAK9863HA74DAT3CT,SHKQTDQ7642CA9752,SQJT74H98532DK8CK,S52HJ6DJ95CQJ8643|sv|n|rh||ah|Board%202|mb|P|mb|1S|an|Major%20suit%20opening%20--%205+%20!S;%2011-21%20HCP;%2012-22%20total%20points|mb|2N!|an|Unusual%20notrump%20overcall%20--%205+%20!C;%205+%20!D;%2012+%20total%20points;%20forcing|mb|4S|an|Preemptive%20raise%20--%204+%20!S;%209-%20HCP;%206+%20total%20points|mb|5C|an|3+%20!C;%2011-%20HCP;%2012-%20total%20points|mb|5S|an|5+%20!S;%2011-21%20HCP;%2012-22%20total%20points|mb|P|mb|P|mb|P|pc|D4|pc|D8|pc|DJ|pc|DA|pc|S3|pc|C2|pc|SQ|pc|S2|pc|DK|pc|D5|pc|D3|pc|D7|pc|S4|pc|S5|pc|SA|pc|C5|pc|DT|pc|DQ|pc|ST|pc|D9|pc|H2|pc|H6|pc|HA|pc|HT|pc|CT|pc|CA|pc|CK|pc|C8|pc|HK|pc|H3|pc|HJ|pc|H4|pc|HQ|pc|H5|pc|C6|pc|H7|pc|C9|pc|S7|pc|C4|pc|S6|mc|10|]399|300[/hv]

 

The first 3 bids are the same as the board I started this thread with, but here the bot bids 5♣, and with the explanation 3+♣, 11- HCP, 12- total points.

 

The explanation does match the original hand, so I assume there could be a difference in the explanation and the actual rule for selecting 5♣

 

Looking at the rules in the free bot there are the following for bidding 5♣

 

 

database search: 1S.2N.4S.5C -> 6

-- 5C..4 .S... ..... ..... ...... 1025 (59): ~b==#1~@05~.>24-#0(2,13)~@09+~((h=.)|1)~#g~#28(b)-h>a+e-2~

-- 5C..4 .S... 8.... ..... ...... 1034 (58): ~b==#1&(b)+k>a+e-2+13+#3-#13~

-- 5C..4 .S... ..... ..... ...... 292 (44): ~b==#1~@05~.+#0(2,13)[#17(5),30]~

-- 5C..4 .S... ..... ..... ...... 295 (30): ~b==#1~@05~.+#0(2,13)>=#17(5)~

-- 5C..4 .S... ..... ..... ...... 297 (5): @05~.+#0(2,13)>=#17(5)~@09+~.>6-#0(2,b)~#b

-- 5C..4 .S... ..... ..... ...... 296 (4): @05~.+#0(2,13)>=#17(5)~@09+~.>6-#0(2,b)~#b

-- 3+ C

 

and the bot will bid 5♣ with both hands, but the BOT at BBO will only bid 5♣ with the last hand

[smerriman]

Ah, you're right, I make a mistake earlier - the rules you listed from the old bot are that you require a strong hand to bid 5♣, so it has P as the book bid for both hands for that reason. They must have changed this on BBO as 5♣ now just shows 3+ ♣. (The rest of the description is just whether you're a passed hand or not.)

 

But regardless, everything appears to be working perfectly, as the advanced bot will bid 5♣ with both hands (as does the old bot, via simulations).

 

If there was one flaw in old GIB, it might be that on very rare occasions it would project a non-passed hand 5♣ being raised to 6. But that's no longer the case on BBO.

 

So still not seeing any flaws, and challenge still remains to suggest why changing the rules would improve the bot here :)

[thorvald]

Well, I hope we can agree it is an error passing on the hand given

 

Then feel free to judge where the error is:

 

In the bidding database

Missing simulation

BBO offering an bot worse than free GIB released 10 years ago

My expectations that the bot is better than a beginner

other

 

It is not important where the error is, just someone would fix it :-)

[smerriman]

The error is that you're playing with free bots. It's fixed by playing with the paid bots. You're the one that needs to fix it, not someone else :)

 

I think you're still totally misunderstanding what the different versions of the robots are. The 10 year old version is the equivalent of today's paid bot, just with an older database. It is NOT the equivalent of today's free bot. The 10 year old version is thus significantly better than today's free bot.

 

To create the free bot, BBO took the 10 year old bot, then intentionally made it worse in three different ways:

 

- crippled it with the -z flag to ensure it *never* runs a simulation during bidding

- reduced the number of simulations to a low number with the -m flag, so it's much more prone to coming up with poor conclusions

- disabled the single-dummy GIBson algorithm (no flag for doing so, must have been done later) which kicks in after a couple of tricks, so the declarer play will always mess things up like delaying guesses rather than make a proper plan

 

This was to ensure it takes up very little computational resources, but makes for a very poor robot. It's free for a reason, and there's no reason to fix things that were only broken because all of its features were turned off.

 

Running the 10 year old robot under these conditions, it passes both hands, and never bids 5♣.

 

Does that help explain why these types of issues aren't actual bugs / flaws in the proper version of the robot?

[thorvald]

The error is that you're playing with free bots. It's fixed by playing with the paid bots. You're the one that needs to fix it, not someone else :)

 

I think you're still totally misunderstanding what the different versions of the robots are. The 10 year old version is the equivalent of today's paid bot, just with an older database. It is NOT the equivalent of today's free bot. The 10 year old version is thus significantly better than today's free bot.

 

To create the free bot, BBO took the 10 year old bot, then intentionally made it worse in three different ways:

 

- crippled it with the -z flag to ensure it *never* runs a simulation during bidding

- reduced the number of simulations to a low number with the -m flag, so it's much more prone to coming up with poor conclusions

- disabled the single-dummy GIBson algorithm (no flag for doing so, must have been done later) which kicks in after a couple of tricks, so the declarer play will always mess things up like delaying guesses rather than make a proper plan

 

This was to ensure it takes up very little computational resources, but makes for a very poor robot. It's free for a reason, and there's no reason to fix things that were only broken because all of its features were turned off.

 

Running the 10 year old robot under these conditions, it passes both hands, and never bids 5♣.

 

Does that help explain why these types of issues aren't actual bugs / flaws in the proper version of the robot?

 

You are missing my point. The error is BBO by offering such a bad bot.

I understand the different versions, and also understand that BBO (hopefully not intentional) have made the freeware version worse by "improving" the bidding.

 

For me that is a bug - if not a software bug - then a marketing bug.

 

The two deals I have shown are both played with the no-pay BOT, and even in a no-pay version they should bid the same. If I disable simulation on the last board it will pass, but as I wrote it actually bid 5♣ in the free daylong, that as far as I know isn't using the advanced robots - but I might be wrong

 

My error is (still) having the energy to report these kind of errors, errors that basically is destroying the fun for thousands of players.

 

Back in 2019 the last time they released a new version Diana wrote

 

After almost a year of tweaking, we've just released a big patch to the robots. GIB version 40 has been released.

This is mostly bug fixes and more than 100 various tweaks and patches.

 

So I expect the effort was made to improve GIB

[smerriman]

So your problem actually has nothing to do with the database; you think BBO should remove the free robot entirely and only offer the paid robot?

 

Well, that's definitely a business decision then. The software is working as intended on this hand - it would be pretty silly for BBO to make database changes that affect free users only, in order to solve problems that already have (and I would say better) solutions in place for paid users. Even if they started making changes to GIB again, that would be the absolute lowest priority of all changes.

 

But as I told you months ago, GIB isn't going to be improved, and the only reason for posting here is if you find these discussions interesting from a theoretical point of view. Not worth spending energy on if you're expecting changes.

[thorvald]

You are right, I better stop posting.

 

At the todays Zenith daylong (Advanced robots) I saw so many errors, that should be corrected first.

 

Perhaps I will return when Argine goes public, there is at least some developers trying to improve it

[pescetom]

To create the free bot, BBO took the 10 year old bot, then intentionally made it worse in three different ways:

 

- crippled it with the -z flag to ensure it *never* runs a simulation during bidding

- reduced the number of simulations to a low number with the -m flag, so it's much more prone to coming up with poor conclusions

- disabled the single-dummy GIBson algorithm (no flag for doing so, must have been done later) which kicks in after a couple of tricks, so the declarer play will always mess things up like delaying guesses rather than make a proper plan

 

This was to ensure it takes up very little computational resources, but makes for a very poor robot. It's free for a reason, and there's no reason to fix things that were only broken because all of its features were turned off.

I find it hard to imagine that disabling GIBson would really save significant CPU time, in the order of simulations.

Maybe they just wanted it to be awful, so that the substantial failure to improve the rental robot would be less obvious :(

[smerriman]

GIBson is really very noticably slower than the straightforward double dummy algorithm.