Revised robot play

[edward5958]

The style of play for the robots in the ACBL tourneys seems to have changed rather dramatically lately in that they are now more arbitrary and difficult to figure out. Whether I'm right in that perception or not, my success rate has been cut by more than 50% as it is now difficult to even scratch whereas I used to finish high on a regular basis. It's very frustrating, to be sure, so much so that I am about to the point of giving it up altogether.

[Bbradley62]

Do you have a specific hand, or are you just venting because you're in a slump?

[edward5958]

Do you have a specific hand, or are you just venting because you're in a slump?

I'm venting, to be sure, but it's not a temporal thing. Since August the change has been most perceptible, IMO, and lately it seems worse than ever. It's difficult to quantify, but I believe there is merit to what I am saying and I was just wondering if anyone else was having the same or a similar impression.

[barmar]

The last change to the robots was in December, when we announced the v20 update.

 

http://www.bridgebase.com/forums/topic/49901-gib-version-20-released/

[Bbradley62]

... and I was just wondering if anyone else was having the same or a similar impression.

I do not have the same impression.

[Bbradley62]

The last change to the robots was in December, when we announced the v20 update.

 

http://www.bridgebase.com/forums/topic/49901-gib-version-20-released/

I'm not at all sure how this helps answer the question, but if you're going to give this you might as well also give:

V19, 10/18/2011: http://www.bridgebase.com/forums/topic/48870-gib-version-19-released/

V18, 9/5/2011: http://www.bridgebase.com/forums/topic/48872-gib-version-18-released/

V17, 6/7/2011: http://www.bridgebase.com/forums/topic/46239-gib-upgraded/

[barmar]

And note that these changes primarily affect bidding, not play. There's a small effect on play because it uses the bidding rules to determine what types of hands are likely, and this is used in the simulations during play and defense.

[pigpenz]

I feel that the robots are always a challenge but they do what they should do as long as you dont lead them astray....sure everyonce in awhile they create a clunker but so do humans

[edward5958]

I do not have the same impression.

Well, thank you for that enlightenment. But I still don't know if anyone else shares my impression, do I?

[edward5958]

I'm not at all sure how this helps answer the question, but if you're going to give this you might as well also give:

V19, 10/18/2011: http://www.bridgebase.com/forums/topic/48870-gib-version-19-released/

V18, 9/5/2011: http://www.bridgebase.com/forums/topic/48872-gib-version-18-released/

V17, 6/7/2011: http://www.bridgebase.com/forums/topic/46239-gib-upgraded/

So do these changes support what I am saying? Remember, I said it goes back to August when my results became perceptibly different even though I was then playing in the same manner I always had played when I was much more succesful.

[barmar]

So do these changes support what I am saying? Remember, I said it goes back to August when my results became perceptibly different even though I was then playing in the same manner I always had played when I was much more succesful.

Your original message didn't mention the time frame, so I assumed you were talking about a change noticed in the past few weeks. As I said, we haven't made any changes to the play. Changes to the bidding rules will affect its inferences about the unseen suits, that's it.

 

If you can't be specific about the manner of the change you perceive, it's hard for us to give a more useful answer. If your results have gotten worse, it could be because the robot opponents are playing better, or because your robot partner is playing worse.

[USViking]

I have also been in a pronounced Robot tournament slump lately. It got so demoralizing

that I have taken a one week+ break from them.

 

However, I think the poor results are due to an extended lapse in concentration on my part,

although I am not conscious of being distracted.

 

I have not noticed any more than usual questionable Robot decisions, and to be fair about

it almost all such that I post here turn out to have some reasonable explanation.

 

FWIW I have kept a spreadsheet record:

 

For 249 tournaments ending 2/15/12 I scored 51.48% in MP competition and +156.37 in IMP.

 

For 20 MP tournaments 2/16 through 2/26 the humiliating MP average was 45.77% with 16/20 below 50% and 17/20 below average in player score ranking. During the same stretch I played in 5 IMP tournaments with 3/5 negative results and 4/5 below average player rank.

 

PS-- now that I think about it I have noticed a few more than usual deals where GiB bids with 3hcp less than bidding box explanation (I wish it could be limited to a max 2hcp "lie"). That is not an excuse for the extent of my recent poor showing, though.

[Bbradley62]

But I still don't know if anyone else shares my impression, do I?

It's been 10 days since you asked; the thread has been viewed over 200 times, and no one has chimed in to say that they do share your impression.

[Bbradley62]

Your original message didn't mention the time frame, so I assumed you were talking about a change noticed in the past few weeks. As I said, we haven't made any changes to the play. Changes to the bidding rules will affect its inferences about the unseen suits, that's it.

 

If you can't be specific about the manner of the change you perceive, it's hard for us to give a more useful answer. If your results have gotten worse, it could be because the robot opponents are playing better, or because your robot partner is playing worse.

OP specified his six-month timeframe in post #3, two days before you made your first reply in which you "assumed [he was] talking about a change noticed in the past few weeks".

 

Also, it appears to me that OP is using the phrase "style of play" to mean "the way GIB plays the game of bridge", not necessarily specifying play of the hand as opposed to bidding.

[barmar]

I confess that I simply missed his reference to August in his second post.

 

Anyway, the updates are SUPPOSED to change the way it plays. Ideally for the better, but if you've gotten used to the old bugs then you may have to relearn (and get used to new bugs).

[edward5958]

It's been 10 days since you asked; the thread has been viewed over 200 times, and no one has chimed in to say that they do share your impression.

 

 

So, does that mean than noone shares it? Is it possible that some who do share it don't read this board, or that some just don't want to comment? All I really know for sure is that you disagree and that's your prerogative but it doesn't invalidate my impression.

[edward5958]

Your original message didn't mention the time frame, so I assumed you were talking about a change noticed in the past few weeks. As I said, we haven't made any changes to the play. Changes to the bidding rules will affect its inferences about the unseen suits, that's it.

 

If you can't be specific about the manner of the change you perceive, it's hard for us to give a more useful answer. If your results have gotten worse, it could be because the robot opponents are playing better, or because your robot partner is playing worse.

 

Is this possible? Do the robots predictably play that differently at different tables? If so, it could very well explain what I have been experiencing lately compared to what it was like before last August. Whatever it is, it just seems that I am consistently in the wrong contract now, or that my p and I don't defend properly etc. on a regular basis. All one has to do it look at my mp totals since August compared to before to see the results of what I am saying. And I am (was) playing the same number of tournaments.

[edward5958]

I confess that I simply missed his reference to August in his second post.

 

Anyway, the updates are SUPPOSED to change the way it plays. Ideally for the better, but if you've gotten used to the old bugs then you may have to relearn (and get used to new bugs).

I have always assumed that what I needed to do was just play Bridge and that the robots would do the same. Indeed, I know the bots have their idiosyncrasies, to say the least, but I never realized so much as I do now that it is not so much Bridge as it is figuring out robot machinations. In any event, I would never tolerate such inconsistencies and unpredictability in any live partner and I don't play to continue to waste time and money fooling around with these machines either.

[edward5958]

I have also been in a pronounced Robot tournament slump lately. It got so demoralizing

that I have taken a one week+ break from them.

 

However, I think the poor results are due to an extended lapse in concentration on my part,

although I am not conscious of being distracted.

 

I have not noticed any more than usual questionable Robot decisions, and to be fair about

it almost all such that I post here turn out to have some reasonable explanation.

 

FWIW I have kept a spreadsheet record:

 

For 249 tournaments ending 2/15/12 I scored 51.48% in MP competition and +156.37 in IMP.

 

For 20 MP tournaments 2/16 through 2/26 the humiliating MP average was 45.77% with 16/20 below 50% and 17/20 below average in player score ranking. During the same stretch I played in 5 IMP tournaments with 3/5 negative results and 4/5 below average player rank.

 

PS-- now that I think about it I have noticed a few more than usual deals where GiB bids with 3hcp less than bidding box explanation (I wish it could be limited to a max 2hcp "lie"). That is not an excuse for the extent of my recent poor showing, though.

Thanks for your post. I hope your problems get straightened out soon. Of course, it could be that you are beginning to experience the same thing I am talking about here and I would appreciate your letting me know if that turns out to be the case.

[barmar]

In the main bridge club, there can be quite a bit of variability in the robots. First, there are 3 kinds of robots:

 

1. Old BBO version. If you use the old, downloaded BBO, it runs the GIB program locally on your PC. The speed of your CPU is a factor, because GIB limits the amount of time it thinks on a hand, so slower computers will do less thinking. You can also configure the thinking settings.

 

2. New BBO, basic robots. These are the robots you get when rent for $1/week. They don't do any simulations during the auction, and use a less sophisticated play algorithm.

 

3. New BBO, advanced robots. These are the robots you get when you rent for $1/day. They simulate during the auction (when the bidding rules allow) and use a more advanced play algorithm after the first few tricks.

 

Also, the simulations used during bidding and play make use of randomly dealt hands, and the random numbers will be different at each table.

 

Things are less variable in robot tournaments. These all use advanced robots running on the BBO servers, and the same random hands in simulations (as long as the human bidding and play are identical).

[pigpenz]

In the main bridge club, there can be quite a bit of variability in the robots. First, there are 3 kinds of robots:

 

1. Old BBO version. If you use the old, downloaded BBO, it runs the GIB program locally on your PC. The speed of your CPU is a factor, because GIB limits the amount of time it thinks on a hand, so slower computers will do less thinking. You can also configure the thinking settings.

 

2. New BBO, basic robots. These are the robots you get when rent for $1/week. They don't do any simulations during the auction, and use a less sophisticated play algorithm.

 

3. New BBO, advanced robots. These are the robots you get when you rent for $1/day. They simulate during the auction (when the bidding rules allow) and use a more advanced play algorithm after the first few tricks.

 

Also, the simulations used during bidding and play make use of randomly dealt hands, and the random numbers will be different at each table.

 

Things are less variable in robot tournaments. These all use advanced robots running on the BBO servers, and the same random hands in simulations (as long as the human bidding and play are identical).

 

maybe there could be a way when looking at results to see what kind of GIB

is playing.....

say

RED-old

GREEN-new

BLUE-advanced

 

I tend to rent GIB by the month and just look at results like playing in a club or open game where there is always an odd result or two.

When playing in ACBL bot games the play is more reliable, but the bots are more advanced.....but some times when I go over results I wonder

why GIB does the things he does at other tables.

[edward5958]

I tried to play a couple of robot tournaments during the last few days and my impression has been confirmed and fortified: playing with these robots is not worth the time, money or effort any longer. The guy who apparently tweaks or manipulates these things is enormously sensitive and defensive about any criticism of his work (see posts above), and I guess the sorry state of these machines reflects his considerable, misguided ego. In any event, IMO he bots used to be much better before all of this "improvement" got under way, and it seems to me that it is an example of fixing something which is not broken.

[barmar]

The guy who apparently tweaks or manipulates these things is enormously sensitive and defensive about any criticism of his work (see posts above), and I guess the sorry state of these machines reflects his considerable, misguided ego.

He hasn't posted in this thread at all, so what are you talking about? I think you're making assumptions about who does what at BBO.

fixing something which is not broken

Huh? Do you read the GIB forum? People report problems every single day. We're not supposed to fix them?

 

One of the tests we do before releasing a new version of GIB is to play a head-to-head match of the old version vs the new version, to ensure that the new version wins overall and analyze the types of hands where the old version wins, in case they reveal changes that are detrimental. We actually pulled out one of the changes in last week's v21 release because of this (in the auction 1Major-1NT(forcing)-2Major, we were going to change whether GIB jumps to game with an invitational 3-card raise, but it ended up missing too many good games).

 

This thread is not very constructive. You've made general complaints about the robot play, which aren't actionable. Post hands in the GIB forum, like others do, and we can potentially act on them. I can't promise that every complaint will be addressed, but without knowing what's wrong, how are we supposed to fix it? We're NOT going to just revert to the pre-August version of GIB.

[Bbradley62]

The guy who apparently tweaks or manipulates these things is enormously sensitive and defensive about any criticism of his work (see posts above), and I guess the sorry state of these machines reflects his considerable, misguided ego.

Too funny!!

[barmar]

For 249 tournaments ending 2/15/12 I scored 51.48% in MP competition and +156.37 in IMP.

 

For 20 MP tournaments 2/16 through 2/26 the humiliating MP average was 45.77% with 16/20 below 50% and 17/20 below average in player score ranking. During the same stretch I played in 5 IMP tournaments with 3/5 negative results and 4/5 below average player rank.

There's two possible explanations for this: your robot partner is getting worse, or your robot opponents are getting BETTER.

PS-- now that I think about it I have noticed a few more than usual deals where GiB bids with 3hcp less than bidding box explanation (I wish it could be limited to a max 2hcp "lie"). That is not an excuse for the extent of my recent poor showing, though.

When simulating, it considers bids whose criteria are within 2 HCP of the hand at the 1-3 level, 3 HCP at the 4-5 level, and 4 HCP at 6-7. But sometimes what a bid shows is slightly different from the criteria for making it (e.g. raising responder's suit shows 4, but may sometimes be made with 3), so the combination of the "lie" and the adjustment for simulation can result in a larger deviation from the explanation. But theoretically, the hand should play about as well as if it had what it showed -- that's what the simulation is supposed to be discovering.