Version 1.47w - please post feedback and suggestions here

[Rain]

A new version of BBO was released today. Version 1.47w. Instant Tournament is the main new thing in this version. For details, see:

http://www.bridgebase.com/forums/topic/62524-new-version-of-bbo-released-instant-tournaments/page__pid__752445#entry752445

 

Please post your suggestions/feedback here!

[strings11]

While it seems like a really interesting concept, I can't help but feel that I should expect to feel the same level of robo-frustration as in any other event where there are 3 robots at the table.

 

Robot declarers are expert at seeking out any potential loser and making it real. It is as if it says to the defender robots: I know you guys are going to find this so take it now! Going down 1 is an art form it seems. In trying to prevent a human from gaining advantage by using a robot partner as declarer, robots routinely shed tricks apparently to negate any advantage to the human due to the robots' oh-so-rare ability to count to 13? This new format Must be "human declares"!

 

To be genuinely interesting, this new format should almost always have the bids of the robots at the table pre-programmed, as reminiscent of and dependent on the recorded actions of humans. Presumably the choices made by 'experts' are on file. Thus, rather than letting the robots do their own version of what is called 'thinking': actions - at least bidding actions - should be selected ad hoc, from historical data.

 

Of course, there will be outliers in any set of data, but that can be either eliminated, or introduced at a statistically experienced rate. (I would argue for elimination since it is not the response of the robots that we seek to measure here, but that is merely my opinion.) Another thought: perhaps the player could be given the choice as to whether he or she wanted to play with either, 'outliers' included, or, eliminated?

 

There is some potential, I would hope, for an attendant attenuation (just had to say that :) ) of the 'crosstalk' that appears to occur, in general, between robots.

 

[i realize that I will, in due course, be assured that robots do not communicate during the hands; but if robots are constrained to take bidding actions inconsistent with their routine programming, this would seem to lessen their ability to 'appear' to communicate. They are all presumably operating under the same programming, and for robots to be constained to take actions conceived of by actual humans in a defined situation, might make the actions of their fellow robots less clairvoyant.]

 

Robots' hand evaluation is suspect at best: They frequently do not hold the hcp ranges indicated in the bidding explanations, and they are as likely to give themselves a point or 2 for distributional features which are neutral or negative, as for those that are positive.

 

On a similar point, robots seem constrained completely, by their previous interpretations during the hand. In other words, once a robot has decided that partner's bid means this or that, the robot does not have the ability to discard that impression in the light of new possibly inconsistent information. Once it has decided on the size of the box that partner's hand belongs in, partner cannot then escape said box. Creativity, once the 'box' has been robo-defined is summarily punished.

 

If this new format is to be a 'referendum' on how well a particular person performs in a situation previously experienced by human experts; (... and since you are awarding 'masterpoints' for the activity), should it not properly be designed so that the robots are merely surrogates for genuine, expert, humans?

[gualtier]

While you can play a few hands instantly, which may be convenient to some, calling this a "tournament" is a hoax. You get compared with other humans who played the same hands in prehistoric times, against much weaker versions of GIB (that's how they made that contract...).

 

About as meaningful as Stac week...

[GreenMan]

During the instant tournament I can look at the play at other tables on the boards I've completed, but as soon as it's over I can only review my own table. Is that intentional?

[Rain]

During the instant tournament I can look at the play at other tables on the boards I've completed, but as soon as it's over I can only review my own table. Is that intentional?

 

After the tourney, your results and travellers can be found in both www.bridgebase.com/myhands as well as in your My BBO->Hands and Results->Recent Tourney folder on BBO.

[Rain]

While you can play a few hands instantly, which may be convenient to some, calling this a "tournament" is a hoax. You get compared with other humans who played the same hands in prehistoric times, against much weaker versions of GIB (that's how they made that contract...).

 

About as meaningful as Stac week...

 

Your partner robot is also stronger now. But yes, we have been able to make some good progress in our robots these 2 years. :)

[daswing]

i originally liked the idea, but being able to compare my scores with other tables just proves to me (again!) that the robot bidding and leads are very inconsistent. On one board, everyone else's robot rebid a spade allowing all other tables to reach a 4S contract, while my opps jumped to 4D and my robot p left it in:(. Then there was two hands where my robot opps lead was the one that make us go down, while everyone else's opp had a different lead---on the EXACT SAME BIDDING!!!!

 

(and i play with robots because i find people frustrating???)

[GreenMan]

After the tourney, your results and travellers can be found in both www.bridgebase.com/myhands as well as in your My BBO->Hands and Results->Recent Tourney folder on BBO.

 

I know that. That wasn't what I was asking. I like reviewing the hands in My Results. Did you really intend to allow review during the IT but cut it off as soon as it ends?

[GreenMan]

i originally liked the idea, but being able to compare my scores with other tables just proves to me (again!) that the robot bidding and leads are very inconsistent. On one board, everyone else's robot rebid a spade allowing all other tables to reach a 4S contract, while my opps jumped to 4D and my robot p left it in:(. Then there was two hands where my robot opps lead was the one that make us go down, while everyone else's opp had a different lead---on the EXACT SAME BIDDING!!!!

 

(and i play with robots because i find people frustrating???)

 

Sounds like we played in the same tournament. :D

[barmar]

I know that. That wasn't what I was asking. I like reviewing the hands in My Results. Did you really intend to allow review during the IT but cut it off as soon as it ends?

No, you should be able to look at the other tables after the tourney ends.

 

We've noticed a bug that sometimes affects the last board, we're working on a fix.

[GreenMan]

No, you should be able to look at the other tables after the tourney ends.

 

We've noticed a bug that sometimes affects the last board, we're working on a fix.

 

Thanks!

[nathan2008]

I must admit it is a fantastic idea !!! Much better than Iphone 5 lol. Now i can play at any time with immediate result.

 

Just a question, I notice that in fact i compare with other 14 people.. How do u choose these 14 players? In general, in robot duplicate games, we have more than 14 players. If you choose the 14 worst scores, i will definitely play this tourney .... maybe it is a bug? (You can check it in My Hands) I prefer we compare with all players.

[granguru]

A new version of BBO was released today. Version 1.47w. Instant Tournament is the main new thing in this version. For details, see:

http://www.bridgebase.com/forums/topic/62524-new-version-of-bbo-released-instant-tournaments/page__pid__752445#entry752445

 

Please post your suggestions/feedback here!

Have just played 4 hands and have a general feeling about this feature:

 

0. I will start saying I found the game addictive and with many advantages, but here I am to bury Caesar, not to praise him. The option to play whenever you want at your own pace is wonderful.

 

1. My main concern is that it is much harder to win this type of tourneys. It seems the 14 other competitors are the top of the list of the real tourney, which is unfair. Should be a random selection or a distribution that will produce a 50% average rank. It may not be so, but I had the feeling.

 

2. Bots of the tourney should have the same level of skill (SW version) as the one we are playing so that they do the same when we play as when they played the real tourney. I played 1NT and made 3, a top surely. No: 37.5% Why? Most of the games were 1NTx made 1, even 1 was 1NTxx. Strange, to say the least, as humans dont X 1NT with 14, and usually declare with 5 points at the other side and a 5th card suit.

 

3. (Spanish) Translations are very very poor. A warning is missing for trick one when it starts from the right. Some texts in the same line or area are in English and some in spanish. A text says that cards are being exchanged to permit you to BID, when it really is to play the hand. I undestand that this is not the definitive version, but this a failure seen since some years ago in many places. If you touch the button for explanations on the game the whole text is in English, untranslated.

4. Even what I said in number 1, I feel this is an easy way to earn points, and fast. You might get points with just a 51% result. For the ones fishing for them this is an excelent alternative, but not for the prize seekers. It should be a BBO$ reward, even more if it so hard to win. Even this could be a virtual carrot (players will win once in 100 games or such a low percentage) would make the game more challenging and attractive. Even a hall of fame could be installed. A simulation with different level of players, or a real statistic might give you some figures to think about this point.

 

5. There could be a ranking of the day/week of the higher %.

 

6. Once I got a message that the tourney could not be loaded. Retried and it worked. Warning message should be improved. Will this be a frequent error? Probably not, but I can not tell.

 

7. This point is not particulary restricted to IT, but to all bot play. Fred informed some time ago that bots wouldnt continue till the 7 level if necessary when

a. one competes to the 4 level with a minor suit

b. one double with a gosh hand and later show the suit and a strong hand. Bot continues up to the seven level resisting to let partner play the hand.

c. after a, say, 2NT overcall for minors, if p chooses D and opponents continue bidfing to game, if you give him an opportunity to play in 5C (with a 5-6 or 5-7) it takes it as a cue and jumps to 6D which is obviously doubled. Frustating thing is that 5Dx is a top and 6Dx is a zero.

 

8. Bot playing is very very good. Bot bidding is also pretty good. Bot defending has the major fails. It is time to improve their defensive methods in at least the signaling area.

2 basic signals would improve and solve 80% of the defensive play.

1. Understand the trick one signal: Hi=encouraging Lo=Discouraging. (in future versions you could give player a window to select the signaling system he/she wants). A poll could be made to use the commonest method used by BBO players.

2. Understand the SPS signal when partner gives him a ruff chance. Lo=low suit return Hi=highest side suit return Middle= no special interest or trumps (if a valid option)

Probably there is a work to do in coding what is encouraging when a 6 (f.i.)is returned. No doubt, but a percentage play can be detected. If bot led the 3 and has seen the 4 from declarer and dummy has the 5 there is a very high chance that the lowest side suit is asked. Finally, anything is better than a blind return, as today, which is very frustrating. This implies also a discarding technic for the declarer-bot according to the signaling defense is using. Discarding hi under Hi-low if declarer wants continuation(and viceversa). Using UDCA bot must discard Lo to encourage continuation.(same method used by the defense).

 

I understand that there are some players who dont care about signaling or, worst, bot will most of the time make better decisions than following partner's desires, This can be solved permitting the player a "no signals" option. Then the logic does not considers analyzing partners cards. As in real life the signal is an ask not an absolute order. Bot might consider, for instance because he has no more trumps, that a different return is better, and do so. But most of the time both analysis, human and bot's, should match.

[keeper2]

It seems that all the robot players in the original tournaments had the same random "seed" for their simulations (and so take the same action under the same conditions), whereas the new instant player is playing against robots with a different "seed" (and as a result the robotsbbb often take different actions from those in the original tournament) which can lead to extreme variability in results. In addition there have been GIB improvements over the years. This randomness takes away from the fun of what is otherwise an interesting innovation. It is less of a good "test" than standard robot tournaments where all face the same conditions.

[42krunner]

It is a neat idea, but unless it is slow on BBO (or very late), I am not sure I get why I would want to play in an instant tourney instead of a regular robo-dup.

[Bbradley62]

It is a neat idea, but unless it is slow on BBO (or very late), I am not sure I get why I would want to play in an instant tourney instead of a regular robo-dup.

If you play quickly in a regular robo-dupe, you may have to wait 10 minutes for results to be posted, whereas they are available as soon as you are finished in the instant tourney.

[wayne_b]

Since the Instant Tournament affects only a single player, is it possible to have the ability to suspend play for the duration of a live tournament? That would make these tournaments truly "filler."

[1eyedjack]

It seems that all the robot players in the original tournaments had the same random "seed" for their simulations (and so take the same action under the same conditions), whereas the new instant player is playing against robots with a different "seed" (and as a result the robotsbbb often take different actions from those in the original tournament) which can lead to extreme variability in results. In addition there have been GIB improvements over the years. This randomness takes away from the fun of what is otherwise an interesting innovation. It is less of a good "test" than standard robot tournaments where all face the same conditions.

 

Agreed. I am not paying any more money to play in a robot tourney where at different tables robots vary their decisions in identical conditions. Just played one where my robot partner was dealer. Half the room opened a weak 2D in first seat and half the room passed. How random is that? Think I will just wait an extra 5 mins for a level playing field.

[geitsab]

I am upset with the new game for the following reasons

 

(1) unable to review the hands after the 8th game.

 

(2) I get a bottom board for not making an additional trick when everyone else with the same bidding gets a friendly lead and mine is a defensive gem.

 

Why

[OBSugar]

Hmm. I know I should try again...but after two odd tops and three odder still bottoms in 5 hands, I signed off. Didn't feel like a true competition. Besides, I have grown to like the real-time games against lots of recognizable names. Wishing the best for BBO always, but hoping the convenience of an instant game doesn't pull from the numbers in the traditional games.

[GreenMan]

It's worth 25 cents to me to be able to start a game whenever I want. The robot inconsistencies are annoying, but if I don't get attached to the standings then it's no big deal. I use these as declarer-play exercise, which I need a lot of, and if the years-old bots were putting everyone else in a better-scoring contract then it's no big deal to me. I also like being able to review a hand and make comparisons right after playing it, when it's fresh in my mind.

 

I haven't had the can't-review-hands-at-the-end problem lately, so apparently that's an intermittent thing (or it got fixed).

[barmar]

Have just played 4 hands and have a general feeling about this feature:

 

0. I will start saying I found the game addictive and with many advantages, but here I am to bury Caesar, not to praise him. The option to play whenever you want at your own pace is wonderful.

 

1. My main concern is that it is much harder to win this type of tourneys. It seems the 14 other competitors are the top of the list of the real tourney, which is unfair. Should be a random selection or a distribution that will produce a 50% average rank. It may not be so, but I had the feeling.

This is not true. The players are chosen totally at random. For those curious about the algorithm, it's whatever MySQL's "ORDER BY RAND()" does.

 

2. Bots of the tourney should have the same level of skill (SW version) as the one we are playing so that they do the same when we play as when they played the real tourney. I played 1NT and made 3, a top surely. No: 37.5% Why? Most of the games were 1NTx made 1, even 1 was 1NTxx. Strange, to say the least, as humans dont X 1NT with 14, and usually declare with 5 points at the other side and a 5th card suit.

This isn't really feasible. We don't keep all the old versions of GIB online when we deploy new versions.

 

3. (Spanish) Translations are very very poor. A warning is missing for trick one when it starts from the right. Some texts in the same line or area are in English and some in spanish. A text says that cards are being exchanged to permit you to BID, when it really is to play the hand. I undestand that this is not the definitive version, but this a failure seen since some years ago in many places. If you touch the button for explanations on the game the whole text is in English, untranslated.

Other than a couple of new messages about loading the boards, all the messages in this application are the same as for other robot tourneys. Are you seeing different translation issues here?

 

6. Once I got a message that the tourney could not be loaded. Retried and it worked. Warning message should be improved. Will this be a frequent error? Probably not, but I can not tell.

We think we've fixed this.

 

7. This point is not particulary restricted to IT, but to all bot play. Fred informed some time ago that bots wouldnt continue till the 7 level if necessary when

a. one competes to the 4 level with a minor suit

b. one double with a gosh hand and later show the suit and a strong hand. Bot continues up to the seven level resisting to let partner play the hand.

c. after a, say, 2NT overcall for minors, if p chooses D and opponents continue bidfing to game, if you give him an opportunity to play in 5C (with a 5-6 or 5-7) it takes it as a cue and jumps to 6D which is obviously doubled. Frustating thing is that 5Dx is a top and 6Dx is a zero.

Do you have a question about this, are you saying that the bot bid a stupid grand slam? If you have a complaint about a particular hand, please post the details in the GIB Robot Discussion forum.

 

8. Bot playing is very very good. Bot bidding is also pretty good. Bot defending has the major fails. It is time to improve their defensive methods in at least the signaling area.

2 basic signals would improve and solve 80% of the defensive play.

1. Understand the trick one signal: Hi=encouraging Lo=Discouraging. (in future versions you could give player a window to select the signaling system he/she wants). A poll could be made to use the commonest method used by BBO players.

2. Understand the SPS signal when partner gives him a ruff chance. Lo=low suit return Hi=highest side suit return Middle= no special interest or trumps (if a valid option)

Probably there is a work to do in coding what is encouraging when a 6 (f.i.)is returned. No doubt, but a percentage play can be detected. If bot led the 3 and has seen the 4 from declarer and dummy has the 5 there is a very high chance that the lowest side suit is asked. Finally, anything is better than a blind return, as today, which is very frustrating. This implies also a discarding technic for the declarer-bot according to the signaling defense is using. Discarding hi under Hi-low if declarer wants continuation(and viceversa). Using UDCA bot must discard Lo to encourage continuation.(same method used by the defense).

Teaching the robots to understand signals is a hard AI problem. Unfortunately, we don't have any programmers with AI expertise.

[1eyedjack]

This isn't really feasible. We don't keep all the old versions of GIB online when we deploy new version

I keep an open mind but currently am unconvinced about this. Seems to me that all that is required is to restrict the population of available historical tourneys to those more recent tourneys that used the most recent implementation of GIB. There would still be no shortage of tourneys to select, and this would address the single most glaring criticism of the format.

[cwfromoz]

I absolutely convinced its much harder to have a good result in these tournies vs regular robot tournies. Ive played about 50+ of the instant tournies and score masterpoints at a fraction of the rate I do in regular robot tournies. There must be reason for this. The only reasonable assumption is robots were much weaker at the time of the orig tourney.

[barmar]

I absolutely convinced its much harder to have a good result in these tournies vs regular robot tournies. Ive played about 50+ of the instant tournies and score masterpoints at a fraction of the rate I do in regular robot tournies. There must be reason for this. The only reasonable assumption is robots were much weaker at the time of the orig tourney.

I haven't played in as many as you, but I've noticed the same thing. I usually scratch in robot tourneys, but I think I've only been above average in an instant tourney once.

 

FYI, all the tourneys currently being used are from the 2nd half of 2009. So there have been LOTS of robot improvements since then.