[eagles123]

{comments}

---

looks like a bug - why doesn't GIB repeat the heart finesse?

thanks,

Eagles

[gwnn]

This is how you post the link without it looking ugly:

(hv=http://long-url-you posted)400|300(/hv)

but with square brackets [].

[gwnn]

---

like this.

[Lovera]

I think that GIB is feared about impasse (in this case against King of heart retained for a second round losing than 1 heart+2 diamonds=3tricks if..vs -2). Infact i had observed that when King is onside i lead a little card because often GIB cover with Ace making my King winner.

[1eyedjack]

Agree - Bug.

[lycier]

I don't believe this is a bug,sometimes Gibs play worse,maybe this is just so-called skill?
However I am sure that Gibs play so better than before.
Let Georgi ,Stephen Tu or Jdonn tell us.

[1eyedjack]

Pure semantics.

How bad does a line of play have to be before it qualifies as a bug? Is it just a matter of degree? If not, can any line of play ever be described as a bug.

One of the encouraging points about jdonn's recent post was that, popular perceptions to the contrary, and acknowledging that it is a difficult task, GIB's card play does receive some attention.

[Stephen Tu]

Is this advanced GIB or basic GIB? advanced GIB shouldn't have problem repeating finesse. Basic GIB sometimes have this problem as this kind of thing is endemic to double dummy reasoning vs. single dummy. Don't play basic GIB if this thing going to bother you.

Quote

Let Georgi ,Stephen Tu or Jdonn tell us.

Why you lump me in with them? I don't work for BBO. I am just familiar with GIB since use it since beginning, and follow the GIB discussion board from when Ginsberg talk about how it work and what sorts of things it has problems with.

[lycier]

Stephen Tu, on 2015-May-17, 06:03, said:

Is this advanced GIB or basic GIB? advanced GIB shouldn't have problem repeating finesse. Basic GIB sometimes have this problem as this kind of thing is endemic to double dummy reasoning vs. single dummy. Don't play basic GIB if this thing going to bother you.


Why you lump me in with them? I don't work for BBO. I am just familiar with GIB since use it since beginning, and follow the GIB discussion board from when Ginsberg talk about how it work and what sorts of things it has problems with.

I disagree.
Whatever it is basic or advanced,BBO will try to improve them,please don't be too strick to speak,you would better suggest us,ok?

[Stephen Tu]

lycier, on 2015-May-17, 07:09, said:

I disagree.
Whatever it is basic or advanced,BBO will try to improve them,please don't be too strick to speak,you would better suggest us,ok?

I'm having trouble parsing your English here, I understand it's not your native tongue. What are you asking me to do, can you try rephrasing?

What I'm suggesting is that this problem of not repeating working finesses isn't a "bug" at least as "bugs" are traditionally defined in computer programming, i.e. the program is not behaving as designed because of a mistake in the implementation code. Rather it's a known inherent limitation of how the basic GIBs work, the program is working as designed, it is not a "bug". This version simply cannot do it. They already know the "fix": use advanced GIB. But they do not use advanced GIB because too expensive (memory, CPU time) to run advanced GIB for everyone at the lower $1/week cost. Maybe when computing power becomes significantly cheaper in the future they can do it.

Asking to "fix" basic GIB is like asking for someone to "fix" a car to fly like an airplane because you don't want to pay for airplane.

With double dummy analysis, with the basic GIB engine, there is no memory of whether a finesse worked or not. The computer is, at every turn to play, analyzing what it thinks is best to do at that point. So at the critical juncture, basically its random sample suggested drop was better than finesse, not knowing at all that the finesse worked before. Now why it thinks drop is better on this hand, I don't know, perhaps its limited sample size not accurate (remember basic GIB also running very fast, small sample). Also, the fact that it thinks that it will get double dummy decisions correct basically makes it do a lot unnecessary flawed things on a single dummy basis (on this hand, playing unnecessary rounds of trumps risking extra undertrick), because it assumes it will know whether to finesse again or not later. Playing the extra trump can't hurt double dummy, so it is "equivalent" double dummy, so it will randomly choose between that and playing another heart to finesse.

The advanced GIB however, doesn't play this way, it has true single dummy engine, and uses some sort of "planning" mechanism which I don't fully grasp so it's hard for me to explain well. Basically it decides what to play for in terms of distributions of the opp's cards and location of key honors, and once it decides on a line it will follow it as long as what it decided on isn't inconsistent with what the opps show up with. So somehow it avoids these problems. I tried this hand on my personal copy of GIB (old 6.1.3, running on my laptop, not a BBO GIB robot), it handles this hand perfectly fine. It doesn't play extra trumps, it takes one heart finesse, then claims since it can discard the second heart on the club and ruff all the diamonds.

[lycier]

Too many words,many of those are useless because Moscow Distrusts Tears !
Maybe there are many things which you don't know,Mr.programmer.
For example,many of my members,their most commonly used word is silly,only ! as long as talking about basic robot.
And I have tried to persuade them,the results unfortunately like yours to little avail.
If the public have no opinion of basic robot,who is willing to use advanced robot?

[Stephen Tu]

The basic robot is lobotomized. The higher thinking portion of its brain has been removed on purpose. So it cannot simply be "fixed". If you "fix it", you then have the advanced bot which is more expensive to run. If people are unwilling to try the advanced bot to see that it is indeed better, then it's their loss. They are the ones being silly, just as silly as basic GIB can be sometimes, if they think that a computer using way more advanced programming algorithm, more CPU time & memory cannot perform better. Because basic bot play bad they think advanced bot must play bad also? Why are they not even willing to pay a couple times for advanced bot just to form an opinion whether worth it to pay more? To me it doesn't take long to see the difference.

You can complain about the basic bot play all you want, but you'd be wasting your time posting about it. What do you expect to happen? It cannot be fixed without turning it into advanced bot. It's not like there is a mistake in one line of code somewhere that can be fixed to make the bot repeat finesses. Correcting this kind of error is basically requiring thinking about the hand in an entirely different way. Computers do not think like humans. Sometimes what is simple from a human standpoint to learn is extremely complicated to get a computer to do.

In the years you've been following this forum, how many play/defense engine improvements have you seen BBO announce? I can't remember any. Maybe fixing what card is led on opening lead from certain holdings as that's conventional and rule based. And I guess leading aces vs. 7nt. But otherwise I don't remember fixes in this area. Josh said to go ahead and post play bugs, but I guess I'll believe him once they actually post a version with claimed play/defense improvements. To me if/when they actually do so, the problems are basically known so it's not likely to be in response to specific complaints, like if you post a particular play error they will come up with some fix specific to that error, that they would not have found if you hadn't brought up the hand in question. Play errors don't lend themselves to situational rule based fixes, it's a totally different kind of problem than just adding a bidding rule to handle a specific sequence of bids.

On the other hand, complaining about bidding bugs has resulted in plenty of specific fixes although certainly only a percentage of them have been addressed. And better bidding rules improve both the basic and advanced bots (advanced bots have better bidding judgment towards the end of the auction when simulation is allowed, but better rules keep both bots on saner courses of action throughout).