[lycier]

---

Look,Gib play this hand randomly---ruffing ♥A and ♣K,I really don't believe Gib can't make 5♠.

[steve2005]

has 11 top tricks barring a compression play, like ruffing a good card or east having all 4 spades

[hotShot]

I assume this was done by a Basic GIB.

Ruffing the ♥A seems to be a typical error caused by a sample size that is to small .

[Bbradley62]

 steve2005, on 2014-May-15, 08:52, said:

has 11 top tricks barring a compression play, like ruffing a good card or east having all 4 spades

Given the 2-2 trump split, he has to make 2 compression plays to go down in 5.

[steve2005]

 Bbradley62, on 2014-May-15, 09:27, said:

Given the 2-2 trump split, he has to make 2 compression plays to go down in 5.

Off AK♠ so only needs to sluff 1 trick to go down in 5♠

[Bbradley62]

 steve2005, on 2014-May-15, 14:41, said:

Off AK♠ so only needs to sluff 1 trick to go down in 5♠

Yes, semantics; it has so many winners that after losing ♠AK it has to ruff two good cards (as it did) in order to go down one.

[Antrax]

I think it could go down in 4 by giving W a diamond ruff and their trump ace, then ducking a trump to E. So, good effort GIB but there's still room for improvement 7/10.

[uday]

this is the basic bot, which suffers from a lack of brainpower when compared to the advanced bot. So i can't worry too much that it messes up more hands in the play than the advanced bot would, but even so, I don't like it when the play drifts from bad to insane. We'll take a look - but we're trying to understand why a lobotomized player plays badly, which is harder than it seems.

The last time I tried this, week or so ago, when a GIB took an antipercentage hook at trick 2 ( QTx opp AKxxx ), I lost 4 days of my life trying to sort it all out, and while that was really interesting in its own way, the answer turned out to be "sample size too small for this problem." At least I learned a little bit about random numbers along the way. Enough to know I don't know anything. Sheesh.

I'm guessing the answer will be much the same here, tho I'm not sure. The hand has some characteristics that don't fit the usual pattern.

[helene_t]

GIB does many things that look completely silly but just happen not to cost a trick in its small sample. Ruffing its owns winners for no good reason, or failing to give partner a ruff. Probably not much to do about other than waiting until teraflops become cheaper so the sample size can be increased.

Uday: Sounds like it was my diamond slam that cost you four days of work. I am sorry about that.

[uday]

Lycier, I can't find this hand in myhands. When did u play it, do u remember ?

[lycier]

 uday, on 2014-May-16, 07:31, said:

Lycier, I can't find this hand in myhands. When did u play it, do u remember ?

yes,yes.

I also never find those hands which I play with Gib alone in the myhand.
but the exact time of this hand I played ：
Beijing Time ：about 8.15 PM on 15th May
NY Time：about 8.15 AM on 15th May

[Bbradley62]

Lycier: Is this using the "Just Play Bridge" feature?

[lycier]

 Bbradley62, on 2014-May-16, 10:20, said:

Lycier: Is this using the "Just Play Bridge" feature?

It should be "Play Bridge 4"

[Bbradley62]

Then how did it get to be Board 41? Doesn't that game play only four boards, then start over from Board 1 if you play again?

[hotShot]

 uday, on 2014-May-16, 07:14, said:

this is the basic bot, which suffers from a lack of brainpower when compared to the advanced bot. So i can't worry too much that it messes up more hands in the play than the advanced bot would, but even so, I don't like it when the play drifts from bad to insane. We'll take a look - but we're trying to understand why a lobotomized player plays badly, which is harder than it seems.

The last time I tried this, week or so ago, when a GIB took an antipercentage hook at trick 2 ( QTx opp AKxxx ), I lost 4 days of my life trying to sort it all out, and while that was really interesting in its own way, the answer turned out to be "sample size too small for this problem." At least I learned a little bit about random numbers along the way. Enough to know I don't know anything. Sheesh.

I'm guessing the answer will be much the same here, tho I'm not sure. The hand has some characteristics that don't fit the usual pattern.

I would think the usual pattern is "the played cards often do not matter" which reduces the usefulness of the samples.

Declarer misses Trump AK. These tricks are lost the cards played to these tricks do not matter. Trumps are 2-2 it even matters less.
Declarer has ♣AK (single) ♦AKJ ♥AKQ (void) In the top tricks and the ruffs the cards off defenders don't matter much.

So I would say the pattern is many quicktricks or ruffs => No significant differences between the tested plays
=>
a) one unusual sample gains a lot of weight
b) an inferior play has the same weight as the good play and is randomly picked

[lycier]

Here is this hand link ：

http://www.bridgebas...2D6SASJH7S7DQH3

[uday]

The interesting thing is that nothing matters unless trumps are 4-0 ( or something unusual along those lines)

If they're 4-0 offside, you can still make it double-dummy by stripping out that hand's site suit winners and then exiting with a high trump for a trump end play to compress AK92 into 2 tricks.

What it's doing is running into some rare hand types like this in its sample (25 hands, for basic bot) and allowing those to influence it.

In other words, something like:

"Well, on 24/25 hands, anything works. On the last hand, playing a trump fails. So I guess I won't play a trump"

This appears happen more frequently than I'm comfortable with, and the influence is certainly greater than I like. But I'm not sure this can be effectively addressed. I'm fiddling w/the code to find out if this is just the basic bot being too basic ( shrug ) or something deeper.

[uday]

My summary of what happened:

the approach taken by the basic bot is inherently inferior to that taken by the advanced bots.

On hands like this, where it is advantageous to make a plan ( win against anything except 4-0 trumps offside ), the approach taken by the advanced bot is simply better.

The basic bot's approach ( play what is DD effective at every turn, no plan per se ) usually works ( 89% of the time ) but it is an inferior approach, and will fail some of the time; there isn't anything to be done about it .

So I'm going to write this one off.

One suggestion I hear is to eliminate the dumbots altogether but they're a *lot* less resource intensive than the advanced bots.

U