[jamegumb]

This was from a Robot Reward I just completed:

All vul, Dealer South

Link here: http://tinyurl.com/yyofjbk7

Robot (N)

S AQ43
H AQT94
D 4
C K54

jamegumb (S)

S 9
H K83
D AKQ
C AQJT82

Bidding (opps silent): 1C / 1H / 3C / 3S / 4H / 4NT / 5C / 5NT / 6D / 7H / P

Perhaps I should have figured out the Robot had the Club King and bid 7NT. But it's a reward race, so I'm giving myself maybe 1-2 seconds per bid. I'd think the Robot should in theory be the captain, as I've described my hand.

Lead is the Diamond 8. Robot wins in dummy, crosses to the Spade Ace, and leads the Heart 9 from hand. Low (Heart 5) from East. And the Robot runs the 9!

Thankfully, it works this time (East had J5 in the suit) and we quickly rack up 2210. But what calculation possibly makes this happen? Please to explain. Thanks!

[johnu]

Seems clear to me. Eight ever, nine never

[DaveRolyat]

Is it a safety play against a 5:0 trump split?

At IMP you can afford it, but at MP giving a trick away could be costly.

[pescetom]

Are we going to miss such spectacular plays when we get Argine?

[jamegumb]

DaveRolyat, on 2019-February-22, 02:09, said:

Is it a safety play against a 5:0 trump split?

At IMP you can afford it, but at MP giving a trick away could be costly.

Maybe this is what it was "thinking"? But, of course, this is total points, so all that matters is making the contract. Yes, this does allow the Robot to pick up 5-0 Hearts with East. So it's seemingly a pretty perfect 50% chance. Though playing low to the King allows for picking up all 3-2 splits, 4-1 onside, 5-0 onside, and 4-1 offside with the Jack dropping. Or 68.7+14.15+1.95+2.83 = 87.63% if I'm doing the math properly. (Actually a little less, because you risk a club ruff crossing to dummy for a finesse on the 4-1 or 5-0 onside splits.)

[HardVector]

DaveRolyat, on 2019-February-22, 02:09, said:

Is it a safety play against a 5:0 trump split?

At IMP you can afford it, but at MP giving a trick away could be costly.

There is no safety play in a grand slam.

[Stephen Tu]

jamegumb, on 2019-February-22, 12:44, said:

Maybe this is what it was "thinking"?

Presumably it's a bug or fluke in the deal sample generating algorithm running at very high speed/low resources generating a very tiny sample, not big enough to consistently reflect true probabilities, that happened to contain more 0-5 / 1-4 splits with the jack with East than deals with the J with West?

[jamegumb]

Stephen Tu, on 2019-February-22, 13:50, said:

Presumably it's a bug or fluke in the deal sample generating algorithm running at very high speed/low resources generating a very tiny sample, not big enough to consistently reflect true probabilities, that happened to contain more 0-5 / 1-4 splits with the jack with East than deals with the J with West?

This (sample generating algorithm) also makes sense, but surely it couldn't be more than a handful of samples before the odds overwhelmingly favor playing for the drop. Unless I'm missing some tipoff the robot derived from the early play.

[smerriman]

jamegumb, on 2019-February-22, 14:04, said:

This (sample generating algorithm) also makes sense, but surely it couldn't be more than a handful of samples before the odds overwhelmingly favor playing for the drop.

Given you know East doesn't have a void, I calculate:

- a 13% chance you must finesse
- a 52% chance you must not finesse
- a 35% chance both options work

If you simulate 10 hands under these conditions, about 5% of the time it will give the wrong conclusion.
If you simulate 20 hands, it's about 1%.

Nobody knows how many hands GIB simulates - but it's suspected to be quite small - while 1-5% is a low probability, it still means it will be making a glaring mistake like this 1 to 5 times every hundred deals (and those one in a hundred appear on the forum )

[jamegumb]

smerriman, on 2019-February-22, 14:46, said:

Given you know East doesn't have a void, I calculate:

- a 13% chance you must finesse
- a 52% chance you must not finesse
- a 35% chance both options work

If you simulate 10 hands under these conditions, about 5% of the time it will give the wrong conclusion.
If you simulate 20 hands, it's about 1%.

Nobody knows how many hands GIB simulates - but it's suspected to be quite small - while 1-5% is a low probability, it still means it will be making a glaring mistake like this 1 to 5 times every hundred deals (and those one in a hundred appear on the forum )

Appreciate the insight - what calculation did you do for the probability? My binomial calculator stops once the third variable (both options working) is introduced.

[smerriman]

My programmer side beats my mathematical side hands down this time, I just ran a large number of tests of x simulations until the probability flatlined.

[jamegumb]

smerriman, on 2019-February-22, 16:36, said:

My programmer side beats my mathematical side hands down this time, I just ran a large number of tests of x simulations until the probability flatlined.

Fair enough. Thanks.

[johnu]

smerriman, on 2019-February-22, 16:36, said:

My programmer side beats my mathematical side hands down this time, I just ran a large number of tests of x simulations until the probability flatlined.

You can use Pavlicek's card combination calculator. If you assume East led a singleton diamond, a first round finesse is the percentage play, but I don't think that is a correct assumption.

http://rpbridge.net/cgi-bin/xcc1.pl

[smerriman]

We were referring to the probability a simulation would give the incorrect results, not the probability the finesse is the right play.

[johnu]

smerriman, on 2019-February-22, 17:12, said:

We were referring to the probability a simulation would give the incorrect results, not the probability the finesse is the right play.

You can run a billion simulations and it won't do you any good if your modeling assumptions aren't any good