PDA

View Full Version : SnG $EV calculations are still wrong



Mild Delirium
03-09-2010, 01:05 AM
I thought this bug was fixed long time ago.

All the stuff below happens in DoNs; can't say anything about other types of SNGs but I assume they suffer from the same problem.

The thing is, $EV diff numbers shown in tournament hands table are inconsistent with the numbers shown when you press ICM button in the replayer. Most severe differences occur when two hands have high odds of tying in the all-in (the problem is not limited to the cases when they actually tie, though - results are still very different if one of the hands wins).

E.g. a recent KQo vs KQs all-in (two-way, so no multi-way problems):



Payouts
1 $20.00
2 $20.00
3 $20.00
4 $20.00
5 $20.00

Start EV End EV Diff Player
$8.45 $9.20 $0.76 hero
$12.23 $12.14 ($0.10) player1
$15.56 $15.50 ($0.05) player2
$9.33 $9.20 ($0.13) player3
$13.84 $13.37 ($0.46) player4
$16.74 $16.92 $0.18 villain
$16.35 $16.31 ($0.04) player5
$7.50 $7.35 ($0.15) player6

SNG EV analysis
Board -
Prizes $20.00 $20.00 $20.00 $20.00 $20.00

EV Result Luck Hand Player
$8.65 $9.20 $0.55 KhQs hero
$12.23 $12.14 ($0.09) player1
$15.58 $15.50 ($0.08) player2
$9.29 $9.20 ($0.08) player3
$13.47 $13.37 ($0.09) player4
$17.00 $16.92 ($0.07) QcKc villain
$16.38 $16.31 ($0.07) player5
$7.42 $7.35 ($0.07) player6

hand order KhQs > QcKc probability 2.1732%
0 stack 2140, icm ev $14.04
1 stack 1580, icm ev $11.76
2 stack 2520, icm ev $15.18
3 stack 1070, icm ev $8.87
4 stack 1860, icm ev $13.00
5 stack 2145, icm ev $14.06
6 stack 2875, icm ev $16.02
7 stack 810, icm ev $7.06

hand order QcKc > KhQs probability 7.1574%
0 stack 0, icm ev $0.00
1 stack 1580, icm ev $13.56
2 stack 2520, icm ev $16.66
3 stack 1070, icm ev $10.47
4 stack 1860, icm ev $14.76
5 stack 4285, icm ev $18.80
6 stack 2875, icm ev $17.33
7 stack 810, icm ev $8.41

hand order KhQs = QcKc probability 90.6695%
0 stack 1070, icm ev $9.20
1 stack 1580, icm ev $12.14
2 stack 2520, icm ev $15.50
3 stack 1070, icm ev $9.20
4 stack 1860, icm ev $13.37
5 stack 3215, icm ev $16.92
6 stack 2875, icm ev $16.31
7 stack 810, icm ev $7.35


So, ICM in the replayer says our $EV diff is -$0.55, but HM shows $EV diff -$2.53 for that hand. They differ by $2, which is about 20% of tourney buy-in. A bit too much, I would think, considering winning regulars get only 5-10% of buy-in per tourney at this limit. Other all-ins (where odds of one hand winning are greater) have the same problem, but the difference is smaller. E.g. for TT vs AT all-in ICM in the replayer has $5.73 difference, but HM hands view shows $5.87. 14 cents, I know. Not much but they still differ.

Of course those inconsistencies should cancel each other out in the long run, but then again, in that long run there's no point of having the red line displayed on the profit graph at all.

Now the question is, why do results of the same calculations in the same program differ unless they are calculated differently and one of methods is wrong? And if it is wrong, why hasn't it been fixed yet when you already have another method that works fine?! And speaking of programming design decisions, I can't help but wonder how two different implementations of the same algorithm could end up in a single program in the first place... See what happens when you do that?

I remember writing you guys about this bug like a year ago,. You received the HHs and said it would be fixed in the next (back then) release so I assumed it was, since ICM calculations for 2-way all-ins in a small field aren't really rocket science.

I appreciate all new features you add and everything, but frankly I'd rather have a program with limited functionality that produces results I can trust.

So please, fix the $EV stuff. I assume other people who reported the same thing have sent you enough hand histories to be able to reproduce it.

P.S. It looks like manually imported (via import file/folder) hands don't have this problem, that is, $EV diff values in hands table and in replayer's ICM for such hands are the same.

Patvs
03-09-2010, 08:00 AM
but then again, in that long run there's no point of having the red line displayed on the profit graph at all.



What's wrong with long term results? Read this: http://forums.holdemmanager.com/manager-bugs/26732-pokerstars-double-nothing-ev-calculation-wrong.html


And we'll look into again why the ICM numbers don't match the EV numbers in HM.

Patvs
03-13-2010, 04:32 PM
So, ICM in the replayer says our $EV diff is -$0.55, but HM shows $EV diff -$2.53 for that hand.


Can you post the actual handhistory.

KQ (1070 stack = HERO) vs KQ (3215 stack)

(2.1732% * $14.04) + (7.1574% * 0) + (90.6695% * $9.20) = $8.6466

This hand resulted in a splitpot? So the EV Diff is $0.55 ($9.20 - $8.6466)
Can you post a screenshot where HM states is -$2.53 for this one hand?