HM1/HM2 Icm '$EV Diff' wrongly imported?

[kampiuceris]

Thanks for the answer. If you guys need help like pseudo code or testing new build, let me know.

[kampiuceris]

Hey, I've quickly ported my algorithm to silverlight app (it may contain minor bugs because I had to rewrite some code (fix language differences)). But you can see all calculations below the results. This tool includes Regular Ev (the one you guys are trying to fix) and Full Ev (street by street implementation). Supports only PokerStars handhistory format so far. Let me know if you have questions. Here it is: http://chainer.puslapiai.lt/CloudEv/CloudEv.html

[udbrky]

I have forwarded this to Patvs.

[Patvs]

I've updated HM-5469.

[kampiuceris]

Long time no see. I still see lookup table inaccuracy which leads to different "EV diff" (usually some cents) on many hands. My question is: will this inaccuracy (probably approximations in lookup table) will be fixed in the future? Or are you guys happy with what's done today?
In any case, this time I came up with hand which seems to be incorrectly calculated not due to lookup table. It's probably some error in the pot distribution algorithm or whatever.

HandHistory: (Fifty50, ICM = { 60 10 10 10 10 })

PokerStars Hand #103742561448: Tournament #784392110, $95.86+$4.14 USD Hold'em No Limit - Level XI (150/300) - 2013/09/06 17:02:54 EET [2013/09/06 10:02:54 ET]
Table '784392110 1' 10-max Seat #10 is the button
Seat 1: kampiuceris (151 in chips) 
Seat 2: Method999 (3825 in chips) 
Seat 4: ChuLai63 (2177 in chips) 
Seat 5: m@chaon (3375 in chips) 
Seat 6: wieselsen (2177 in chips) 
Seat 7: burnyo26 (1291 in chips) 
Seat 10: marin06 (2004 in chips) 
kampiuceris: posts the ante 40
Method999: posts the ante 40
ChuLai63: posts the ante 40
m@chaon: posts the ante 40
wieselsen: posts the ante 40
burnyo26: posts the ante 40
marin06: posts the ante 40
kampiuceris: posts small blind 111 and is all-in
Method999: posts big blind 300
*** HOLE CARDS ***
Dealt to kampiuceris [9h 2h]
ChuLai63: raises 1837 to 2137 and is all-in
m@chaon: folds 
wieselsen: folds 
burnyo26: folds 
marin06: folds 
Method999: folds 
Uncalled bet (1837) returned to ChuLai63
*** FLOP *** [9s Ac 9c]
*** TURN *** [9s Ac 9c] [Qc]
*** RIVER *** [9s Ac 9c Qc] [2c]
*** SHOW DOWN ***
ChuLai63: shows [Jc Jh] (a flush, Ace high)
ChuLai63 collected 378 from side pot
kampiuceris: shows [9h 2h] (a full house, Nines full of Deuces)
kampiuceris collected 613 from main pot
*** SUMMARY ***
Total pot 991 Main pot 613. Side pot 378. | Rake 0 
Board [9s Ac 9c Qc 2c]
Seat 1: kampiuceris (small blind) showed [9h 2h] and won (613) with a full house, Nines full of Deuces
Seat 2: Method999 (big blind) folded before Flop
Seat 4: ChuLai63 showed [Jc Jh] and won (378) with a flush, Ace high
Seat 5: m@chaon folded before Flop (didn't bet)
Seat 6: wieselsen folded before Flop (didn't bet)
Seat 7: burnyo26 folded before Flop (didn't bet)
Seat 10: marin06 (button) folded before Flop (didn't bet)

Currently using:
HM1 1.13.02 (gives -$19.03)
HM2 2.0.0.7852 (gives -$19.03)

When clicking "ICM" button in hm1 hand replayer it gives correct numbers (I actually think it always gives correct numbers)

Payouts
1	$575.16
2	$95.86
3	$95.86
4	$95.86
5	$95.86

Start EV	End EV	Diff	Player
$13.72	$50.46	$36.74	kampiuceris
$212.10	$197.88	($14.22)	Method999
$147.96	$146.22	($1.74)	ChuLai63
$195.89	$192.25	($3.64)	m@chaon
$147.96	$142.62	($5.34)	wieselsen
$100.97	$94.80	($6.17)	burnyo26
$139.99	$134.37	($5.63)	marin06

SNG EV analysis
Board - 
Prizes $575.16 $95.86 $95.86 $95.86 $95.86 

EV	Result	Luck	Hand	Player
$7.73	$50.46	$42.72	9h2h	kampiuceris
$200.23	$197.88	($2.35)		Method999
$171.59	$146.22	($25.36)	JcJh	ChuLai63
$194.77	$192.25	($2.52)		m@chaon
$146.74	$142.62	($4.11)		wieselsen
$98.84	$94.80	($4.04)		burnyo26
$138.70	$134.37	($4.34)		marin06

hand order 9h2h > JcJh probability 15.0633%
0 stack 613, icm ev $50.46
1 stack 3485, icm ev $197.88
2 stack 2215, icm ev $146.22
3 stack 3335, icm ev $192.25
4 stack 2137, icm ev $142.62
5 stack 1251, icm ev $94.80
6 stack 1964, icm ev $134.37

hand order JcJh > 9h2h probability 84.4461%
0 stack 0, icm ev $0.00
1 stack 3485, icm ev $200.66
2 stack 2828, icm ev $176.17
3 stack 3335, icm ev $195.22
4 stack 2137, icm ev $147.48
5 stack 1251, icm ev $99.57
6 stack 1964, icm ev $139.49

hand order 9h2h = JcJh probability 0.4906%
0 stack 306.5, icm ev $26.76
1 stack 3485, icm ev $199.04
2 stack 2521.5, icm ev $161.48
3 stack 3335, icm ev $193.49
4 stack 2137, icm ev $144.62
5 stack 1251, icm ev $96.75
6 stack 1964, icm ev $136.46

My calcs:

Board[] Dead[] Pot=613
kampiuceris          9h 2h      15.31%     93.84     
ChuLai63             Jc Jh      84.69%     519.16    

Board[] Dead[9h2h] Pot=378
ChuLai63             Jc Jh      100.00%    378       

Total chips in all pots: 991
Payouts: { 0.6 0.1 0.1 0.1 0.1 }
Buy In: $100.00 ($95.86+$4.14)
Total Prize Pool: $958.60


PlayerName        Pocket       Chips Diff      Chips Won       Chips Ev        Chips Ev Diff   Icm Ev Diff $         
---------------------------------------------------------------------------------------------------------------------
kampiuceris       9h 2h        462             613             93.84           -519.16         -0.0445695  $-42.72    (Hero)
Method999                      -340            0               0               0               +0.0024561  $+2.35     
ChuLai63          Jc Jh        38              378             897.16          519.16          +0.0264593  $+25.36    
m@chaon                        -40             0               0               0               +0.0026236  $+2.52     
wieselsen                      -40             0               0               0               +0.0042926  $+4.11     
burnyo26                       -40             0               0               0               +0.0042118  $+4.04     
marin06                        -40             0               0               0               +0.0045261  $+4.34     
---------------------------------------------------------------------------------------------------------------------
kampiuceris       9h 2h 15.3%  462             613             93.84           -519.16         -0.0445695  $-42.72    (Hero)
Method999                      -340            0               0               0               +0.0024561  $+2.35     
ChuLai63          Jc Jh 84.7%  38              378             897.16          519.16          +0.0264593  $+25.36    
m@chaon                        -40             0               0               0               +0.0026236  $+2.52     
wieselsen                      -40             0               0               0               +0.0042926  $+4.11     
burnyo26                       -40             0               0               0               +0.0042118  $+4.04     
marin06                        -40             0               0               0               +0.0045261  $+4.34     




##################### Full Ev #####################
##################### Preflop (Evaluations 1712304) #####################

Player Name         Icm Ending   ($)            Weight ($)            
kampiuceris         0.0080655337 ($7.73)        100%   ($7.73)        
Method999           0.2088827158 ($200.23)      100%   ($200.23)      
ChuLai63            0.1789969379 ($171.59)      100%   ($171.59)      
m@chaon             0.2031787986 ($194.77)      100%   ($194.77)      
wieselsen           0.1530754309 ($146.74)      100%   ($146.74)      
burnyo26            0.1031054457 ($98.84)       100%   ($98.84)       
marin06             0.1446951374 ($138.70)      100%   ($138.70)      

-------------------------------------------------------------

9h2h  JcJh      probability 84.44610% (1445974/1712304)
1     0     
Player Name         Stack       Icm                 Weight    Icm Weight          
kampiuceris         0           0.0000000000        100%      0.0000000000        
Method999           3485        0.2093280749        100%      0.2093280749        
ChuLai63            2828        0.1837779383        100%      0.1837779383        
m@chaon             3335        0.2036545510        100%      0.2036545510        
wieselsen           2137        0.1538539762        100%      0.1538539762        
burnyo26            1251        0.1038693870        100%      0.1038693870        
marin06             1964        0.1455160726        100%      0.1455160726        


9h2h  JcJh      probability 15.06327% (257929/1712304)
0     1     
Player Name         Stack       Icm                 Weight    Icm Weight          
kampiuceris         613         0.0526350475        100%      0.0526350475        
Method999           3485        0.2064266571        100%      0.2064266571        
ChuLai63            2215        0.1525376178        100%      0.1525376178        
m@chaon             3335        0.2005551787        100%      0.2005551787        
wieselsen           2137        0.1487828680        100%      0.1487828680        
burnyo26            1251        0.0988936210        100%      0.0988936210        
marin06             1964        0.1401690100        100%      0.1401690100        


9h2h  JcJh      probability 0.49063% (8401/1712304)
0     0     
Player Name         Stack       Icm                 Weight    Icm Weight          
kampiuceris         306.5       0.0279181584        100%      0.0279181584        
Method999           3485        0.2076341845        100%      0.2076341845        
ChuLai63            2521.5      0.1684533876        100%      0.1684533876        
m@chaon             3335        0.2018434856        100%      0.2018434856        
wieselsen           2137        0.1508638252        100%      0.1508638252        
burnyo26            1251        0.1009286280        100%      0.1009286280        
marin06             1964        0.1423583307        100%      0.1423583307        






##################### Regular Ev #####################
##################### Preflop (Evaluations 1712304) #####################

Player Name         Icm Ending   ($)            
kampiuceris         0.0080655337 ($7.73)        
Method999           0.2088827158 ($200.23)      
ChuLai63            0.1789969379 ($171.59)      
m@chaon             0.2031787986 ($194.77)      
wieselsen           0.1530754309 ($146.74)      
burnyo26            0.1031054457 ($98.84)       
marin06             0.1446951374 ($138.70)      

-------------------------------------------------------------

9h2h  JcJh      probability 84.44610% (1445974/1712304)
1     0     
Player Name         Stack       Icm                 
kampiuceris         0           0.0000000000        
Method999           3485        0.2093280749        
ChuLai63            2828        0.1837779383        
m@chaon             3335        0.2036545510        
wieselsen           2137        0.1538539762        
burnyo26            1251        0.1038693870        
marin06             1964        0.1455160726        


9h2h  JcJh      probability 15.06327% (257929/1712304)
0     1     
Player Name         Stack       Icm                 
kampiuceris         613         0.0526350475        
Method999           3485        0.2064266571        
ChuLai63            2215        0.1525376178        
m@chaon             3335        0.2005551787        
wieselsen           2137        0.1487828680        
burnyo26            1251        0.0988936210        
marin06             1964        0.1401690100        


9h2h  JcJh      probability 0.49063% (8401/1712304)
0     0     
Player Name         Stack       Icm                 
kampiuceris         306.5       0.0279181584        
Method999           3485        0.2076341845        
ChuLai63            2521.5      0.1684533876        
m@chaon             3335        0.2018434856        
wieselsen           2137        0.1508638252        
burnyo26            1251        0.1009286280        
marin06             1964        0.1423583307        




Regular EV = 
(0.84446 * 0.00000000) +
(0.15063 * 0.05263505) +
(0.00491 * 0.02791816) 
= 0.00806553369434663 ($7.73)
Hero after hand = 0.0526350474679012 ($50.46)
Regular EV Diff = 0.00806553369434663 - 0.0526350474679012 = -0.0445695137735545 ($-42.72)

[Patvs]

The lookup table is already an improvement over the old lookup table.
Despite some hands may still result in a few cents off, we're happy with this.

Similarly: take 'Run It Twice' hands =, HM2 only counts the first run for EV, but the EV Diff is still based on the sum of both outcomes. It isn't perfect because we completely disregard the fact that 1-5 cards have already been revealed (and thus can't appear on the second run). Though the EV is still going to be pretty close despite of this, so we see no real urgency to fix this.


--


Having said that, there still are numerous real EV bugs.
The splitpot bug HM-5469, Omaha EV is mostly always wrong,

[zamb3zi]

I recently re-imported my whole hand history archive and I noticed a big change in my cumulative $EV Diff. My history is about 240k hands over the last 10 months.
In my old DB I have ICM Diff $1044 (which is massive at my stakes) and in the newly imported DB it's $695.
Is this explained by improvements during this year? Is the latter number more accurate, and just how accurate can I expect it to be now? Thanks.

[Patvs]

What types of games are you playing exactly on which site?

[zamb3zi]

What types of games are you playing exactly on which site?

PokerStars turbo 6-max STT. Mostly all $7, but also some $3.5 and $15.

[Patvs]

Turbo or Hyper Turbo?

Most EV fixes tackled bugs with calculating EV for hyper turbo's.
Though whether you play Turbo or Hyper Turbo, the latest HoldemManager build has the most accurate EV.
(there still are some bugs.... only affecting splitpots... and Omaha)

So the $695 is the accurate EV Diff. And $1044 was bugged.