All-In Adjusted off?

[mamba]

My month in HM2:
http://s7.directupload.net/images/120317/3c43btn3.png

In PT4:
http://s1.directupload.net/images/120317/hqrdzjac.png

Currency=MyCurrency - so not the cause.

Do both programs have completely different methods to determin adj. AI?
HM1 matches HM2 BTW.

[kraada]

We don't have access to their algorithm completely so I cannot say with 100% certainty exactly what differences there are.

One thing I know about our algorithm which comes to mind which may be applicable to Omaha is that if you are all in on a street and someone has called that all-in and that someone later folds, PokerTracker does not adjust winnings in that case and I have heard that (at least at some point in time, I admit I have not researched this recently) they do calculate. This difference in methods would cause divergent results especially when playing short stacked in a game where there are a lot of folds.

Assuming this is correct here is an example of what I mean:

You open shove preflop and get two callers A and B. After the flop A bets and B folds. We do not adjust winnings here because B had cards (and equity) at the time of the all in and we cannot calculate how much.

That said, it may also be something else entirely. If you can narrow down the list of hands further and get the discrepancy down to just a few hands we'd be happy to look into how our calculations are done. With a discrepancy that large if you start by breaking it down by session I'd expect to see at least one session with a big difference in numbers, so it might not take too long to find an obvious culprit or two.

[mamba]

Thanks for the detailed reply.

I don't have the time to make a huge investigation, but I compared yesterdays hands PT4 vs HM2 by sorting all hands by AI Equity.
Here is a hand that looks suspicious to me.
While HM2 has it as a 59% AI (which seems correct), the AI Equity in the PT4 report is blank. This means PT4 assumes I got it in with 100%, right?

PokerStars - $1 PL Hi (6 max) - Omaha Hi - 6 players
Hand converted by PokerTracker 4: http://www.pokertracker.com

UTG: $48.50
MP: $187.02
CO: $59.61
Hero (BTN): $110.78
SB: $101.39
BB: $47.76

SB posts SB $0.50, BB posts BB $1.00

Pre Flop: (pot: $1.50) Hero has Ac 9c 3d Kd

fold, fold, CO raises to $3.00, Hero raises to $9.00, fold, BB calls $8.00, CO calls $6.00

Flop: ($27.50, 3 players) 5d Ad 9h
BB checks, CO bets $26.26, Hero raises to $90.50, BB calls $38.76 and is all-in, CO calls $24.35 and is all-in

Turn: ($167.48, 3 players) 4d

River: ($167.48, 3 players) 4s

CO shows 7d Tc As 9s (Two Pair, Aces and Nines) (Pre 25%, Flop 21%, Turn 3%)
Hero shows Ac 9c 3d Kd (Flush, Ace High) (Pre 36%, Flop 59%, Turn 97%)
BB shows Jh Jc 7h 8h (Two Pair, Jacks and Fours) (Pre 39%, Flop 21%, Turn 0%)
Hero wins $164.68


And another one, 54% AI Equity, but shows up blank in the AI Equity column:

PokerStars - $1 PL Hi (6 max) - Omaha Hi - 6 players
Hand converted by PokerTracker 4: http://www.pokertracker.com

BTN: $111.41
SB: $163.72
BB: $23.50
UTG: $110.93
MP: $96.31
Hero (CO): $122.63

SB posts SB $0.50, BB posts BB $1.00

Pre Flop: (pot: $1.50) Hero has 7d 6s 8s Th

UTG calls $1.00, fold, Hero raises to $4.50, fold, fold, BB raises to $15.00, UTG calls $14.00, Hero calls $10.50

Flop: ($45.50, 3 players) 3s 9c 5h
BB bets $8.50 and is all-in, UTG raises to $68.95, Hero raises to $107.63 and is all-in, UTG calls $26.98 and is all-in

Turn: ($245.86, 3 players) 5c

River: ($245.86, 3 players) 8d

UTG shows Ac Qh 9s As (Two Pair, Aces and Fives) (Pre 42%, Flop 29%, Turn 61%)
Hero shows 7d 6s 8s Th (Straight, Nine High) (Pre 38%, Flop 54%, Turn 33%)
BB shows 3h Ks Ah 6h (Two Pair, Fives and Threes) (Pre 20%, Flop 17%, Turn 6%)
Hero wins $243.06

[Zangeeph]

We don't have access to their algorithm completely so I cannot say with 100% certainty exactly what differences there are.

One thing I know about our algorithm which comes to mind which may be applicable to Omaha is that if you are all in on a street and someone has called that all-in and that someone later folds, PokerTracker does not adjust winnings in that case and I have heard that (at least at some point in time, I admit I have not researched this recently) they do calculate. This difference in methods would cause divergent results especially when playing short stacked in a game where there are a lot of folds.

Assuming this is correct here is an example of what I mean:

You open shove preflop and get two callers A and B. After the flop A bets and B folds. We do not adjust winnings here because B had cards (and equity) at the time of the all in and we cannot calculate how much.

Your algorithm is not correct.

The point at which a player goes all in is NOT where the equity should be calculated. EV calculations must be done at all points where all the remaining players in the hand (i.e. those who haven't folded) have their cards face up. In your example the winnings should be adjusted. This is an issue that needs resolving immediately. I'm shocked to learn that PokerTracker has been doing EV calculations incorrectly this entire time.

EV adjusted winnings should be the same in both HEM and PT3. EV calculations are not a matter of opinion. Just like the bb/100 is the same in HEM and PT, EV should be the same.

[WhiteRider]

mamba - we currently have a known issue with 3-way pots. This is in our system to be fixed.

Zangeeph - please check out this tutorial about all-in EV calculations.

[Zangeeph]

mamba - we currently have a known issue with 3-way pots. This is in our system to be fixed.

Zangeeph - please check out this tutorial about all-in EV calculations.

In the "All-ins with unknown hole cards" section, in the given example, it is correct to state that no EV calculation should be made. However, it incorrectly concludes that no EV calculation should be made when an all in is made without the knowledge of all the cards at the point the all in occured. That section of the tutorial needs to be removed.

HEM correctly calculates EV. PT4 doesn't. That's unacceptable. People such as myself rely on seeing their adjusted bb/100 to determine their win rate and you've completely messed up the algorithm. The algorithm you have doesn't actually cause a bias but it is not correct. It's ignoring cases when it shouldn't.

Look at the following example I've made.

BTN: $100 J 2
SB: $50 5 6
BB: $100. A K

SB posts SB $0.50, BB posts BB $1.00

Pre Flop: (pot: $1.50)

BTN Raises to $4, SB calls, BB calls.

Flop (pot: $12) 2 A T
BB Checks, SB bets $46 and is all-in, BTN calls, BB calls.

Turn (pot: $150) 8
BB bets $40, BTN folds.

River (pot: $150) 7

SB shows 5 6 and won $100 net
BB Shows A K and won -$50 net

In this example, the EV calculation should be done on the turn. The EV calculation should be 5 6 against A K on a board of 2 A T 8 with no dead cards. The SB has an equity of 18.182% and a net EV of 0.18182*$150-$50=-$22.72700.

According to your algorithm, no EV calculation would be done because when the all in occurred we didn't know the cards of the BB and the BTN. I assure you that my is calculation is correct. If you don't believe me, you should post on 2+2's Probability forum, Poker Theory or Science Math and Philosophy forum where they will kindly explain how to do EV calculations. Or you could use Google and learn how to do it that way.

If PT4 doesn't fix this, I will need to use HEM to calculate my EV. I don't want to do that. Nobody wants to do that. I don't want to have to import my hands into HEM just to do something that PT4 should have correct. This issue explains the discrepancies between HEM's luck adjusted line and PT4's in SnGs. Let's get this issue fixed then we can move on.

If you are not willing to fix it then you should at least disable the EV calculations as they are not correct.

[kraada]

We respectfully disagree with your analysis. For a more detailed explanation on our thoughts, please see Jeff Hwang's article on the subject, reprinted (with permission from CardPlayer Magazine) here.

[Zangeeph]

We respectfully disagree with your analysis. For a more detailed explanation on our thoughts, please see Jeff Hwang's article on the subject, reprinted (with permission from CardPlayer Magazine) here.

Jeff Hwang is not a statistician. He is not qualified to talk about how EV should be calculated. Watch this TED talk by a statistician about the dire consequences when people think they can do calculations that only statisticians should be doing.

I decide to analyze this hand CardrunnersEV (notice the presence of EV in the name of the software). Their entire software is dedicated to EV analysis. And they charge $100 for this software. If anyone got this right, it'd be them.

Click here to see the analysis, notice the arrow.

Notice the -22.25 there? Once you add in the small blind of $0.50, you'll reach -$22.75. Does that number sound familiar? It may do, because in my analysis I arrived at -$22.73. The small difference of $0.02 is because it uses monte carlo.

EV is a tiny, almost insignifcant part of PokerTracker. It's understandable that there has been an error in the algorithm used to calculate EV. What's not understandable is refusing to fix the error once someone has told you about it.

[kraada]

I've seen the video before, and I'm not going to talk about an incomplete screenshot of a piece of software I do not have personal intimate knowledge of.

Please review the third example in this post.

In the example given there there are three potential outcomes:
(1) neither player improves, the hand is checked down and we win
(2) A improves and bets and we lose
(3) B improves and bets and we lose

Since the example is completely specified there are 10 known cards (Hero's two, A's two, B's two and the board's 4) so there are 42 potential outcomes. We know that (1) occurs 14 of those times
(2) occurs 15 of those times
(3) occurs 13 of those times.

Assuming this setup happened an infinite number of times your math would show: 14/42 * (1/3) + 15/42 * (61.4) + 13/42 * (61.4) or 41.04% equity in the long run.

That is far from the 33.33% actual equity in this particular scenario.

[Zangeeph]

I've seen the video before, and I'm not going to talk about an incomplete screenshot of a piece of software I do not have personal intimate knowledge of.

It's not an incomplete screenshot. It's a full analysis of the hand, from start to finish showing all the actions of every player. It's disappointing to hear that you're not familiar with CardRunnersEV, I would've thought that being aware of competing products was quite important.

Please review the third example in this post.

In the example given there there are three potential outcomes:
(1) neither player improves, the hand is checked down and we win
(2) A improves and bets and we lose
(3) B improves and bets and we lose

Since the example is completely specified there are 10 known cards (Hero's two, A's two, B's two and the board's 4) so there are 42 potential outcomes. We know that (1) occurs 14 of those times
(2) occurs 15 of those times
(3) occurs 13 of those times.

Assuming this setup happened an infinite number of times your math would show: 14/42 * (1/3) + 15/42 * (61.4) + 13/42 * (61.4) or 41.04% equity in the long run.

My math would not show that. The correct algorithm actually makes no calculations in the example you gave, so in this case your incorrect algorithm gives the same result as the correct algorithm.

EV calculations should be done in the following cases: All players who have not folded have their hands face up.

You would then input all the board cards and the hole cards into an equity calculator (such as the one included in PT4) to determine their equities and then multiply those numbers by the size of the pot to get their EVs.

I think that you think that if player A goes all in on the flop is called by B and C then B and C go all in on the turn, the equity calculation for A should be done for the board on the flop (i.e. ignore the turn card), and B and C's equity calculation should be done using the flop and turn cards. This would be incorrect as the percentages would not add to 100. All players, A B and C, need their equities calculated on the turn using all board cards.

Your EV calculations are at odds with HEM, CardRunnersEV and the rest of the poker world. How much longer are you going to stand by your broken algorithm?