PDA

View Full Version : HM2 cEV algorithm.



YurkoPurko
01-23-2018, 12:38 PM
Hey guys,

trying to figure out how HM2 cEV algorithm works. Below is attached file with two hands. If you check those hands you will see that they are exactly same, only River actions are different. But for the first hand HM2 decides that all-in street is Preflop, equity is taken from turn. This hand looks strange, why equity is taken from turn? On the other side second had looks good and stats looks correct too.

Would you please explain what is difference between this two hands and why HM2 calculates it in different ways?

Best regards
Igor

YurkoPurko
01-23-2018, 12:49 PM
And second question is about folded but known cards. Will HM2 take into account known folded cards when calculating equity? Below is attached another sample hand ([Ks Kh] folded). Right now it looks like HM2 ignores folded cards, but why if those have been known before other/all-in action?

fozzy71
01-26-2018, 08:23 AM
Hey guys,

trying to figure out how HM2 cEV algorithm works. Below is attached file with two hands. If you check those hands you will see that they are exactly same, only River actions are different. But for the first hand HM2 decides that all-in street is Preflop, equity is taken from turn. This hand looks strange, why equity is taken from turn? On the other side second had looks good and stats looks correct too.

Would you please explain what is difference between this two hands and why HM2 calculates it in different ways?

Best regards
Igor

EV is calculated from the point we know hole cards for everyone that is all-in. That being said, this appears to be a bug in HM2 (HM3 uses a more advanced/accurate EV code) as neither of the KJ hands should be calculating EV since the 3rd player didn't fold until the river when all board cards had been shown.

Here you can see the results for these hands when I tested them in HM2 vs HM3:
http://i.imgur.com/fQRE979.png (https://imgur.com/fQRE979)
http://i.imgur.com/6kXuzm8.png (https://imgur.com/6kXuzm8)



Here is an older explanation of how EV works from one of our previous employees that was quite well educated on the subject, with a couple relevant bits in bold:



Watch this video - FAQ - Hold'em Manager (HM1) Poker Tracking Software :: EV Explained (http://faq.holdemmanager.com/questions/120/EV+Explained)


**CASH EV**

What EV can do / how it can be used:
Any allin situation, before the river, where a player has at least one "out" results in a EV $ Diff.
$Won + EV $ Diff = $USD EV
(AA vs KK is the most fun example... AK vs QQ is the most common "coinflip")

The "problem" what some people don't understand:
A: If a player has 0 outs, or
B: the allin situation takes place on the river there is NO EV DIFFERENCE.
C: If a shortstack goes allin preflop, and is called by two bigstacks. And the two bigstacks continue to bet on the flop, turn or river, this situation is treated as situation B. (EV = 0)
Fozzy : "You can't calculate all-in equity if you don't know the hands you are up against."
D: If you commit 80% of your stack with the best hand, but your last 20% goes allin with the worst hand, the $EV Difference will be calculated by your entire stack.

Why "EV by street" (which people who often see situation D want) is a bad thing:
Best explanation here - http://forums.holdemmanager.com/3rd-party-programs-compatible-holdem-manager/16665-enter-sect-7.html

Summary of that thread:
Example:
- you have AA, you raise to 80% of your stack, donkey calls, flop comes K83 rainbow.
- you then go all-in, no matter the flop, because you're committed.
- out of 100 times, 88 times donkey folds.
- 12 times donkeys calls with a set (33/88/KK).
What shall EV by street bogusly do? It shall do no computations for the 88 times where donkey folded--> "no more calculation".
What shall EV by street do the 12 times where donkey calls with a set? "Show that donkey sucked out and that you got unlucky".


So although you ran obviously uber-good by having donkey folding 88 times out of 100. EV by street focuses on the 12 times where donkey hit his set and tells that you're running below EV.
This is a well-known gambler fallacy. And this is why "EV by street" is completely bogus and should not be implemented.

Note: Tristanblue writes "it's precisely because EV by street does nothing to your adjusted-graph on these cases where the opponent folded that it is completely bogus."
But what if your opponent never folds? Suppose there are two players A and B.
Player A has AA, B has KK. (both have $100 stacks). They commit half their stack preflop and the flop comes AK6. Player B (KK has 1 "out") to win the hand.
If I would play this hand I would always make sure I'm allin on the turn.
However Player X always commits the rest of his stack on the flop and turn *except for one dollar*. And he commits on the river.
Of course, 4% of the time, the rivercard is the case King. Player X's EV Diff is always 0.
My EV Diff is -$4 (96 out of 100 times) and +$96 (4 out of 100 times)
So our EV graph actually looks the same after 100 of these hands.

**TOURNEY EV**

*Single Table Tournaments only (includes: Double or Nothings, Headsup, etc.... excludes: single table satellites!)
*cEV Diff (chip EV difference)....: this works exactly the same as the $EV Won for cashgames.
*$EV Diff (dollar EV difference)
*$Won + SUM $EV Diff = $EV Won
*The same "why EV per street is bad" applies for tourneys too.

-->Confusion: $EV Won vs $USD EV vs EV $ Diff vs $EV Diff vs cEV Diff vs $Won vs Luck Adjusted Winnings vs All-in EV (do we need eight different terms for three different things?)

*In addition: the $EV Won can be confusing when it's negative and a larger number than the actual buyin, or positive and a larger number than the actual first price money.
You have some extreme situations (often in Double or Nothing and in SUPER TURBO SNGs)

Extreme examples:
$169 buyin SUPER TURBO, $Won = $720 (1th place)---> $EV Won = -$196 (a negative value, that is higher than the actual buyin)
$5.20 buyin DoN, $Won = $10---> $EV Won = $11.35 (more than you can actually win)

So EV becomes quite a meaningless number if you focus on one game (or in the "case King hits the river" AA vs KK example on that one hand)
The more you play, the more accurate it gives a representation of your overall "luck".

And it doesn't take into account coolers. There was a program called set-o-meter (worked with PT2) where you could see how often you'd hit a set.
EV doesn't take into account how often (and how much) you win/lose with an overpair against a set (and vice versa).

http://forums.holdemmanager.com/manager-bugs/26732-pokerstars-double-nothing-ev-calculation-wrong.html





And second question is about folded but known cards. Will HM2 take into account known folded cards when calculating equity? Below is attached another sample hand ([Ks Kh] folded). Right now it looks like HM2 ignores folded cards, but why if those have been known before other/all-in action?

No, we do take any known cards into account because depending on the site's rules and hand history formatting we can't consistently know hole cards for all hands even if all the actions for 2 different hands from 2 different sites. WPN for example almost never writes all-in opponents hole cards to the hand history file, even if they may have chosen to shown them to you at showdown instead of mucking them.