Is there anybody here who originally used my code for 6-max SNGs (ie: with the correct Entry Fee Percentage) but has now switched to using HEM? If so, do the graphs look approximately the same?

Juk

I must be doing something wrong, i've played 14$+1$ super turbo and ran horribly, still your program shows me that i ran 800$ above EV... (i run it wih the 7$ .bat since the entry fee ratio is the same)



I've put all my HH in the folder, there are almost 500 for ~240 super turbos, most files are almost empty, like 0 or 1 ko... could this be the reason?

EDIT : After sorting out the HH only, it doesn't change a thing. There must be a problem, i've run so sickly it was insane, no way i ran twice better than EV.

Sorry for my english, what "like 0 or 1 ko" means?

Don't put summary files in programm folder, this may cause weird results.

Files that weigh 1kb or 2kb. (1 kilo-octet
But i sorted them out, it didn't change a thing.

There's definitely a bug, after trying out my 705 sit n goes @ 7.5$ in which i won 200$, it shows that i lost 6$ and that i should have lost 1470$. I ran soooo good.....


Or not.

Any ideas ?

When I open allin_luck_resulsts_by_hand there are a lot of dates in the second raw (raw B)... seems weird.

If you play short sessions, try 20-30 games to see if "Actual Won" show right value.
Weird results happens on big number of games, but i can't catch this bug. My 1K sng's test show nothing.

You can send you HH to me (aruta[at]sibmail.com), then i can try find what's wrong.

Well yeah still kinda wrong even on 30 games samples. (actual winnings are true, but still saying i'm running above EV even while losing 250$ on 15$ games in 30 games lol (should run @-350$ lol, how is this even possible, that's almost -100% ROI, like i was drawing dead preflop all the time))

Probably just isn't working well with ST or im making something wrong.

Hello Juks .

is it possible to put PS 18 men sng in sng luck ?.i've tried to create a new .bat but there is no ratio for 18 men(3.40$ and 6.50$) in the read me .i had the 1.03e version

ty .

It's not possible to use my code with more than 3 places paid atm (and also it's not possible to use $EV luck-adjustment until the final table anyway because you don't know all the other player's stacks). Assuming the total number of chips in play doesn't exceed 32k then the cEV luck-adjustment calculations should still work OK (although I've never tried it...)

Juk

ok ty JUK for your answer

the download link is broken. Where can i get it?

See post number 1207 above for the URL of the download.

Juk

anything for mtts yet?

Who can we talk to about explaining the excel files that are created? Is there a way to get the data in a graph?

They should be fairly self-explanatory, but if not then I can answer any questions about them.

To get a graph then just plot the 2nd, 3rd and 4th columns from "allin_luck_results_by_sng.csv" in Excel.

Juk

Hey juk and all,

I just read the whole thread... Took me quite some time

At one point someone wrote:

That seems insane. Juk can correct me if I'm wrong, but I believe that hypothetically you could play one $60 SNG and have the program show a -$1000 EV result. I have no idea what normal variance is for luck in these things.

To which juk answered:

I've got absolutely zero problem with losing more than the SNG buyin itself...

However, I'm a bit confuzzabled: is it's possible to lose more in $EV than: (first price - buy-in)?

Maybe the following scenario could help you explain me what I'm getting wrong...

Scenario: heads-up Sng, no buy-in, no entry fee (yup, it's simplified). Payout: $20 to the winner (example is oversimplified on purpose so that computations for the sake of this example stay simple).

Total chips: 3000
Hero chips: 2250
Villain chips: 750

Then comes a deal with an all-in: hero's AKo all-in vs villain AQo: 74% vs 26% (crude approx for the sake of this post, and we ignore ties).

Case 1: hero wins and hence wins $20. So hero "really" won $20*0.74 + $20*0.26*0.5, which gives $17.4 right!? (74% of the time hero win $20, 26% of the time he'll be at the same number of chips as villain, so we consider hero is entitled to 50% of those 26%) [anyone can correct me here if my math is off, I'm sleepy and it's very late, this thread was lloonng ]

Case 2: villain sucks out and wins the AKo vs AQo with his AQo, both hero and villain now have 1500 chips.

Then right after AKo vs AQo, next deal is all-in again this time JJ for hero vs villain's TT (JJ is 82% favorite vs TT, and we ignore ties again).

How bad should a SnG analyzer consider we're running if we now lose this JJ vs TT and lose the game?

Is a SnG luck analyzer saying: "we ran bad by $17.4 plus we ran bad by $16.4 ($20*0.82), meaning overall we ran bad by $33.8 in a SnG where the first place prize is $20" (which indeed would be kinda mindboggling).

Or is it saying about $16.932:

$20*0.74 + $20*0.26*0.5*0.82 which gives $16.932 ?

Or ?

Thanks in advance for any info,

TacticalCoder

Hand 1: $EV_luck = Actual final equity - Expected final equity = 10 - (0.74*20 + 0.26*10) = -$7.4

Hand 2: $EV_luck = Actual final equity - Expected final equity = 0 - (0.82*20 + 0.18*0) = -$16.4

Adjusted prize $ won = Actual prize $ won + Sum[$EV_luck] = 0 + (-7.4 + -16.4) = -$23.8

So to answer this:
Yes, it's possible to lose more than first prize.

Juk

I can see where you are getting at with this question too (I've considered this in the past...), but sadly the idea doesn't work. It appears to work (semi-)OK for your simple example:

Adjusted prize $ won = Actual prize $ won + P(enter state 1)*$EV_luck_1 + P(enter state 1)*P(enter state 2)*$EV_luck_2 = 0 + 1.0*-7.4 + 1.0*0.26*-16.4 = -$11.66

But if you were to try to apply this idea to a whole SNG then very quickly the probabilities of entering the states would get close to zero; meaning that you'd end up putting massive weight on early bad luck and almost none on late-game bad luck.

The best way to think about the way I've implemented it is this: if a "magic" entity were to give you back what you had lost in $-equity each time you went all-in (ie: using ICM for the n-way $EV hands and cEV for the 2-way hands) then the final value I store for each SNG (ie: -$23.8 in the simple example) would be what you ended up with at the end (ie: the actual prize plus the sum of all the $-equity you've lost).

Juk

Damn, must be a slow day now for I just woke up, thanks for the answer: first thing I checked when waking up

Actual final equity in case 1 is $20, 74% of the time hero wins $20, 26% of the time he wins $20*0.5. So $EV_luck = -$17.4 not -$7.4!?

I just woke up so I don't see how here can have an $EV_luck of only -$7.4 on AKo vs AQo and then -$16.4 on JJ vs TT when in both case he has respectively 74% then 82% change to win $20 (going to make myself another coffee : )

But then the corollary "it's possible to win more in $EV_luck than the first prize" it true too? You'd be slowly losing your stack in non-allins pot and constantly sucking out when not favorite and shortstacked.

Isn't the possibility to win more than the pot in $EV_luck less likely than the possibility to win less than the pot in $EV_luck and hence the results skewed towards showing good players (which are more likely to be on bigger stacks over the long term) running bad on AIEV?

I'm asking all this because I'm sitting on a correct and very fast algorithm to compute correct AIEV-adjusted even in the most complicated multi-ways pots one can imagine (involving different players all-ins at different streets etc.) that I'm using for Xypto and I was wondering how I could apply this to graph some SnG luck analyzer

What if I were to generate one million SnG made of only all-ins and analyzing the $EV_luck of all the players? What should I see?

The actual final equity you end up with after losing the first all-in is $10 (ie: 1500 chips). Your expected final equity for the hand is (0.74*20 + 0.26*10) = $17.4, but the difference (ie: the amount lost to luck) is 10-17.4 = -$7.4.

I've seen no sign of any bias and have run on nearly 70k of my own SNGs and have also tested it using huge simulated all-in samples generated from my own all-ins.

My code (and HEM's afaik) can deal with any all-in; on any street, and with any number of players involved. The only slight difference between mine and HEM's is that I just use Monte-Carlo roll-outs (instead of enumeration) when there are >=4 players all-in pre-flop or when the all-in occurs postflop (it makes hardly any difference anyway and that's why I've never bothered to implement the enumeration...).

I'm not sure if your Xypto app only works for cash games (ie: cEV), but if you want to apply it to SNGs you'll have to deal with all of the possible outcomes of an all-in separately (ie: you can't just deal with averaged equity like you can for cEV situations as the ICM function is non-linear).

Juk

PS: Sorry, I missed this bit:

The reason is that in the first hand the effective stack size is only 750 chips (which limits the amount of equity you can lose to bad luck), whereas after losing the all-in the effective stack size goes to 1500 chips...

Ah of course juk, it was a slow day indeed, now that caffeine kicked in and that I got your detailed answer it's all crystal clear, thank you very much

Great!

Offtopic but: Monte-Carlo when the all-in occurs postflop? There's at most C(45,2) cases to evaluate when there's an all-in postflop. I'd brute-force these 990 cases, not MC them!?

Yup, I really did read the whole thread yesterday and I know I have to deal with the different possible outcomes separately.

Thanks a lot for taking the time to answer my questions,

TacticalCoder

Yeah, it was just laziness really (I did plan on using enumeration, but never got round to it).

Juk

Will this work with PT3 processed hands?

a mirror??

It should do unless PT3 somehow changes the files?

See here: http://www.jukofyork.com/Juks_BackTe...uck_v1_03e.rar

Juk