EV Software
12-06-2009
, 03:17 PM
This might help, but it just does the calculations using the hand histories themselves (ie: no need for PT3 or HEM):
http://forumserver.twoplustwo.com/45...alyzer-598964/
Be advised though that what you are calling "true EV" has some pretty fundamental flaws associated with it (ie: it can be badly biased and make good players appear to always be unlucky...). Read the posts by contravariance in that thread (or just do a search of 2+2 for: "street EV", "street by street EV", "weighted EV", etc) before putting too much confidence into the results you get.
All-in EV adjustment on the other hand, does not suffer from these biases so you have the choice of either looking at all-in EV only and being 100% sure that the adjustment is mathematically correct or looking at street-by-street EV; which will remove more of the winrate variance, but at the cost of never being sure if the results are meaningless because of the biases associated with the method.
Juk
http://forumserver.twoplustwo.com/45...alyzer-598964/
Be advised though that what you are calling "true EV" has some pretty fundamental flaws associated with it (ie: it can be badly biased and make good players appear to always be unlucky...). Read the posts by contravariance in that thread (or just do a search of 2+2 for: "street EV", "street by street EV", "weighted EV", etc) before putting too much confidence into the results you get.
All-in EV adjustment on the other hand, does not suffer from these biases so you have the choice of either looking at all-in EV only and being 100% sure that the adjustment is mathematically correct or looking at street-by-street EV; which will remove more of the winrate variance, but at the cost of never being sure if the results are meaningless because of the biases associated with the method.
Juk
12-06-2009
, 06:26 PM
Hi there, I've written a tracker & HUD for OS X that correctly computes all-in adjusted EV. It's not as advanced yet as what jukofyork wrote for SnG (that is: I'm only doing it for cash games at the moment, no SnG AIEV yet) but it's still a topic I know a bit and I'd like to contribute to this thread
I 100% side with what jukofyork said: what you call "true EV" has some pretty fundamental flaws.
Sidetracking a little bit: some computational models make assumptions: for example the ICM assumes that players are of equals skills.
If we now assume that players are of equals skills for cash games, I can tell you what your "true EV" is, taking all the deals into account (both deals that went to showdown and those that didn't): its "- rake" (or -(rake - rakeback) but lets keep this simple).
That's the one true "expected value": in the end all players of equals skills are break even and losing the rake.
So if you're running above "- rake", you're running good while if you're losing more than the rake, you're running bad.
It may seem a little bit far fetched, but it really aint': it is only assuming players are of equal skills.
Now people can say: "wait, I only want the 'true EV' for deals that went to showdown" and that's precisely where the problem lies.
You cannot do such a thing without suffering bias (due to all the hands your opponents folded which you'll never know) and hence there's no way to know if the results have any significance at all or not, as said jukofyork:
So you can compute "something", but you wouldn't be able to tell if that something as any meaning or not.
In any case, the "something" you could compute would not be the "true EV", that's a certainty.
To me, in addition to being surely not the "true EV", it has very little meaning and hence it's a feature I'm not putting into Xypto (my tracker & HUD).
Interestingly enough, it means that "adjusting EV" cannot be done for fixed limit.
For No-Limit, you could look at your AIEV results and consider them to be your true results while looking at your real $ results and consider them to be your variance-influenced results.
Here's a cropped screenshot of one of my $25NL fullring session seen in my own tracker:
In addition to being a super-nit at 11.9/7.4 and only being involved in 9 all-ins over 1935 deals, it shows me I'm running very lucky (making $185.25 instead of $137.66). I consider the $137.66 to be my "real results" and the $185.25 I really made to be "all-in variance-influenced".
I'm mathematically inclined and "exactitude" is very important to me: that's why instead of displaying seemingly random numbers you can hover over any stat and see a "xxx deals contributed to this value (for example over ATSB it shall show 55/212). Then, for example, you can open any deal contributing to an AIEV and open it in the replayer:
(numbers add up to 99.99% and not 100% just like in PokerStove sometimes they add up to 99.999% and not 100%, but that's just a display thing)
It's not to plug my tracker (it's OS X only anyway) but it's really to emphasis that AIEV is about math. It's backed by indisputable numbers. Any AIEV number displayed in Xypto can be explained and verified. And if someone finds a mistake, I'll fix it.
While any other kind of "adjusted EV" is not about math: it's vague, very vague. There's no accepted formula. On the contrary, quite a lot of people dispute the very idea of any other kind of EV being meaningful and some even dismiss the concept entirely.
The only thing we know for sure is that any kind of non-AIEV is suffering from bias. It's related to the incomplete information nature of the Poker games and the fact that the only moment where complete information is available is when there's an all-in with cards to come.
I don't want to produce biased graphs in Xypto, so I decided to only display the AIEV graph (and one of these days I'll extend that to SnG, like jukofyork did).
Of course AIEV doesn't tell everything: for example you'd be unlucky to face AA 10 times when you have KK all-in preflop while only ever facing "AA vs AA" when you have AA preflop.
To summarize, as jukofyork said, AIEV is mathematically correct while any other kind of EV suffers from fundamental flaws and is biased.
To me the only 'true EV' is the AIEV and it is only applicable to no-limit cash games and, in a more complicated version, to no-limit tourneys/SnGs.
And what about all the cases where you put $1 preflop and everyone folded? Or the case where you put $2 on the flop and everyone folded?
That's the fundamental flaw in computing such kind of EV: you're biasing the results towards hands that go to showdown. These shall tend to be the hands where the opponent either hit or hit a draw (and the fact that you can also plot a non-showdown graph doesn't make non-AIEV less biased).
It has been pointed out that such flaws in reasoning have been detailed in articles by the Pokersleuth author, notably in his article about the Defense Attorney's fallacy.
What you want to do is to focus on the unlikelyhood of an event (opponent sucking out while he was only 20% favorite) while neglecting the information available that make this event more likely.
What makes that event more likely is that in many deals your opponents folded and you're not taking these deals into account.
Non-AIEV is biased and the very concept of a non-AIEV is based on the Defense Attorney's fallacy.
Unless you assume players of equals skills, where hence the one true EV is -(rake - rakeback) and which is a very simply line to plot
jukofyork adviced you to search the forums: you'll see lots of logical explanation detailing why non-AIEV is biased and exactly zero logical or mathematical proof explaining how to compute a non-biased non-AIEV.
oh well, that's was my contribution and cute screenshots to this thread
TacticalCoder
I 100% side with what jukofyork said: what you call "true EV" has some pretty fundamental flaws.
Sidetracking a little bit: some computational models make assumptions: for example the ICM assumes that players are of equals skills.
If we now assume that players are of equals skills for cash games, I can tell you what your "true EV" is, taking all the deals into account (both deals that went to showdown and those that didn't): its "- rake" (or -(rake - rakeback) but lets keep this simple).
That's the one true "expected value": in the end all players of equals skills are break even and losing the rake.
So if you're running above "- rake", you're running good while if you're losing more than the rake, you're running bad.
It may seem a little bit far fetched, but it really aint': it is only assuming players are of equal skills.
Now people can say: "wait, I only want the 'true EV' for deals that went to showdown" and that's precisely where the problem lies.
You cannot do such a thing without suffering bias (due to all the hands your opponents folded which you'll never know) and hence there's no way to know if the results have any significance at all or not, as said jukofyork:
In any case, the "something" you could compute would not be the "true EV", that's a certainty.
To me, in addition to being surely not the "true EV", it has very little meaning and hence it's a feature I'm not putting into Xypto (my tracker & HUD).
Interestingly enough, it means that "adjusting EV" cannot be done for fixed limit.
For No-Limit, you could look at your AIEV results and consider them to be your true results while looking at your real $ results and consider them to be your variance-influenced results.
Here's a cropped screenshot of one of my $25NL fullring session seen in my own tracker:
In addition to being a super-nit at 11.9/7.4 and only being involved in 9 all-ins over 1935 deals, it shows me I'm running very lucky (making $185.25 instead of $137.66). I consider the $137.66 to be my "real results" and the $185.25 I really made to be "all-in variance-influenced".
I'm mathematically inclined and "exactitude" is very important to me: that's why instead of displaying seemingly random numbers you can hover over any stat and see a "xxx deals contributed to this value (for example over ATSB it shall show 55/212). Then, for example, you can open any deal contributing to an AIEV and open it in the replayer:
(numbers add up to 99.99% and not 100% just like in PokerStove sometimes they add up to 99.999% and not 100%, but that's just a display thing)
It's not to plug my tracker (it's OS X only anyway) but it's really to emphasis that AIEV is about math. It's backed by indisputable numbers. Any AIEV number displayed in Xypto can be explained and verified. And if someone finds a mistake, I'll fix it.
While any other kind of "adjusted EV" is not about math: it's vague, very vague. There's no accepted formula. On the contrary, quite a lot of people dispute the very idea of any other kind of EV being meaningful and some even dismiss the concept entirely.
The only thing we know for sure is that any kind of non-AIEV is suffering from bias. It's related to the incomplete information nature of the Poker games and the fact that the only moment where complete information is available is when there's an all-in with cards to come.
I don't want to produce biased graphs in Xypto, so I decided to only display the AIEV graph (and one of these days I'll extend that to SnG, like jukofyork did).
Of course AIEV doesn't tell everything: for example you'd be unlucky to face AA 10 times when you have KK all-in preflop while only ever facing "AA vs AA" when you have AA preflop.
To summarize, as jukofyork said, AIEV is mathematically correct while any other kind of EV suffers from fundamental flaws and is biased.
To me the only 'true EV' is the AIEV and it is only applicable to no-limit cash games and, in a more complicated version, to no-limit tourneys/SnGs.
Quote:
I want to know that if I put in $1 pre-flop as an 80/20 favorite and another $2 on the flop as a 60/40 favorite and then no money on the turn and river and ended up losing the hand that my actual result was -$3 but my "Sklansky" result was +$1 (which I believe should be calculated .8 minus .2 x $1 + .6 minus .4 x $2).
That's the fundamental flaw in computing such kind of EV: you're biasing the results towards hands that go to showdown. These shall tend to be the hands where the opponent either hit or hit a draw (and the fact that you can also plot a non-showdown graph doesn't make non-AIEV less biased).
It has been pointed out that such flaws in reasoning have been detailed in articles by the Pokersleuth author, notably in his article about the Defense Attorney's fallacy.
What you want to do is to focus on the unlikelyhood of an event (opponent sucking out while he was only 20% favorite) while neglecting the information available that make this event more likely.
What makes that event more likely is that in many deals your opponents folded and you're not taking these deals into account.
Non-AIEV is biased and the very concept of a non-AIEV is based on the Defense Attorney's fallacy.
Unless you assume players of equals skills, where hence the one true EV is -(rake - rakeback) and which is a very simply line to plot
jukofyork adviced you to search the forums: you'll see lots of logical explanation detailing why non-AIEV is biased and exactly zero logical or mathematical proof explaining how to compute a non-biased non-AIEV.
oh well, that's was my contribution and cute screenshots to this thread
TacticalCoder