Add this month's results so far (where i've run at an awesome 0.94ROI over 733 SnGs) to my last post:





Luck: -849.6%
Bucks: -3105.32$

i run so good

</end of "i hate poker boo hoo hijack rant">

May I ask where your ROI of 0.94% would end if you add these 3105.32$ as earnings?

Would it be a ROI that you think is realistic for your poker skills?

I'm not sure - i don't know how to work that out. All I know is that my roi for the first sample of 5815 hands was probably somewhere between about 7-9%. (I can check the exact number later this evening).

It's still using 60/40 for all 6-max? If you have to pick one can we do 65/35 since that's what Stars/FT use?

It's the dooooooooomswitch! I just wish they would turn mine off for a little bit...

Juk

My exact ROI for my first sample of 5815 hands was 7.44%. Assuming this is my accurate current roi (it's probably higher because this sample set included a learning curve) then my expected earn for my second sample set would be $2690.38 (the second sample set is now 751 SnGs @ 1.21roi - i've played a few more since my previous post).

So therefore, in answer to the question, if i add the 3105.32% it would give a realistic roi for my poker skills.

Has anyone else tried this software with a decent sample size of party hands? I'm curious as to what sort of results other people have.

I wonder how much that people have folded hands before we push ("bunching effect"?) affects the results...

I am more likely to be pushing all in if there has been no action before me. There being no action before me I think indicates there is a smaller chance of high cards being in the folded players hands, and so a higher chance of high cards in the remaining deck. As the callers of our pushes are more likely to do so with high cards, their hand may connect with the remaining deck more frequently meaning that their equity is actually higher (and so ours lower) than in a simple random deck with only the knowledge of each players hole cards.

Thoughts?




With regard to the date selector, I tried using "Windows Classic" theme rather than using the standard vista aero thing and the date selector works fine.

I think this won't make much of a difference, especially considering the most equity jumps around when it's three or four handed.

Are you having the feeling it doesn't use the 65/35? Actually it should choose the right payout structure automatically:

Paradise Poker: 70 | 30 - Full Tilt: 65 | 35 - Pokerstars: 65 | 35 - Party: 60 | 40 - Ultimate Bet: 70 | 30 - Absolute Poker: 65 | 35 - Prima: 65 | 35 - Poker Room: 65 | 35 - Cryptologi: 65 | 35 - B2B: 65 | 35 - All others: 65 | 35

But I haven't double-checked many 6-max games yet. Possible that it is still buggy in that point. Also, I'm not quite sure if all of these are correct. I keep those updated in this little FAQ

Not sure if I got you right.

In most cases the biggest part of an estimated +EV push comes from the fact that we expect our opponent to fold.

If our opponent calls with the top 10% of hands and we are holding aces which are a substantial part of his calling range then his calling propability is only 7.8% instead of 10%. Right?

Now, you are saying that IF we get called, we shouldnt expect a random deck but a more high-cards weighted deck?

I don't see how this has an effect on our luck. If you have the better hand you are expected to win. If you loose, you are unlucky.

I was rather saying that on the occasions we push we shouldn't expect a random deck (which only matters to SnG LA when we are called). For example it is folded to us in the SB 6 handed, we push 72o and BB calls with AKo. It seems much more likely that the 4 opponents who folded were holding sevens and twos, than aces or kings, and so 7s and 2s more likely "dead", and so the value used for Win % slightly inaccurate.

I tried playing around with PokerStove though and the effect on equity seems smaller than I imagined. Also, as Slim Pickens says equity changes most when we are short handed and the effect of folded hands is even smaller then.

The effect is probably so incredibly small it isn't worth thinking about.

The effect is also reversible when you are the one calling, typically your own hand selection for calling will be tighter than average, although you don't call as often the effect would be greater. So even if the effect was significant it should even itself out to some extent.

Sorry, made a computational mistake as usual. It looks fine.

Bodypull, What does the "Ended" filter do?

I've only imported my last two days of 22s so far:



I'm down $1088 (in Sklansky bucks) in 234 SNGs, but only down ~$450 in real cash, so if all had gone as expected I should be up $638?

Has anybody else tried this with a bigger sample of Party hands and actually been lucky? I have about 25000 Party 11s/22s/33s that I could run it on, but they are all in my PGSQL DB and it's sooooo slow to re-import them into Access... Any ETA on the PGSQL support?

I also have (or had) about 8000 Pacific SNGs in an Access DB - will it work with Pacific as there are no summaries for Pacific SNGs?

EDIT: Also, to be sure this is working properly, would it not be possible to backtest it using datamined hands? Party shows mucked hands at showdown, so there is no reason you couldn't backtest all the all-ins and see if your app is actually converging.

Juk

PS: I'm just in the process of importing ~2.5k SNGs from December (where I think I ran well). Will post the results when Access finally stops chugging...

The real-time monitoring is a useless feature and I think it might be behind a lot of the random freezing. I would like to disable it. It's way more important to include the non-Hero all-in hands in the luck calculation.

Agree that real-time is not important to me. Especially if it affects the performance.

Well after 4 hours of importing, I managed to get 2858 SNGs into an Access DB:



What's interesting is that the hands I imported I felt like I had ran well (the hands from today, I knew I had run bad), but the graph just seems to be slowly diverging?


Well while I was waiting for PT to do it's import, I quickly hacked together some of my code to try to repeat your experiment:

A) I used my own Party HH parser, and read the hands direct (so no possibility that PT is causing problems).
B) I only considered heads-up all-in situations (not sure if you are considering 3-way+ all-ins).
C) I also just broke all-ins down into:

i) Open: you open push all-in and go into the flop vs a single opponent.
ii) NoOpen: you push (or call) all-in over previous bet(s) and go into the flop vs a single opponent.

What's amazing is I get almost the same "gut-wrenching" bad luck using my code!

NumSNGs = 2858
TotalLuck = -6639% (Open=-5676%, NoOpen=-962%)
PerHandLuck: Open=-2.0022%, NoOpen=-0.480762%
(So on average, I'm losing ~2% for every push and ~0.5% for every call.)

And just to see, I also ran using cEV instead of $EV:

NumSNGs = 2858
TotalLuck: -20139597 (Open=5498829, NoOpen=-25638427)
PerHandLuck: Open=19.3962, NoOpen=-128
(So on average, I'm getting an extra 19.3 chips for every push and losing 128 chips for every call.)


So now I'm sure we are getting similar results, I tried re-running on a much bigger set of SNGs:

NumSNGs = 21772
TotalLuck: -43422% (Open=-36790%, NoOpen=-6631%)
PerHandLuck: Open=-1.63908, NoOpen=-0.436752
(basically this starts to converge onto the numbers quite quickly and I doubt running many more would alter the output much...)

NumSNGs = 21772
TotalLuck: -175183873 (Open=28041787, NoOpen=-203225661)
PerHandLuck: Open=12.493, NoOpen=-133.842


So either:

1) We have both got the same bug (unlikely).
2) We have both messed up our calculations in the same way (possible).
3) My sample of 20k games is not really that big and if I were to add more it would start to converge (very unlikely).
4) There is some fundamental flaw in this idea and it won't converge to "even luck" however many samples (seems strange that cEV diverges too...).

I'll have another look though my code again tommorow (when I'm not so sleepy...), just to make sure I haven't messed up somewhere, but I'm 95%+ confident I haven't.

Any ideas on the phantom Party Poker divergence?

Juk

PS: And I would also suggest getting away from using DBs... It took me 4 hours to import just 2.5k SNGs and then SnG Luck Analyzer spent well over 1/2 an hour reading them back in - without using PT I managed to parse and backtest vs all 21k SNGs in less than 10 minutes.

Scrub that, there's a bug with my cEV amounts... will fix and re-post tomorrow after some sleep.

Juk

Well I've fixed yesterday's bug and seem to be getting much more sensible results now:

NumSNGs = 22076 / NumDataPoints = 53470
$EV Luck: -55% (Open=1035%[0.0325791%] NoOpen=-1090%[-0.0502656%])
cEV Luck: 400900t (Open=698536t[21.9818t] NoOpen=-297635t[-13.721t])
Mean P(lose) difference: -0.324163%



After running on 22000 SNGs the final results are within 55% of "even luck" and the graph looks "sensible" based on my own feelings about luck (ie: ran well to start with, then started to run a bit worse, then very recently ran very bad).

One thing that's interesting is the "Mean P(lose) difference" is still slightly off even after all those games and it appears that I've been winning slightly more than I should have been (ie: -0.3% means that if I should lose an all-in 60% of the time, I'm actually losing only 58.7% of the time). I also plotted the cumulative version of this and it does look like it needs more data to converge and from just this it's not obvious if some bias is or is not effecting it (maybe I was just lucky over those 22k SNGs...):




I also ran against the same data as I imported into an Access DB for "SnG Luck Analyzer" to look at. Here are my results for just that data:

NumSNGs = 2858 / NumDataPoints = 7011
$EV Luck: -1042% (Open=-691%[-0.168734%] NoOpen=-351%[-0.120523%])
cEV Luck: -280108t (Open=-198708t[-48.501t] NoOpen=-81399t[-27.934t])
Mean P(lose) difference: 0.643722% {Note how "Mean P(lose) difference" is +ve for this sample...}



Compair this with the output of "SnG Luck Analyzer", which seems to just be slowly diverging towards negative luck and has me down as having nearly 3x worse luck (-2864.6% instead of -1042%):



I couldn't face importing the whole 22k SNGs into Access DB, but it does seem that we are getting very different results for my recent data. I also "feel" that the results of my last 2858 games fits much closer to my own graph... I felt like I ran fairly hot in December and then started running really bad in January and my graph shows that almost exactly. Also, I don't think I'm down nearly $6k from just the recent downswing (-$2K seems much more sensible based on my past experiences, winrate, etc...)

I'll look into posting the exe sometime later today (or tomorrow), but atm I'm having issues with the Intel Compiler not wanting to let me revert projects back to use plain VS2005 (my current exe will only execute on a Core 2 Duo CPU...).

Also, don't expect anything flashy - all my exe outputs is the results and a ".csv" file with some debugging info (so the results can be compared manually with the respective HH file) and the cumulative EVs, etc that I plotted above. My goal is not at all to replace this great util (and idea!), but simply to verify it's results for Party SNGs.

Juk

EDIT: Posted bad cEV values and bad seperated "NoOpen" EVs - fixed now.

Excellent post. And thanks for sharing your outputs. No worries, I'm happy to have co-existent solutions. One reason I'm developing this tool is because I NEED IT and it doesn't exist yet...

One question before we start comparing: are your results taking only 2-way all-ins in consideration?

Hi,

I created a new DB for only my stars sngs (250) and tried to import, see what happend:

Weird error message...

Are you sure, that your new database is working correctly even in Pokertracker?

Sometimes, if you create a database from scratch wihtout copying any settings from other databases, it doesn't set up well.

I usually copy at least the "import settings" from an existing database just to avoid this.

Just wanted to add my thanks for creating such a great tool. I am a degree qualified Chartered Engineer with a solid understanding of advanced maths but the intelligence required to undertake these tasks blows my mind, and to start bringing it all together into a program like this takes some serious brainpower and talent. Well done! It's worked great for me so far. All the best with it for the future. The market has been crying out for a tool exactly like this for ages.

hehe, just imported the Stars Summary but no hands!! That's the problem. My Fault

Yep, I'm only considering 2-way all-ins. I class a 2-way all-in as either:

A) We put all our chips in the pot on our first action (push or call all-in). Whatever the action was before or after, it's still a 2-way all-in so long as only one opponent remains when we leave the pre-flop stage (all dead bets from other players are properly handled, so there can be raises and folds from other players, eg: UTG raises, we push, BB calls, UTG folds, etc are all handled correctly...).

B) We make a call for less than our total stack and there are no non-all-in players left (ie: this can't handle BTN pushing and us just calling from SB as BB isn't all-in already, but it can handle say BTN push all-in, SB fold and we call in BB).

(I can fix these to handle any 2-way all-in where we are all-in with a single a opponent as we leave the pre-flop stage, but atm that's what it does and that's why I'm getting slightly less data-points per SNG...).

I also take ties into account and have three separate outcomes for every all-in (ie: we lose, we win, we tie).

I think a much easier way will be to compare outputs for the same HH files. What I've done it output a ".csv" file which has a row for every all-in, eg:

When I post the exe you should just be able to run it on your own hands and compare my output with yours.

EDIT: Made my example ".csv" file easier to understand - I was using the term "EV" when I meant "Equity" and since dollar equity is measured in terms of "% of prize pool", it seems clearer to measure chip equity in terms of "our total chips" rather than "our chips gained/lost since start of hand".

Juk