Duvet Stuffing Part One

[peterbirks]

Just to show how frustrating this R&D stuff can be, here's an hour's work (without looking at a single specific hand, just looking at the stats).

I analyzed 27,000 hands from Jan 06 to Nov 30, and 16,000 hands from Dec 01 06 to the present, all on Ultimate Bet, at all levels, where I raised first in, regardless of position, and there had been no previous callers.

I analyzed it for all hands, and specifically for AKo.

For these two periods I was up $540 (in open play) for the first period, and down $301 for the second period. The net result after bonuses and rakeback was plus approx $1800 for the first period and plus $$500 for the second period.

One would, therefore, expect a deterioration between period one and period two.

Not so.

1: OVERALL FIGURES

Jan-Nov 2006: All hands, +0.38BB per hand (+$2,495).
Jan-Nov 2006: AKo, 0.35 BB per hand (+$187.50)

Since Dec 2006: All hands, + 0.51BB per hand (+$2082.75)
Since Dec 2006: AKo, +0.66BB per hand (+$149.25)

Jan-Nov 2006: Won when saw flop: 47.72%
Jan-Nov 2006: AKo won when saw flop: 51.28%

Since Dec 2006: Won when saw flop: 52.29%
Since Dec 2006: AKo won when saw flop: 57.14%

Jan-Nov 2006: Went to showdown: 33.36%
Jan-Nov 2006: Won $ at showdown: 61.52%
Jan-Nov 2006: AKo went to showdown: 36.36%
Jan-Nov 2006: AKo won$ at showdown: 62.5%

Since Dec 2006: Went to showdown 36.39%
Since Dec 2006: Won $ at showdown 62.99%
Since Dec 2006: AKo went to showdown 43.24%
Since Dec 2006: AKo won $ at showdown 62.5%


In other words, there's a consistent improvement of my performance across the board when I have raised first in.

I broke this down into hands that went to showdown, hands I won without showdown, and hands where I folded pre-showdown.

SHOWDOWN HANDS

Jan-Nov 2006: Went to showdown: won 1.23BB per hand ($1,901)
Jan-Nov 2006: Went to showdown: AKo won 0.93BB per hand ($120)

Since Dec 2006: Went to showdown won 1.38BB per hand ($1,454)
Since Dec 2006: Went to showdown AKo won 1.27 BB per hand ($80)

Interesting to note that AKo actually does worse than average at showdown than the "average" raising hand. However, figures are once again better for the past five months than the previous eleven months.



I FOLD PRE-SHOWDOWN

Jan-Nov 2006: Folded for -1.81BB per hand (minus $3116)
Jan-Nov 2006 AKo folded for -2.02BB per hand (minus $193)

Since Dec 2006: Folded for -1.76BB per hand (minus $1,706)
Since Dec 2006: AKo folded for -1.91BB per hand (minus $90)

Although the figures are an improvement, and although the net figures for AKo are positive, it's interesting that AKo wins less than average when you go to showdown, but loses more than average when you fold. Is this inevitable? Or am I continuation-betting too often? Note that AKo wins pots more often than the average hand when you see a flop, so the only conclusion you can reach is that it wins small pots and loses bigger ones, because (presumably) of the continuation bets when you miss.

A possible conclusion from this is that I shoulc continuation bet less. I've already abandoned a CB if I have three or more opponents and I miss. I'm wondering about abandoning it if I have two or more opponents and I miss, particularly if the texture of the board is distinctly horrible (say, J98 two of a suit or three of a suit).

The final numbers just complete the picture, hands where I won without showdown:

Jan-Nov 2006: Won without showdown for 1.16BB per hand ($3,720)
Jan-Nov 2006: AKo won without showdown for 1.16BB per hand ($246)
Since Dec 2006: Won without showdown for 1.17 BB per hand ($2354)
Since Dec 2006: AKo Won without showdown for 1.16BB per hand ($126)

Is this a statistical anomaly that all the numbers are so similar? Probably not.

For all my frequent raising hands, the range was from 0.96 to 1.5 BB per hand (we can perhaps look at the performance of various hands when we do go to showdown at another time). For my "rarer" raises (poorer hands that I would only raise with as a steal, and then not 100% of the time), the average win per hand looks slightly higher, but that might be counterbalanced by a greater number of preflop folds and a poorer performance at showdown. That might be worth looking at.

So, conclusions.

My raising first in has improved, so this is not the reason for the deterioration in my results.

However, some general points are of interest. 1) AKo does worse than average when you raise with it first in. (Interestingly, it does better than average if you raise first in with it from the CO or the button). I suspect this might be because a number of opponents "put" you on AKo as the default if you raise from a non-steal position.

Possible conclusion. Limp first-in with AKo from non-steal positions?

Or, possibly, limp first in with AKo half the time? Then you can either call or reraise a raiser behind (depending on that raiser's aggression factor and the number of other players in). If there is no raiser behind, you just get away from the hand if you miss. If you hit, play it aggressively to the end, going into call-down mode if there is a lot of action against you.

And I must remember that this is just for Ultimate Bet. It will be interesting to see how the numbers for Party (where I have a far larger hand database) and Full Tilt (smaller database) compare.

Comments

[jellymillion.livejournal.com]

Possible conclusion. Limp first-in with AKo from non-steal positions?

Alternative conclusion. Accept that you're playing AKo pretty well and move on?

This is good stuff. Make some more categories and dig some more.

[(Anonymous)]

Pete,
I have hard time to understand this analysis of yours. What exactly are you trying to find from these numbers?

You conclude that "My raising first in has improved, so this is not the reason for the deterioration in my results." It's easy to improve that more, only raise with Aces. Sadly that improvement is not directly transferable to your bank account, quite the contrary.

I would recommend to put the actual $$ results aside for awhile, and attack the problem from "how I played then vs now / how they played then vs now" basis.

In 2006 the pots which were open raised and won without the showdown were typically won either preflop (+0.75BB) or flop (+1.25BB when BB defends). Same goes on this year too ;)

gl
Aksu

[peterbirks.livejournal.com]

Hi Aksu:

You didn't understand my analysis, and I didn't understand your response!

I put the actual dollar figures in as well, as well as the hand sample sizes.

Look at those (actual $$ numbers) before you criticise my statement "My raising first in has improved, so this is not the reason for the deterioration in my results".

Why spend time looking at particular hands in this section if my overall results have got better? Sure, perhaps they could get better still, but I'm looking for areas of deterioration to explain the fall-off in the bottom line (see opening paragraph).

When I find that area, then I look at the "how I played then and how they played then vs how I play now and how they play now".

I'd love to do that for every type of hand, but I have to prioritize, and prioritizing in an area where my results have got better would be stupid.

I don't understand your final sentence, I fear. Where do you get those numbers from? Where's your data? What's the sample size? What do you mean by "typically"? (See my previous post about banishing such words from poker analysis).

I'm sorry that you don't understand the analysis. It really is laid out very clearly. Perhaps you mean that you don't understand why I took this methodology? That's a different statement.

Pete

[peterbirks.livejournal.com]

Fuck, I got so worked up thinking about this and was in such a rush to post something, that I cut myself shaving! God, the grief that this blog can bring me.

Referring to your last statement first, Aksu, it's what I'd call a typical 2+2 witticism, in that it's short, pithy, and untrue.

It implies, for example, that the Big Blind is 0.5B worse off if he defends than if he folds. Clearly if this were the case, the big blind would never defend. Also (clearly), there is an empirical diffierence between his result when he folds (-0.5BB) and when he defends (< -0.5BB). And this is a very important area for research. It really doesn't deserve such a dismissive comment.


To further demolish your intimation that my percentages have improved because I have tightened up my raising standars (even though that line is comprehensively disproved by the cash figures mentioned above), here are some stats. I try to keep my posts to a manageable length, so I do not include numbers that (I feel) do not add to the information gleaed.

Jan-Nov 2006: VPIP 16.62%: PF raise 10.92%. Went to SD 28.7% Won $ at SD 59.36%
Since Dec 2006: VPIP 16.88%: PF Raise 11.42% Went to SD 29.89% Won $ At SD 57.65%

So, my raising range has, if anything, widened slightly (I've raised with more hands, not fewer). By your argument, this should mean that my BB per 100 performance in the sample should deteriorate (even though the bottom line might improve), since if you tighten up to just AA, your BB improves, but your bottom line deteriorates.

However, as the numbers show, my BB per 100 has got better.

Therefore, any changes that I have made have been for the better, not worse.

Your "look at how they play and how I play now compared to then" line sounds attractive, but what about the sample size? How does one know that an individual hand is "typical". Looking at the numbers as a whole gives a better overall picture of what is happening, without spending an hor looking at 12 hands that might be atypical.

Eventually you do need to look at specific hands, but that is down the line.

And now I've missed my train, and i'm still bleeding. Oh bollocks :-)

PJ

[(Anonymous)]

Pete,
my last statement refers strictly to this:
"
The final numbers just complete the picture, hands where I won without showdown:

Jan-Nov 2006: Won without showdown for 1.16BB per hand ($3,720)
Jan-Nov 2006: AKo won without showdown for 1.16BB per hand ($246)
Since Dec 2006: Won without showdown for 1.17 BB per hand ($2354)
Since Dec 2006: AKo Won without showdown for 1.16BB per hand ($126)

Is this a statistical anomaly that all the numbers are so similar? Probably not."

I think it is/was both, I did not understand the analysis nor the methodology.

Now that you have put in the PF raise percentages it becomes more understandable, but still I doubt that one can draw much conclusion from the results itself. In original analysis you have the actual $$ numbers for AK and all hands. You do have the meta information to interpret this number, but I do not. What does cumulative sum of winnings from two different sized datasets across various limits mean? To me it means nothing.

Sample size? 43000 hands is fine to try to determine your overall win rate, given that games are stable. Now you split the data for two quite randomly and throw most of the hands to garbage bin and concentrate on specific type of hands. The fact that this stabilizes some of the conditions is in your side. But still I think that you have way too much noise to make any conclusions from the results. Also you are not doing this only for raising first in hands, right? Given that you do this for limped in, called cold, 3 bet defending SB and bunch of other situations, it's quite clear that the differences you'll find have not much value (except maybe if they are huge).

Now do not get me wrong. There is much information you can find from PT data. And I'm sure you have already find some gems. But do not let results to drive you analysis this much.

About the witticism. I hope that my comment do not sound that mean spirited to you anymore. It was not meant to that very witty (a bit yes I admit). Anyhow my comments have certain kind of bias. I'm too lazy and bad writer to give much positive feedback. I enjoy the block, keep it going.

Aksu

[peterbirks.livejournal.com]

The 43,000 hands is the total dataset. I split this between Jan to Nov 2006 and since Dec 2006 because that was when the overall results seemed to start deteriorating. It wasn't a random split. However, it might have made things easier to understand if I had split the data into two equal data sizes. To be frank, I just didn't want to go back to start again.

However, the percentage of raises first in (i.e., the subsets) are proportional for both periods.

I could have mentioned that the proportion of limits played, etc, had not changed, but this would have made the post even longer than it already was.

In other words, if other conditions had changed (the levels I was playing, my proportion of raises, etc), I would have mentioned them.

"Also you are not doing this only for raising first in hands, right? Given that you do this for limped in, called cold, 3 bet defending SB and bunch of other situations, it's quite clear that the differences you'll find have not much value (except maybe if they are huge)."

I'm not sure what you mean here. Do you mean my plans for the future, or what I have already done?

If you mean my plans for the future, then, yes, I hope to look at any situations that I can easily extract from PT. Each of which takes time. What I'm looking for is a situation where my returns have deteriorated significantly, and then to search for the reasons for that deterioration (am I dooing something different? Are my opponents doing something different? Have I just run bad there?)

I'm still surprised that you feel that the differences I find will not have statistical value and that the dataset is too small, with any alterations being attributable to "noise". Put plainly, I really think that you are wrong, here.

I was going to analyse the same situations in Party, to see if any differences cropped up between Party and UB.

Through this analysis I come up with hypotheses of future strategies. Then I test the hypothseses (via more checking of the data, this time on a lower level of individual hands, the performance of other players who have used such strategies and, eventually, by trying it out myself). I think that this is a more rigorous method than a sort of wooly replaying of a small sample of hands and trying to come up with some magical "oh, if I had done this, I would have won" or "oh, he three-bet here with AJo, that means that's what all players are doing to me now".

I do appreciate the feedback Aksi - perhaps 6am when I am bleeding from the chin and rushing for a train is not the best time to catch me... :-)

But I'm always surprised when good and successful players pooh-pooh what to me seems a sensible method of analysis. Perhaps my brain just works a different way, but I really do think that the stuff that can be gleaned from these numbers is useful...

PJ

[(Anonymous)]

I'm not really saying that all alterations are attributable to "noise". I am saying that you just can't tell if it is noise or genuine find. Hence I recommended to hop right to the "am I dooing something different? Are my opponents doing something different?"-phase.

My use of term random is quite broad. By saying that you split the data in two randomly, I'm implying that your later results maybe due to bad luck and therefore split point is random. In any case there really is not strict time point when the games change from this kind to that kind.

For comparison I splitted 100k database in two (completely artificial timepoint split) and checked same numbers as you:
First group: All hands +0.42BB AKo +0.40BB
Other group: All hands +0.32BB AKo +0.87BB

I cannot conclude anything from these numbers. I do not even have any feel for variance of these estimates. And even if I did theres a slight problem. Imagine that I do 100 statistical analyses for finding differences in 100 situations. Using the convential 5% significance rule, I would find roughly 5 significant findings from complete noise. That's what I was speaking in multiple scenario analysis part.

Now, of course you may find something truly important this way, but I imagine that you would know where to look for it allready if it's in there. In any case you really cannot hop to conclusions based on the small differences in the results, like you partly did with the limping AK early thing.

BTW if you are handy with the ACCESS databases (or that other one PT uses) and have time and will, you could export data to some stat prog and do simulations and such to get some feel to variances etc. http://cran.r-project.org/ is good and free toy for that kind of purposes.

Anyway, report the findings, I'm nearly allways ready to crush 'em =)

Aksu

[peterbirks.livejournal.com]

OK, I'll concede that your numbers are interesting there.

However, if I take a number of scenarios (the above one, and various other ones which you mention, plus a few others that I'm thinking about) and some scenarios show an improvement, while others show a marked deterioration, is it not more likely that the area of marked deterioration is a contributory cause to the overall deterioration?

So, I look for areas of deterioration and I study those areas closely first. If I jump straight into the "am i doing something different, are my opponents doing something different?" phase (which, I should point out, I hypothesised in a a certain area in the first bit above) then there is no methodology. It's just some kind of heuristic mess as you play hands back. I can't replay ALL the hands, and then take notes, trying to gather significant findings from them at the end. First, this is so slow that I get bored very quickly. Second, I am even more at risk of taking small samples as "representative", when they may not be representative at all.

It is also possible that I'm just in a long-term bad run, but my feeling is that is not so. My best way of finding out how it is not so is to analyze the data that I have. Sure, this is subject to noise, but if one result is "5" and the other is "10", with a noise factor of 50, that still means that the first result is more likely to be less than the second.

You have to work with the data that you have.

Unfortunately, I'm not a wizz with the Access databases (and I don't even have Access on my machines), and I don't have the time and will, so an export to a stat problem (which would probably bamboozle me anyway) is not an option.

PJ

[(Anonymous)]

For Pete's sake ... er .. Pete,

stop wasting all this time analysing the square-root of fuck all and come down to a real casino and pick up some of the money that people are chucking away with no thought at all. There's less to this than meets the eye.

DY