View Single Post
  #1  
Old 09-02-2005, 08:07 PM
Guruman Guruman is offline
Senior Member
 
Join Date: Jan 2005
Posts: 228
Default reducing relevant sample size for analysis (long)

I'm currently working on a way to analyze my play on a periodic basis that reduces the effect of insufficient sample size. I'd like some opinions as to how instructive this little exercize would be.

Here's the setup:

Expert (and winning) poker play comes from one's ability to do a couple of things better than your opponents. In order of importance those things are:

1)put your opponent on a range of hands

--IMO one of the single most important parts of poker. If you could see your opponents cards almost every decision becomes incredibly obvious. Sklansky's FTOP speaks to this effect.--

2)Make the correct decisions based your read of the opponents hands.

--If I have TPTK, you have a set, and I know you have a set, then I shouldn't be checkraising you. Also, I'd better have the technical knowledge to understand that my middle pair and overcard with a nut-flush draw has tons of equity versus your drawless top pair no kicker hand.--

3)disguise your hand

--this means slow-playing when it will get more action and bluffing when it will make better hands fold. This does not mean slow-playing when your opponent will pay you off anyway or bluffing when he loves his hand. Obviously this should also be heavily tied to hand reading.--

Now comes the challenge. If I want to formulate a reasonable way to measure my effectiveness at executing these most important parts of poker, I can think of no other way to do so than (gasp) <font color="green"> results based thinking! </font> [img]/images/graemlins/shocked.gif[/img]

Here's my theory:

Since hand-reading is the basis for all expert poker play, I think that carefully measuring my ability to make correct decisions on every street is easily at least as important as calculating win-rate, managing bankroll, etc. The problem is: to use a raw number database set up the way that pokertracker is means that you must play multiple tens of thousands of hands against countless opponents to get even a basic feel for the health of your game (and gives just about no info as to how well you're adjusting to your opponents). Most people that posts PT stats on a limit hold'em forum gets "SAMPLE SIZE" screamed back at them, almost regardless of the question. Even after a million hands, would you have a statistically significant sample of how well you're playing T9o from the small blind vs an aggressive blind stealer? Probably not, but good poker players figure this stuff out well in advance of their first million hands.

I believe that there is more information readily available there than I am currently utilizing. Here's a rough sketch of how I hope to squeeze it from my sessions:

First, I would play a session of a few hours at my regular game. I would do my best to stay on my A game, and try to play solid poker.

After the session, I would replay it hand-by-hand in poker tracker. I would make sure that I could see both my own cards and my opponents known cards. I would then rate my play in the following manner:

I would calculate a Decision Making Score (or DMS) for each street. This score would be kept much like a batting score - with a 1 being awarded for a "hit" and a 0 being issued for a "miss." An at-bat would be recorded for every action on every street. This means that betting, raising, and capping a made boat on the flop vs an over pair would yeild a score of 3. More on the specifics per street. Because some hands play out in such a way that enough information is not available to rate them with any

preflop

This is the hardest street to rate for me, since there is so little specific info available to me at this part of the hand. I'd appreciate any help in improving my rating methods on this or any street. [img]/images/graemlins/grin.gif[/img]

I think I'd generally award a hit if I stick to the chart in SSHE, and I'd allow the following exceptions:
a)I make an off-chart raise in order to isolate a player and the isolation play works.
b)I make an off-chart raise with what I percieve to be a better hand and either the villain folds or he calls with a better hand.
c)I make an off-chart fold to a tight pf raiser and he ends up having better cards than I did.

Any other off-chart moves I make should be issued a 0. This includes ill-advised raises, calling when I should raise, and folding when i should at least call. I'll rate all of my preflop hands.

flop

Here's where the analysis gets interesting:

I'll obviously not rate any hands where I didn't make it to the flop. On those flops that are multiway, but I don't know all of my opponents cards because some did not show down, I'll rate my play as follows:

--I'll use pokerstove to calculate my equity versus the known hands plus this range for all unknowns: all pocket pairs, suited Aces down to A7s, suited Kings down to K9s, and any two broadway cards. This range will be too wide for some players and too narrow for others, but it should be close enough to get a read on how profitable my play was there.

--If I have enough equity I get a hit for a bet or a call, and a miss for a check-through or a fold.

auto hits on this street per action should include:
-betting out and winning the pot right there
-raising with equity
-check-raises with equity count as two hits
-bets or raises that cause one's equity to disproportionately increase due to some opponets folding

auto misses include
-betting or raising without equity and getting called
-check/folding with equity
-folding the best hand HU
-calling without positive equity and insufficient pot odds.

You'll note here that some actions that are clearly -EV become DMS neutral using this system. Something like holding a TT overpair vs QQ and check/calling is -EV, but yeilds a DMS of .5 (1 for the check and 0 for the call) However, taking the same line with a QQ overpair vs villains TT overpair is as +EV as the first scenario was -EV, but the DMS is again neutral (a 0 for the check and a 1 for the call)

I think that this type of rating accurately reflects a WA/WB line, and the hero could even increase his score in either scenario with a read and an appropriate check/fold in case 1 or check/raise in case 2. This seems very poker appropriate to me.

turn

same as flop

river

hits include:
calling with the best hand
betting out and winning the hand
betting and getting called by a beaten hand
checking and having a beaten hand bet (this turns to a miss if you fold)
any subsequent bet that gets called by a beaten hand

misses include:
folding the best hand
betting or raising with a beaten hand
calling a bet with a beaten hand
checking through with the best hand

The river is the easiest street to define with this method, and it's where my whole thought process started when I was dreaming this up. This is a brutal self analysis of one's play, and multiple bad moves can kill yor DMS score in a hurry here.
---------

Once I've recorded my scores, I'd calculate an aggregate DMS score, a preflop DMS, a flop-turn DMS, and a river DMS.

A score of 1 on any street is obviously unattainable. A score of .500 is break-even play. Anything under .500 and you're burning Sklansky bucks at a high rate.

This system rewards the player who can adjust to his opponents. If you auto-raise any suited connectors but like to slow-play AA and AK, my ability to adjust to your style will be directly reflected in this score. If you refuse to pay me off when I have you beat, you're hurting my DMS score and I need to adjust. If you're a loose-passive fish, then I can pump up my score (and my bankroll) by value betting and raising you at every opportunity.

This system is still being fleshed out, and probably has some obvious flaws. The purpose of my methodolgy is to get more immediate, more relevant info from reviewing my hh than poker tracker can give, and to do it in a significantly smaller sample size.

Please flame on, point out my errors, tell me im nutz, and help me perfect this thing.

thx!
Reply With Quote