| User | Post |
|
11:42 am
October 22, 2009
|
sbaker
| | |
| Guest
| | |
|
|
First, I want to say the book is very well written from a pure entertainment perspective. The ideas presented sound powerful. They encompass many different scenarios you are likely to encounter playing poker. The idea of using various stats to gain reads on players sounds very desirable and reasonable.
The authors, however, seem to have not accounted for freedom loss. One of the authors played some 300,000 hands and claims a positive winrate. Three hundred thousand hands seems like a lot until you consider all of the NUMEROUS variable parameters inherent in such a system. Such parameters include: number of players, position, your hole cards, your opponents' range of cards, how often and how much each opponent raises, calls, how often they fold, etc.
Each of those parameters is also free to vary depending on the outcome of the flop, turn, and river. Not counting your own hole cards, there are 19,600 three card combinations (flop), 230,300 four card combinations (turn), and 2,118,760 five card combinations (river). When you factor in all of the parameters that the authors consider when making a decision in relation to the combinatorial explosion of community card possibilities, the methods outlined are SEVERELY constrained by freedom loss.
In short, no confidence can be placed in the statistical validity of their methods due to the HUGE loss in degrees of freedom.
|
|
|
12:40 pm
October 22, 2009
|
mullethaiku
| | |
| Member | posts 96 | |
|
|
sbaker said:
First, I want to say the book is very well written from a pure entertainment perspective. The ideas presented sound powerful. They encompass many different scenarios you are likely to encounter playing poker. The idea of using various stats to gain reads on players sounds very desirable and reasonable.
The authors, however, seem to have not accounted for freedom loss. One of the authors played some 300,000 hands and claims a positive winrate. Three hundred thousand hands seems like a lot until you consider all of the NUMEROUS variable parameters inherent in such a system. Such parameters include: number of players, position, your hole cards, your opponents' range of cards, how often and how much each opponent raises, calls, how often they fold, etc.
Each of those parameters is also free to vary depending on the outcome of the flop, turn, and river. Not counting your own hole cards, there are 19,600 three card combinations (flop), 230,300 four card combinations (turn), and 2,118,760 five card combinations (river). When you factor in all of the parameters that the authors consider when making a decision in relation to the combinatorial explosion of community card possibilities, the methods outlined are SEVERELY constrained by freedom loss.
In short, no confidence can be placed in the statistical validity of their methods due to the HUGE loss in degrees of freedom.
Just curious, in your opinion, how many hands does it take to get an "accurate" measure of success? 1 million hands? 2 million? 3 milllion? 5 million? If all poker authors did this, the authors would have no time to write the book! Or by the time they finished all of those hands, broke down the results, and then wrote the book, it could be 3-4 years later, and the knowledge would be outdated.
Did the book help your game? If so, who cares about how many hands the authors played?
All of the trolls burn this next statement into your brain…
The best poker players do not automatically equal the best poker teachers/authors
This fact can apply to ANY field. In fact many times, the absolute best in a certain field make for a terrible teacher because their ego gets in the way.(see Micheal Jordan and Wayne Gretzky as Coaches and GM's) Not to mention the best players can also make TERRIBLE writers. Research this in any field..the top teachers/authors are rarely the absolute best on the subject.
Now given this fact..who cares about the author's exact stats/# of hands? (which you are never going to find out anyway, and even if you did, you would not believe it)
Ed Miller did not receive widespread acclaim on SSNLHE, NL Theory and Practice, SSHE, and PNL 1 because he was considered the best NL cash game player in the world. He got it because he considered one of the best poker AUTHORS around. In my opinion, his books are clear and concise, full of good info, and he makes some complex concepts easy to understand. That makes him better than probably 90% of poker authors right now (I'm sure he is a good player too, I'm just making a point)
It seems many of you guys feel that the best poker books must come from only the best players with the best stats. If so, have fun reading Phil Hellmuths library on tournament poker  Before you laugh at me, check Phil's stats, I mean he has 11 WSOP bracelets right? and about 10 million in tourney cashes? He must make great poker books!!! WRONG
…and yes I know some highly successful online pros now have online e-books and teaching videos that are also very good…but have fun spending $700-1000 +. Their videos are not for made for the regular small stakes player's budget. Unless you are a big winner, and if are already a big winner, then why do you need help from this book? 
|
|
|
8:56 am
October 23, 2009
|
Matt Flynn
| | |
| Admin
| posts 115 | |
|
|
Hi sbaker,
We're the only book authors out there (at least as far as I know) who took a stand, in writing, on REAL bankroll requirements. The fact that we recommended having 100 buyins and gave some stats to back it up should demonstrate that we are attuned to stats issues.
As for 300,000 hands being an inadequate sample size to prove a poker method, I agree. This does not mean the concepts presented aren't valid.
Matt
|
|
|
10:35 am
November 11, 2009
|
eli
| | |
| Guest
| | |
|
|
Matt Flynn said:
Hi sbaker,
We're the only book authors out there (at least as far as I know) who took a stand, in writing, on REAL bankroll requirements. The fact that we recommended having 100 buyins and gave some stats to back it up should demonstrate that we are attuned to stats issues.
As for 300,000 hands being an inadequate sample size to prove a poker method, I agree. This does not mean the concepts presented aren't valid.
Matt
Matt,
Just curious how you decided to recommend 100 buyins? Are you saying using the methods in the book will give someone at least a 1% edge? How did you determine this edge?
Thanks!
|
|
|
11:02 am
November 12, 2009
|
Jim
| | |
| Guest
| | |
|
|
sbaker said:
First, I want to say the book is very well written from a pure entertainment perspective. The ideas presented sound powerful. They encompass many different scenarios you are likely to encounter playing poker. The idea of using various stats to gain reads on players sounds very desirable and reasonable.
The authors, however, seem to have not accounted for freedom loss. One of the authors played some 300,000 hands and claims a positive winrate. Three hundred thousand hands seems like a lot until you consider all of the NUMEROUS variable parameters inherent in such a system. Such parameters include: number of players, position, your hole cards, your opponents' range of cards, how often and how much each opponent raises, calls, how often they fold, etc.
Each of those parameters is also free to vary depending on the outcome of the flop, turn, and river. Not counting your own hole cards, there are 19,600 three card combinations (flop), 230,300 four card combinations (turn), and 2,118,760 five card combinations (river). When you factor in all of the parameters that the authors consider when making a decision in relation to the combinatorial explosion of community card possibilities, the methods outlined are SEVERELY constrained by freedom loss.
In short, no confidence can be placed in the statistical validity of their methods due to the HUGE loss in degrees of freedom.
This is interesting. I made some conservative assumptions and ran some numbers. Here's what I come up with:
your position out of the players at the table (n-1 = 5)
169 unique hole cards possibilities (n-1 = 168)
a sinlge opponent's range [we'll say 1-10%, 10-20%, 20%+] (n-1 = 2)
3 possible actions [fold, call, raise] (n-1 = 2)
19,600 flops excluding your hole cards (n-1 = 19,599)
5 * 168 * 2 * 2 * 19,599 = 65,852,640 degrees of freedom consumed.
Therefore, we would need to play at least 65,852,641 hands just to have any degrees of freedom left over. Even if you had the stamina of "Seo Awesum" you would need to play 1 million hands per year for over 65 years before you could consider yourself anything other than lucky.
*Gulp* This is rather discouraging.
|
|
|
3:05 pm
November 22, 2009
|
margot
| | |
| New Member | posts 1 | |
|
|
5 * 168 * 2 * 2 * 19,599 = 65,852,640 degrees of freedom consumed.
Therefore, we would need to play at least 65,852,641 hands just to have any degrees of freedom left over. Even if you had the stamina of "Seo Awesum" you would need to play 1 million hands per year for over 65 years before you could consider yourself anything other than lucky.
No you would not. I'm not sure if you've played much poker but the total number of possibilities is not a good metric to use.
|
|
|
8:41 pm
November 22, 2009
|
Matt Flynn
| | |
| Admin
| posts 115 | |
|
|
- redundancy
- the central limit theorem
- the pokerspace in no-limit is far larger than anyone in this thread has accounted for
|
|
|
1:24 pm
November 24, 2009
|
Jim
| | |
| Guest
| | |
|
|
- the pokerspace in no-limit is far larger than anyone in this thread has accounted for
Okay, then that just makes it all the more likely these methods are weak and will not hold up long enough to account for all the parameters.
|
|
|
6:32 am
November 25, 2009
|
Matt Flynn
| | |
| Admin
| posts 115 | |
|
|
Not true. The error you are making is assuming sampling is not valid in poker.
To give you an analogy, take driving a car. Over your lifetime, the potential number of driving situations you can get into, when you account for all variations like "four cars on the road near me at the following distances, 8ft, 17ft, 23ft, 48ft, one of which is accelerating by me, on slightly wet pavement with minimal oil making a left turn" is effectively infinite if you include enough details. By the argument presented, we can never know if a driver is truly safe.
Sampling, however, is good enough to let us know over a few years (say 300,000 minutes of driving) with high likelihood whether a driver is a winning driver who is much likely to have fewer at-fault accidents than average over a year. The safe driver may not actually win that year – he could have an accident – but he is likely to win.
Poker is the same way. So many situations are similar in the pokerspace that they are essentially redudant, so small samples like 300,000 work well enough. This is measured by variance. A degrees of freedom argument falls apart because it assumes sampling is not valid.
|
|
|
3:10 pm
November 25, 2009
|
Sunny Mehta
| | |
| Admin
| posts 67 | |
|
|
Great explanation, Matt.
To add to what Matt said, note that variance rates seem to converge pretty quickly. While we observe madass swings in winrate over, say, 100k hand samples, we see fairly constant SD's.
This was my point in the other thread where someone dismissed my 28k hand screenshot as insignificant. Say a breakeven player played an infinite number of hands, and we decided to look at a whole bunch of 28k hand samples. The CLT tells us that, regardless of what the actual distribution of winnings looks like, the means of 28k hand samples will be normally distributed. Further, if we assume an SD of 50 PTBB/100 (which is way on the high side for 6-max, but i'm erring on the side of caution here) we'd expect 95 percent of those 28k hand samples to show a winrate between -5.86 PTBB/100 and 5.86 PTBB/100. Note that my actual observed winrate of 8.4 PTBB/100 over 28k hands is way outside that range, and it is even outside the 99 percent range of +/- 7.7 PTBB/100. Therefore, it is very, very unlikely that I was a breakeven player over that stretch and pretty damn likely I was a strong winner. So even though it was a small sample, it happened to be very significant.
|
|
|
2:30 pm
December 11, 2009
|
Amaryllis
| | |
| Guest
| | |
|
|
Hi,
Not a probability guru, but trying to follow.
Shouldn't the 95% certainty be between – 2 SD and + 2 SD? Meaning between -100 PTBB/100 to +100 PTBB/10 instead of +- 5.86? I guess in other words I should ask whether you can show me how I should calculate this.
Thanks
Amaryllis
|
|
|
6:12 pm
December 11, 2009
|
JJS
| | |
| Member | posts 48 | |
|
|
Post edited 2:13 am – December 12, 2009 by JJS
Post edited 2:14 am – December 12, 2009 by JJS
Amaryllis – The SD of 50 PTBB/100 represents the SD for a sample of 100 hands. Sunny is talking about a sample of 28,000 hands. To translate this you do:
50 x square_root(100/28000) = 2.988 PTBB/28k hands
The 95% certainty is actually a factor of 1.96, not 2:
2.988 x 1.96 = 5.857
That's where Sunny's +-5.86 comes from.
|
|
|
12:20 am
December 16, 2009
|
Amaryllis
| | |
| New Member | posts 1 | |
|
|
Login
Search 
Home
Print this Post 
Add a New Topic
Reply to Post
Quote and Reply 
Topic RSS 
