[jgilmartin]

OP, I'm not sure that you understand what many people have been getting at in the thread. Please read this: http://en.wikipedia.org/wiki/Data_dredging

[hutennis]

Usma


Please clarify.
How many games out 3600 you mentioned you actually bet on?

[hutennis]

Oh yeah.

I never thought you lied. I am almost positive you are very confused, though.

[usma1992]

Thank you for the insight... I have to think about it... for a day or so to see if that is exactly what I am doing...

[usma1992]

Hu...I started 3 years ago on college basketball after building my first version of college football program. I figured that I could understand what was important in the game of basketball. I flew to vegas spent a week building a program testing it on every game that was being played. I got pounded 1-9 the end of the week flew home a put the program on the shelf. Than I decided to take it off the shelf and begin building the program.

I have bet approximately 1/10 of the games that the program has told me to bet because as I was fine tuning the relationships... I continued to lose and further develop not only what the bets were but how I approached them... how I weighted them... what $dollar value I would put on each bet etc...

Fast forward... I did this for two years... fast forward Since Dec 11th... I have essentially bet every game. Last night I used the 57% system, went 0-3 and lost $525.

For the college football season, I bet every bowl game and the system hit 60%. I made money on college bowls for the first time in a while and I made money during the month of december on basketball. So my training set has been the previous two years... and my test set is this year starting Dec 11th. I continue to fine tune and plan to fine tune further over the course of the season.

Am I data dredging... as some one previously stated... Not sure... I have found a positive correlation 59% over the course of a subset of 1300 games of the 7000-8000 I have tracked. Is my model legit? No idea, this year and next year... will prove it one way or other I suspect.

So yes I have bet every game this year and many of the games last year. I personally don't believe unless you are willing to lay money on the line it has any legitimacy.

Enjoy

[HeeeHAWWWW]

Am I data dredging... as some one previously stated... Not sure... I have found a positive correlation 59% over the course of a subset of 1300 games of the 7000-8000 I have tracked.

Unfortunately, 57% almost certainly suggests that's the flaw. We're not trying to be unnecessarily negative here, but it's a frequent error in sportsbetting models, and high 50s is indeed what it produces (followed by something that quickly converges to 49-51% when you actually try it out).

[Miz]

Unfortunately, 57% almost certainly suggests that's the flaw. We're not trying to be unnecessarily negative here, but it's a frequent error in sportsbetting models, and high 50s is indeed what it produces (followed by something that quickly converges to 49-51% when you actually try it out).

This is very valuable advice, that was learned by first-hand experience, no doubt. I have seen this also

[usma1992]

I understand... but it yielded over 60% over the course of December and I did try it out. I made a couple of adjustments first week of January to try and make it better. Maybe my adjustments were made prematurely but after tracking 8000 games...it seems like a fair sample size...hitting 59% doesn't seem impossible. At what point does your sample size become self validating... 8000...10000...20000 games. Not sure... just curious on thoughts...

Has anyone truly tracked 8000+ games legitimately like I have... maybe... but I doubt many have really put in the effort.

I guess only time will tell...

[jgilmartin]

Where are you getting the 59% figure? Is that from your actual bets or what your bets WOULD HAVE been if you had been using the same formula that you are today?

[usma1992]

Picks for Tonight...

19:00	Depaul	Connecticut	144	N
19:00	Northeastern	Drexel	130.5	N
19:00	Clemson	Duke	132	N
19:00	Alabama	Missouri	132.5	N
19:00	Baylor	Texas Tech	150.5	N
19:00	Niagara	Brown	138	N
19:00	Appalachian State	Wofford	130.5	O
20:00	Drake	Creighton	147.5	U
20:00	Illinois State	Missouri State	127	N
21:00	Pittsburgh	Georgetown	115.5	O
21:00	Ohio State	Purdue	135	N

[HeeeHAWWWW]

I understand... but it yielded over 60% over the course of December ....

Problem is sample size - what's the z-score on that?

[Miz]

I give up

[matthew919]

It's like watching someone try to build a skyscraper out of soggy cardboard.

[hutennis]

Hu...I started 3 years ago on college basketball after building my first version of college football program. I figured that I could understand what was important in the game of basketball. I flew to vegas spent a week building a program testing it on every game that was being played. I got pounded 1-9 the end of the week flew home a put the program on the shelf. Than I decided to take it off the shelf and begin building the program.

I have bet approximately 1/10 of the games that the program has told me to bet because as I was fine tuning the relationships... I continued to lose and further develop not only what the bets were but how I approached them... how I weighted them... what $dollar value I would put on each bet etc...

Fast forward... I did this for two years... fast forward Since Dec 11th... I have essentially bet every game. Last night I used the 57% system, went 0-3 and lost $525.

For the college football season, I bet every bowl game and the system hit 60%. I made money on college bowls for the first time in a while and I made money during the month of december on basketball. So my training set has been the previous two years... and my test set is this year starting Dec 11th. I continue to fine tune and plan to fine tune further over the course of the season.

Am I data dredging... as some one previously stated... Not sure... I have found a positive correlation 59% over the course of a subset of 1300 games of the 7000-8000 I have tracked. Is my model legit? No idea, this year and next year... will prove it one way or other I suspect.

So yes I have bet every game this year and many of the games last year. I personally don't believe unless you are willing to lay money on the line it has any legitimacy.

Enjoy

OK, I see.

It is becoming painfully obvious that you are getting into every trap your brain is setting up for you.
Those traps are cognitive and logical fallacies and illusions.

There is nothing wrong with your brain, of course.
Every brain of every human being is evolutionary programmed to do the the same thing - to fool us.
As paradoxically as it sounds, it is basically a survivor mechanism. It's been very useful for quarter of a million years. It is probably the greatest single reason we were able to survive as a spices.

But while it was so beneficial for our ancestors in a African savannas, it brings nothing but troubles to a modern man.
This is not my opinion. This is a scientific fact. Although behavioral psychology is fairly young, none the less, this fact is well established and empirically proven.

Instead of getting any deeper into this madness you have committed yourself to, I would strongly suggest to put your "modelling" efforts aside and start to educate yourself on these psychological issues.

This is a great starting point.

You Are Not So Smart: Why You Have Too Many Friends on Facebook, Why Your Memory Is Mostly Fiction,
and 46 Other Ways You're Deluding Yourself.
An entertaining illumination of the stupid beliefs that make us feel wise by David McRaney

And, of course, the great "Thinking Fast and Slow" by grand daddy of behavioral psychology himself - Daniel Kahneman.

Read it, then read it some more, then think about it, then read it again.
Then look back on what you are doing now and you will be horrified.

That's the best advise I (or anyone else for that matter) can give you.

[brettd]

Hu...I started 3 years ago on college basketball after building my first version of college football program. I figured that I could understand what was important in the game of basketball. I flew to vegas spent a week building a program testing it on every game that was being played. I got pounded 1-9 the end of the week flew home a put the program on the shelf. Than I decided to take it off the shelf and begin building the program.

I have bet approximately 1/10 of the games that the program has told me to bet because as I was fine tuning the relationships... I continued to lose and further develop not only what the bets were but how I approached them... how I weighted them... what $dollar value I would put on each bet etc...

Fast forward... I did this for two years... fast forward Since Dec 11th... I have essentially bet every game. Last night I used the 57% system, went 0-3 and lost $525.

For the college football season, I bet every bowl game and the system hit 60%. I made money on college bowls for the first time in a while and I made money during the month of december on basketball. So my training set has been the previous two years... and my test set is this year starting Dec 11th. I continue to fine tune and plan to fine tune further over the course of the season.

Am I data dredging... as some one previously stated... Not sure... I have found a positive correlation 59% over the course of a subset of 1300 games of the 7000-8000 I have tracked. Is my model legit? No idea, this year and next year... will prove it one way or other I suspect.

So yes I have bet every game this year and many of the games last year. I personally don't believe unless you are willing to lay money on the line it has any legitimacy.

Enjoy

The underlined screams of back fitting. Trust me man, been there and done that.

Even in validation testing you can convince yourself that you have +EV subsets that you legitimately believe you have edge on; but 9 in 10 times you've just singled out a spurious subset in an overall -EV model.

Answer yourself this:

Whether you have absolute point value estimates or logistic type ATS win % estimations, an increase in the deviation of your plays from the spread should increase your actual win % on plays in your validation set. If this relationship between ATS deviation and actual win % does not exist, your model is highly likely to not be any good.

Also as a general rule of thumb, if you have a model that cannot grade plays in terms of their +EV magnitude; either through your own estimate of ATS win % or a point differential from the line, then your model is also likely to not be much good at all.

[usma1992]

I understand what the board is saying. Paraphrase(though I have identified a subset that seems to correlate to 59%, the likelihood is that is that is just a coincidence and that I really have determined any relationships worth anything). I guess my question is ... at what point do the shear numbers of games tracked

Does it scream of back fitting... absolutely... as you collect data... back fitting has to occur to some degree as u fine tune...

In my defense...
1. How many people here have actually tracked 8000 games with live statistics at the time of the game? My guess... very few, if any. So those claiming back fitting, not have correlations etc... are probably talking about much smaller numbers than I am. Brett, I am curious, how many games did you test? Not trying to get into a contest... I am just curious when people talk about have models. How many games have they actually looked at and analyzed.

2. It is working and has worked over the course of December. How many games? My guess is 500 games. Is it a coincidence that I can show the over the previous 6000 games the relationships work and over the course of a new season... it is still working... maybe...

3. I noticed the lines change in my favor almost every time after I place my bet. There have only been a couple of games that have gone against me in the last couple of weeks and it turned out I lost those games. So the sharps new something I did not.

Starting yesterday.... I will continue to post the picks of the 59% system to see how if fairs. Right now... I am 1-2...Enjoy..

As for the college bowl season, it was great. If not for a couple of crazy plays... I would have been very close to 70%. I believe that my college football system is ready to go with less than a days work. I can provide picks starting week 5 of the season next year. I forgo the first 4 weeks in order to collect data.

Dave

[usma1992]

19:00	Dayton	Virginia Commonwealth	140.5	N
19:00	Rhode Island	Richmond	128	N
19:00	St. Bonaventure	George Washington	135.5	N
19:00	Duquesne	Fordham	148.5	N
19:00	Iowa State	Kansas	149	U
19:00	Syracuse	Providence	136	N
19:00	Rutgers	St. John's	138.5	N
19:00	Boston College	Virginia Tech	144	O
19:00	Nebraska	Michigan	126	N
19:00	Louisville	Seton Hall	138.5	N
19:00	William & Mary	Towson	126	N
19:00	Hofstra	Delaware	135	N
19:00	Butler	Saint Joseph's	125	O
19:00	UAB	UCF	144	N
19:00	Tulsa	Marshall	131	O
19:00	Cleveland State	Wright State	121.5	N
19:00	Western Michigan	Akron	140.5	N
19:00	Northern Illinois	Miami (OH)	125	O
19:00	Ball State	Eastern Michigan	114	O
19:00	Buffalo	Ohio	132	N
19:00	Toledo	Kent State	138.5	N
19:00	North Carolina-Wilmington	James Madison	134	N
19:00	Bradley	Indiana State	125	O
19:00	Georgia	Florida	119.5	N
19:00	Bowling Green	Central Michigan	130.5	N
19:30	La Salle	Charlotte	137	N
20:00	Florida State	Maryland	145	N
20:00	Georgia Tech	North Carolina State	134	N
20:00	TCU	Oklahoma State	117	N
20:00	Southern Miss	Rice	128.5	N
20:00	East Carolina	Memphis	151	N
20:00	Villanova	USF	123	N
20:00	South Carolina	Mississippi State	133	N
20:00	Evansville	Northern Iowa	131	N
20:00	Southern Illinois	Wichita State	128	N
20:00	Valparaiso	Illinois-Chicago	122	N
20:00	Loyola (IL)	Green Bay	123	N
20:00	Mississippi	Tennessee	134.5	N
20:00	UTEP	Tulane	119	N
20:00	Southern Methodist	Houston	134.5	N
20:00	Eastern Kentucky	Murray State	142	N
20:00	Morehead State	Austin Peay	145.5	N
21:00	Nevada	Air Force	138.5	N
21:00	LSU	Auburn	137	N
21:00	Arkansas	Texas A&M	140	N
21:00	West Virginia	Texas	126.5	N
21:00	Minnesota	Illinois	141.5	N
21:00	Virginia	Wake Forest	120	N
22:00	Boise State	Wyoming	125	O
22:00	Cal State Northridge	Long Beach State	155	O
22:00	UNLV	New Mexico	138.5	N
22:00	Washington State	Stanford	122	N
23:00	Washington	California	143	U
23:00	Hawaii	UC Irvine	139	N
23:30	San Diego State	Fresno State	122	N

[roasthawg]

How many people here have actually tracked 8000 games with live statistics at the time of the game?

Interested in what this line means? What are you "tracking at the time of the game"?

[brettd]

Interested in what this line means? What are you "tracking at the time of the game"?

Yeah I was gonna ask the same question.

Also as a general rule of thumb, if you have a model that cannot grade plays in terms of their +EV magnitude; either through your own estimate of ATS win % or a point differential from the line, then your model is also likely to not be much good at all.

I'm also concerned about the fact that your outputs cannot seem to grade plays in terms of EV.

3. I noticed the lines change in my favor almost every time after I place my bet. There have only been a couple of games that have gone against me in the last couple of weeks and it turned out I lost those games. So the sharps new something I did not.

If you can BTCL >80% of the time over the next 300 plays by an average of over a point, you at least have a break even model.

[brettd]

Also, where you getting those lines from?

My sports insights account only shows nine NCABB games having released totals lines.

[usma1992]

betonline.ag... only site that has basketball totals... from what I have found....

[brettd]

Ah damn I didn't know they released before 5dimes. 5dimes popped up with a multitude of totals lines only a couple of hours ago.

Have added BetOnline to my watch list.

[usma1992]

Lines changes... of the games I bet... Bradley moved too much to get that game in... became a neutral bet. Most others moved the way I bet them except Northern Illinois... that moved .5 point against me.

Kansas and Washington lines stayed the same. We shall see.

Good Luck
Dave

[sbrhedge]

walk-forward optimization, out-of-data sampling, re-sampling, etc. are all things you might want to do to verify the volatility of your numbers. you tested apparently over several years, but you might consider these methods within a single season as well (for many reasons).

provided your numbers are good, the performance you report has a plus or minus 1-2% of vol. it's quite a lot considering that at a high confidence level, your performance could drop all the way down to 51% .. at -110, you have a losing tape.

your sharpe and pain-to-gain ratios are also going to determine how good your system really is.

[Miz]

betonline.ag... only site that has basketball totals... from what I have found....

5Dimes does also. Bookmaker has a more limited set, but the lines are softer. I highly recommend you stay away from betonline. See "the industry" forum.

[brettd]

pain-to-gain ratios

Can someone elaborate what this means please, have never come across the term.

[brettd]

5Dimes does also. Bookmaker has a more limited set, but the lines are softer. I highly recommend you stay away from betonline. See "the industry" forum.

They're good for steam off their action.

[339955]

id say it is very possible to make a model which is profitable, and then over optimizie and change the results from slightly profitable to silly profitable.

[Spektre]

Ok, never mind.

Here's what we got.

You are saying that with avg probability of winning the bet = 50% you won
57% or 1131 out of 2300 games and
59% or 767 out of 1300 games.

The probability of getting results like this by luck is nice, big, fat 0.
So, basically you are sport betting genius, the best ever.

Yet, you don't know what an implied probability is, you don't understand importance of posting odds,you've been removed far from statistics
and you have trouble figuring out on your own witch system will yield you more dough in a long run.

And yet, with all this ignorance on a elementary subjects, you did manage to develop
indisputably the best sport betting models of all time.
Models that could be sold/bought for countless millions tomorrow.

I don't know. Does not look good.

I'm more curious about the motivation of this post than the actual content.

A person interested who wants to ask a question about sports betting comes onto a forum dedicated to ... sports betting and ask a question on the forums. . The user's join date is only a few months past and thus he is a new user.

I have read the thread and he certainly seems confused on training data vs. testing data, and he is unskilled in the vernacular of statistics. Nevertheless, he correctly locates the subforum dedicated to algorithm development and asks his question.

To this you decide to:

(1) Call him a liar.
(2) Try to dissuade others from helping him
(3) Call him a delusional dreamer or an attention whore
(4) Decide that by somehow posting some picks here his model would gain credibility (This would be a long thread if we wanted to gain statistical significance)
(5) Advise him to give up a hobby that might bring him enjoyment to read something such that he can look back on this with horror.

Why?

As per my post count, I certainly do not take time to post often, but have contributed to some when I thought I could offer benefit. When I did not, I did not. When I found the post to be not to my liking, simply moved on.

In terms of knowledge benefit, what would you expect a poster's EV, who needs help with a subject someone else is knowledgable on,to be when that someone else responds.

What would it be if that poster did not need help with a subject someone else was knowledgable on.

Would it not thus make sense to help posters who need it (as is demonstrated here) as opposed to offer help where it is not needed?

Once again, more interested in the motivation than the content.

[hutennis]

Once again, more interested in the motivation than the content.

Help. That's the only motivation I had.

And please, post a quote when I called him a liar.

[parlay100]

data mining!

[Spektre]

Help. That's the only motivation I had.

And please, post a quote when I called him a liar.

Look, I don't know, maybe it is just me, but...

The best sport betting model of all time from 4 year old laptop operated by a guy who "has been removed from statistics"
sounds not just too good to be true. It sounds intelligence insulting.

liar (n) a person who tells lies

lie(n) an untrue or inaccurate statement that may or may not be believed true by the speaker

cite to Mirriam Webster's online dictionary

----

You intended to help him with this hobby by suggesting he abandon it?

I don't intend to belabor this further. It simple seems like a series of counter-productive posts.

[hutennis]

Look, I don't know, maybe it is just me, but...

The best sport betting model of all time from 4 year old laptop operated by a guy who "has been removed from statistics"
sounds not just too good to be true. It sounds intelligence insulting.

liar (n) a person who tells lies

lie(n) an untrue or inaccurate statement that may or may not be believed true by the speaker

cite to Mirriam Webster's online dictionary

----

You intended to help him with this hobby by suggesting he abandon it?

I don't intend to belabor this further. It simple seems like a series of counter-productive posts.

Well, I guess instead of focusing on "it sounds too good to be true" (pure semantics) why not focus and discuss why " It sounds intelligence insulting."

Maybe by the end of this discussion you would understand my motives much better.
You would also be able to see the difference between innocent hobby and potentially dangerous and destructive obsession based on self delusion, limited knowledge on a subject, bunch of erroneous believes and complete loss of common sense.

[hutennis]

It simple seems like a series of counter-productive posts.

What do you see to be a productive input?

I can only see my posts to be counterproductive if looked at from books POV.
Would be kinda hard to get money out of him if he, as you say, "abandons his hobby"

[Spektre]

I would not tell someone picking up a violin for the first time, who seemed quite unfit to play decently, to put away his instrument and go read a book on why most people never become a virtuoso.

I might instruct them on how to read music, and the amount of time and effort that would need to be put in to accomplish this, but to simply assume, "You aren't where you need to be so give up now", seems counter-productive to me.