Not a system or strategy exactly but rather more help requested in developing one

[muppet77]

Re: Not a system or strategy exactly but rather more help requested in developing one i have used a random 300 games from the premiership. i have access to a website that gives this of data. i have used home teams all PPG for last 10 games - same for away and also home team's av goals for last 10 the results are attached. the cutoffs can be changed. a small loss. what exactly does the resultant value mean? (to be honest it sounds a bit strange - arctan of an arctan of an arctan. secondly the two variables are correlated and so such a simple model must have strng weaknesses) just a thought.

[Kanga]

Re: Not a system or strategy exactly but rather more help requested in developing one Muppet, are these figures based on the actual table points at the time of the match being played ? - it seems to make a big difference based on previous comments.

[muppet77]

Re: Not a system or strategy exactly but rather more help requested in developing one so no weighting? a game played as match 38 in the PL has the same value as game 1 played 10 months earler?

[Smokin Joe]

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one

True, think you're right mate, no point in reinventing the wheel, but would also like to compare it across divisions/countries/my own ratings model. Next job is to get the odds appended into my db. Anyone done it already or got a function to do it!? The footydata files are all over the place which makes automation a nightmare and why I've put it on the back burner...

If I can confirm that my formula is correct, I can automate for all of the files that football-data have.

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one

so no weighting? a game played as match 38 in the PL has the same value as game 1 played 10 months earler?

Yes, that seems to be what the formula is stating.

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one Using a cut off of 72% and -28%, for the EPL this season: Homes 64 bets, lose £7.79 Aways 32 bets, win £2.38 In fact, I tested for all of the current seasons with the files from the main leagues from football-data using those cut offs, and the loss was just over 4%.

[happygooner]

Re: Not a system or strategy exactly but rather more help requested in developing one Well I've really opened up a can of worms here so that's nice. The thread has been very helpful also in making me see that complicated or not I need to combine many different things and not solely hope to rely on just one or 2.

[Rushian]

Re: Not a system or strategy exactly but rather more help requested in developing one Interesting discussion going on, just a couple of points if I may:

... Where . . PPGDIFF = [Home team’s points per game] minus [Away teams points per game] HGPG = Home team’s scored per game (Cos(-5.363495)^3) +PPGDIFF+Atn(Atn(Sin(Atn(Atn(Atn(Log((Atn(HGPG)^(1.0/3.0))))))))) +(Cos(-5.363495)^3) ...

(a) Any reason for not writing: (Cos(-5.363495)^3) + ... + (Cos(-5.363495)^3) = simply as 2 *(Cos(-5.363495)^3) + ... What I mean is, they are exatly the same, right? Or is there some sort of typo? (b) Since these ratings are used with cutoff points, wouldn't removing the cosine terms alltogether make the ratings much more simpler? Their effect could be included in the cuttoffs. and © can the complicated atan function not be simplified? I've had a look at it's shape and it looks very similar to the logarithmic function, just that it flats out quicker... There must surely be something simpler than that! :eek

[a1ehouse]

Re: Not a system or strategy exactly but rather more help requested in developing one Ha! Certainly have happygooner, sorry to have hijacked, all relevant tho so thanks for raising!! Muppet77, I have run your selections and calculate a 27.1 point profit: 88 home bets, return +16.03 44 away bets, return +11.64 attached including my GeneX. Smokin Joe - think you suggested before you had a function to get the odds, let me know if you willing to share etc. drives me mad with all the different columns in all the data files at least the first few columns are consistent! I think the formula was passed through the GeneX program at the end of the 0506 season, so would be all games at that time. For the prediction model to work in season, I defo think it'd be points per game/goals per game at the time.

[muppet77]

Re: Not a system or strategy exactly but rather more help requested in developing one Ale. How is yours different to mine? (cannot access the excel file as i am on mobile phone web).

[Kanga]

Re: Not a system or strategy exactly but rather more help requested in developing one

Interesting discussion going on, just a couple of points if I may: (a) Any reason for not writing: (Cos(-5.363495)^3) + ... + (Cos(-5.363495)^3) = simply as 2 *(Cos(-5.363495)^3) + ... What I mean is, they are exatly the same, right? Or is there some sort of typo? (b) Since these ratings are used with cutoff points, wouldn't removing the cosine terms alltogether make the ratings much more simpler? Their effect could be included in the cuttoffs. and © can the complicated atan function not be simplified? I've had a look at it's shape and it looks very similar to the logarithmic function, just that it flats out quicker... There must surely be something simpler than that! :eek

Well spotted Rushian ... I hadn't noticed the symmetry before. I dont think it's a typo the webpage is listed above. I'm not qualified to comment on cutt-offs or a simplification, but like you I'm sure there's must be a simpler way to express it.

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one

Ale. How is yours different to mine? (cannot access the excel file as i am on mobile phone web).

A1e, I second that. I get a loss on the EPL for the current season, where are your large profits coming from?

[Rushian]

Re: Not a system or strategy exactly but rather more help requested in developing one Continuing from my previous post, f(x) = ATAN(ATAN(SIN(ATAN(ATAN(ATAN(LOG(ATAN(x)^(1/3)))))))) is extremely close to g(x) = LOG(ATAN(x)^(1/3)) for values of x > 0.5 so for any team that scores on average at least a goal every other game. After a few games, the average will be more than that so I guess this can be simplified. Just my 2p worth.

[a1ehouse]

Re: Not a system or strategy exactly but rather more help requested in developing one

A1e' date=' I second that. I get a loss on the EPL for the current season, where are your large profits coming from?[/quote'] Not really sure chaps, I have built 3 functions: PointsPerGame GoalsPerGame GeneXPro I pass the team, season and match date through the first two making sure the HomeTeam or AwayTeam = the team and this fires a query on the db to bring all home and away games for that team. the GeneXPro function is just the formula, I made no changes. If I run the functions based on todays date I get the correct games played and the correct points tally. Think I differ from Muppets cos he's looking at average of recent form (10 days) whereas I look at season to date e.g. game 15

[servecold]

Re: Not a system or strategy exactly but rather more help requested in developing one not sure if this is all ready being used but for the ref couldnt you add thier (card totals up /number games) maybe award point for yellow,red cards somehow maybe 3pnts yellow 6pnts for red then turn this into a % ??? how these %s would be used i,ll leave upto the mathmations

[Kanga]

Re: Not a system or strategy exactly but rather more help requested in developing one Seems to me the differences we have here is from when the input data is taken from, can you guys confirm the way your calculating this. Must admit my initial reading of the formula was to take the info from the start of the current season, but how would this be done with no games played initially ? and then you have the problem of how do you calculate for promotion or demotion :wall so that can't be right. A1's data base seems to be giving good results (even using muppetts random results gave a 15% ROI) so maybe that's the correct way of doing it ???

[muppet77]

Re: Not a system or strategy exactly but rather more help requested in developing one What precisely are the two variables? Home teams goals at home and also away per game? The second is home teams total points per game minus away teams total points per game? Are these correct? I assume that they must be over a certain amount of games rather than roll from one season straight into the next.

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one

What precisely are the two variables? Home teams goals at home and also away per game? The second is home teams total points per game minus away teams total points per game? Are these correct? I assume that they must be over a certain amount of games rather than roll from one season straight into the next.

I assumed that all the variables are for the season to date. This was reinforced by the blurb where the writer states that no predictions were made until 8 games were played by both teams.

[muppet77]

Re: Not a system or strategy exactly but rather more help requested in developing one Good shout. And the home / away issue?

[muppet77]

Re: Not a system or strategy exactly but rather more help requested in developing one using the suggested cutoffs and home team's PPG (H+A) - away team's PPG (H+A) and home team's GPG (both H+A) i get a yield of -3% for this year's premiership. using maximum odds on betbrain.

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one

Good shout. And the home / away issue?

By that do you mean "average goals socred by home team"? If so then I take it to mean the average goals scored by the home team in ALL their games season to date. That return of -3% is something similar to what I got.

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one Getting 80% of the corrections correct is no good if the average odds of your selections are 1.2 or less (which may not be beyond the bounds of possibility if the author used the highest 200 home predictions that the ratings generated). This shows the folly of writers just quoting the numbers of winners they got, if they don't look at the odds of their selections. Paul Steele was guilty of this as well in his book.

[Kanga]

Re: Not a system or strategy exactly but rather more help requested in developing one Guys the page this was taken from is page II - here's page I http://www.uk-neural.net/football.html It looks like he took data from 03/04 and 04/05 to predict 05/06 season. So maybe A1's correct and explains the profit shown ?

[a1ehouse]

Re: Not a system or strategy exactly but rather more help requested in developing one

Guys the page this was taken from is page II - here's page I http://www.uk-neural.net/football.html It looks like he took data from 03/04 and 04/05 to predict 05/06 season. So maybe A1's correct and explains the profit shown ?

Yes, I read the full monty, it's actually quite an interesting article. All he is doing is training a piece of software over 03/05 and testing the reliability using 05/06 data. This is why it is suggested that only 3/20 inputs were used by GeneX. The package has produced the formula and inputs to use, I have done the same. I am actually looking at the total points or goals divided by the number of games played in that season at the match date: e.g. Match day 15, HomeTeam example: 14 games completed 28 points, 16 goals PPG: 28/14 = 2 HGPG: 16/14 = 1.14

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one

I am actually looking at the total points or goals divided by the number of games played in that season at the match date: e.g. Match day 15, HomeTeam example: 14 games completed 28 points, 16 goals PPG: 28/14 = 2 HGPG: 16/14 = 1.14

A1ehouse, that is the basis that I calculated my ratings, and I get quote a poor set of results using that logic. Will be interesting to see if you have any improvements.

[a1ehouse]

Re: Not a system or strategy exactly but rather more help requested in developing one Re ran again.

[Smokin Joe]

Re: Not a system or strategy exactly but rather more help requested in developing one A1ehouse, your numbers are still wrong I am afraid. Take Liverpool v Aston Villa on 22nd March. Liverpool scored 22 home goals in 14 games and 27 away goals in 15 games prior to this game - 49 goals in 29 matches = an average of 1.69, not the 1.80 that you have. Likewise for the league points: At home they had 8 wins and 6 draws = 30 points Away they had 9 wins and 4 draws = 31 points Total 61 points in 29 games = 2.10 PPG, not the 2.13 that you quote. Villa also has an incorrect PPG figure. It should be 1.79 not 1.73 (IMO). You must still be doing something wrong in your logic.

[Kanga]

Re: Not a system or strategy exactly but rather more help requested in developing one SJ - when are you counting from ? A1 is counting back a couple of seasons as in page I ( as i understand it )

[muppet77]

Re: Not a system or strategy exactly but rather more help requested in developing one agreed- you have PPG values for the first games played? attached is my sheet - no predictions until game 8.