That is the war cry of Wyld and the challenge he has set. I'm possibly "getting above my station" here but I'm happy to adjudicate this competition.

Rules
- The game will start beginning of GW8, after the international break, and run for 10 GWs finishing GW17 (inclusive)
- 25 players will be published in the international break. These will be the same 25 players that will be used throughout the whole 10 GW competition
- By Friday 11.59pm, UK time, each participant must PM me the score they believe the 25 players will achieve that GW. (If there is a midweek GW then players must PM their predictions at least 1 hour before the first kick-off of that GW)
- By Saturday 11.30am I will publish these scores on this forum so everyone can see the predicted scores of each individual player. (For midweek GWs, this will happen ASAP)
- At the end of competition we will crown the person who has the closest % overall.

Prize
- Bragging rights and ultimate respect

I know crispybits is in and I presume Wyld will keep his pledge above, but please sign up below if you are interested in seeing if your formula is the best on the FISO forum.

I'll give it a go if there's enough people doing it... Warning: No formula, just gut! I fancy my chances over any formula , just hope 10 GWs is enough.

If formulas were horses, the statisticians would ride.
We would be having MBA / MA ( Statistics ) from Oxford , Cambridge, MIT , Wharton , Harvard - etc - currently in the Manegerial positions.
But football - is football - and though statistics really help back you up on your decisions - they alone cannot form a basis for decision making.

So whatever the case is ( Human Brain VS The State of England ) - Am in .

What was Wyld's formula anyway ?

I'll give it a try, but as I've stated before, I don't actually put any real faith in my "predictions". If I thought they were accurate, I wouldn't publish them, I'd try to make money off them.

A couple suggestions though: I'd prefer to see a limited the number of defenders and consider leaving keepers out of the game altogether. These are SO heavily dependent on team play that it makes predictions very difficult. Also, consider judging the results in terms of both actual & relative scores. I'm as happy to know that player A will score more than player B as I am that player A will get 8 points while player B gets 6. You might also consider adding a few more to the list (say five) who could be brought in if one of the "starters" gets injured.

I would also like to see a small amount of players included who did not play EPL last year but DID play at the top level in whatever league they were in (VdV for example). We might also want to make sure we have a couple players who have changed EPL teams in the last year or two (you know, like Man City's starting 11).

Netflix offered (and I believe paid) a million dollars to whomever developed an algorithm that could significantly improve upon their baseline method of predicting a user's rating of a movie title. I'm assuming something similar will be offered here?

Edit: I'd like to see the guy who developed the FantasyAdvice website get in on this as well.

Count me in. Won't be using a formula but I'm generally ok at predicting points... unfortunately never for my own team's players.

Who is going to pick the players? I've had a few ideas if it will help at all... It might be useful to separate this into 3 groups of players with 8 defenders, 8 midfielders and 8 forwards to see which position is easier to predict too. Goalkeepers are a pretty pointless exercise, I think. And also spreading the chosen players across teams/abilities so 2 guaranteed starters from current top 5 teams, 2 from 6-10, 2 from 11-15 and 2 from 16-20 but only 1 max from each team in each position.

Hey

Im pretty New to the forums but i like the idea and would love to be involved.

And Of course we all need someone to laugh at for getting the lowest percentage!

Thanks

As someone who has contributed to that thread, I will give it a go. You will need to decide what to do about injured players, whether to replace or remove as Wylds model will still produce a result, and how you will actually measure us, eg standard deviation or sum of squared differences or whatever other measurement there may be

I am defo in this, how will the scoring be done?

Would giving everyody a point for every point they are away from the correct player score, then tallying the results up in reverse order so the lowest score wins, how bout that?

Bring it on!

Thanks to everyone for the suggestions so far - I'll run down a few criteria later with regards the types of players etc - think it's going to be difficult to please everyone with the group of players but do agree that it will include a lot more attackers/midfielders rather than defenders and probably no goalkeepers. I do like cinci's idea about different types of players eg new to EPL, moved teams within EPL etc but I do think this will result in people claiming their formula/guesswork only works on certain types of players.

As for injured players - I might increase the number of players to 30 - 35, so should we have 5-10 players injured, we still have a good spread. Otherwise Wyld can add to his formula 'x 1' if player not injured or 'x 0' if player is

How to score the contest is something I want more clarity on. Do we want it to be that we see who has the closest points total over the 10GWs, or are we more interested in weekly success? For instance, let's just take 5 GWs - If Bale actually scored

4, 14, 2, 7, 11 and therefore got = 38 points but someone actually predicted
21, 2, 2, 12, 1 - even though they got 38 points as well, only GW 3 was a perfect prediction, whilst the other GWs were way off. How do you want to mark the formula?

I ask again. Whats the formula ?

I'd go along with that scoring method, although it might be a lot of work with everyone involved. Are you sure you know how much you've taken on here, stuboy?

It's also going to take a commitment for all 10 gameweeks too so if anyone isn't able to submit their predictions for a week, what contingency do we have? Assigning average points from other players for that week?

I don't think it even needs to be as complicated as this. If the player doesn't play, just assign no score for that player. Replace the player (if injured) with someone from the same team in a similar position if necessary. There isn't a need for a backup list really.

I do think this might end my relationship with my girlfriend but if it means I end up knowing whose formula/guess work I should follow so I win my mini-leagues, surely it's going to be worthwhile!



There is no contingency - anyone is capable of looking ahead of time so if they think they'll be away one week, they PM 2 weeks worth to me. We can have a * next to that week's score showing they predicted early but if people have formulas, then unless the previous GW score is part of it, they should be able to manage.

Make sure she knows your little experiment will only last 10 weeks and then she won't ever have to worry about you stressing over making the wrong choices ever again. I'm sure nothing can possibly go wrong with that...

Kudos for Stuboy for trying to organise this


I personally believe the scores should be marked on how far off you are each week. And not how far off you where at the end of the 10 game weeks.


If that makes sense

I'll state again that I'd like to see two scores. One for however you want to score the actual predictions and one for the relative ranking of the scores.

For example, take my predictions and sort them by points for the week. Then do the same thing after the scores are in and compare the two lists. Subtract a point for each number of "relative" places I'm off. If I give two players with the same score, they would get the same ranked place. Zero (nil) is a perfect score (there may be better ways to score this and I would be OK with whatever seems to give the "best" answer).



To me, a relative score is every bit as good as an actual score as what I really want to know in the above example is that it would be good to have Gerrard, Drogba, Giggs & maybe Arshavin. The actual scores are less important.

This system does seem overly complicated to me could just be my small brain though
If you would like to pm a detailed description of the Rank system i may find this benificial

I think it's worth pointing out that given the spirit of what Wyld suggested, this measure isn't really measuring what he claims. This is a measure for predicting the median score, rather than the mean.

For example, forwards tend to score 2 points a week, unless they score or assist. Unless he's likely to score more than once every other game, your best guess is 2. For example, Andy Caroll has scores of 17,2,8,2,2,5, for an average of exactly six. If you guessed 6 every week, your difference would be 26; if you guessed 2 every week, your difference would be 24. So 2 is slightly better, in spite of the clear fact that 2 was an atrocious guess on average and 6 was bang on. In fact, guessing anywhere between 2 and 5 every week gives exactly the same score after 6 weeks.

(I recognise that you are supposed to alter your guess each week to take into account of opposition strength - but the fact remains you shouldn't be guessing at an average score. For example, if you think "I think Anelka has a 33% chance to score/assist against Arsenal at the weekend", then you should guess 2 points, not 3.3 or whatever).

Probably worth agreeing exactly what it is you are trying to predict, and then come up with a scoring system to match.

This is exactly why I said is 10GWs enough: Over time we can see which formula (if any) is any good by the average ppg off target as a %; so the smaller your average % the better you are doing?
So if you predict 2, 4, 6 and it's 3, 6, 6 you are 50% away, 50% away, 0% away, leaving you 33.3% away overall, is that right??

You know, this stuff all makes so much sense in my head, but seems to get all fuzzy when trying to put it to paper.



Let me try it this way. I don't think anyone can predict with any (consistant) accuracy the points a player is likely to get for any single given gameweek. There's just too much variability. But it would still be useful to know whether player A is going to score more points than player B (in any given gameweek), and I think this approach may have a better chance of success. This isn't exactly what Wyld's formula is designed to do, but might prove useful even if it never gets anywhere near the actual points a player scores in a single predicted game week.


This is true and should be accounted for. The scores need to be shown by week to see if anyone is getting close to a "single game week" predictor. But it's OK to to also take into account a prediction over a period of weeks as most people will hold a player for more than one week. But I agree that this 2nd measurement, although useful, is not exactly what Wyld's formula is designed for.

I'm unlikely to enter, but would suggest that players declare whether their entries are "model" "model plus judgement" or pure judgement. People should be able to enter more than one of these - for instance enter their model "as is" plus another with judgement added.

There was a day when I could not have resisted this.

Alas the days.

Do inform the results to all. And please, please, do collaborate on the best aspects of all the ideas to make one master idea at the end.

Fantatsic work there cincirollers. Basically It wont allow you to say the score a player is going to get ! but may be usefulll in helping choose between to players?

Thanks interesting stuff

This is turning into another good thread – so thanks for everyone’s contribution so far. I think it’s important to remind everyone what Wyld's challenge actually is (and I also hope Wyld posts to acknowledge that his challenge still stands)



So this has to focus on individual GWs and being accurate to that GW - not how accurate you are across a larger period - although a larger period, say 10GW allows us time to see whether the formula is lucky or accurate, nor whether you are good at comparing points for 2 different players etc.



I agree with this too and that's why I want to be able to come up with a marking system for individual GWs



Personally, I think 10 GWs should be enough to fathom whether a prediction formula is accurate or not. Not a fan of your scoring method however. Firstly, taking your example above, if 2nd person’s prediction is 4,8,6 then this means both formulas are similar, which they’re not as one is over predicting. Secondly, if person predicts 3,6,12 does that make them 0%away, 0% away, 100% leaving you 33% away overall? The second formula is more accurate but is deemed the same as first.



Disappointed to read you won’t enter but I had already made the decision that people had to say whether they were using 100% formula, formula with ability to use outside knowledge or pure judgement. As for entering multi teams – I think I’m going to be busy enough so one team per person but people can publish their own additional predictions ahead of time if they wish



I agree with this but I need to be fair to Wyld’s challenge and this was predicting a player’s FPL score next GW.

But how to mark this fairly seems to be the area where I’m almost thinking someone needs to create another formula for.....

So, what do people think of this? The accuracy is done via points with a person trying to score as close to 0 as possible. As has already been shown, it is more likely that players will score relatively low – 6 and below, so predicting under the actual score has a higher penalty than predicting above the actual score. At the end of a GW the totals are added and we have a ranking of how well people did. If you come 1st, you score 1 point, 2nd = 2 pts etc. Person at end of 10 GWs with lowest overall ranking score shows they were best at predicting each week.

In the meantime, please see below my individual GW scoring and let me know if you think this is a fair way of penalising or scoring people – if not, any other suggestions please.

The point table....

Would be good to get Wyld's opinion on how one would measure success, as it was his statment in the first place.

There should be points for correctly predicting the highest scoring player, as this would be beneficial for picking the captain.

Yes, this is basically what I was trying to say...


And this is basically why.

Agree that Wyld should help decide how we best measure the accuracy of everyone's predictions since he has so boldly put the effectiveness of his formula up for public scrutiny.

I'll happily be part of it, and I won't be using a system (although I use proprietary models a lot in my profession), I will be purely discretionary.

Am OK with whatever scoring system. I don't need to win, but would really like to learn something in the process.

I'm in. I'd like to enter a formula and a judgement entry, if that's allowed.