For such things you can check www.11heroes.com . They always have in addition to that Game Odds, goalscoring chances and clean sheet odds for the matchday. For the semi finals it is:

http://www.11heroes.com/article/prev...ro-semi-finals

They also go in some lineup discussion with there partner rooms. For Germany it is expected:

Goal: Neuer
Defense: Hector, Höwedes (slight doubts), Boateng, Kimmich
Midfield: Schweinsteiger (no doubt), Kroos, Özil, Götze, Draxler/Schürrle (tendency goes to Draxler, but BILD, a german newspaper, says Schürrle will play)
Forward: Müller

So it is very unlikely that either Weigl or Can will start.

Can has started and Mustafi who I was expecting to play has not. Thanks for the link and will bookmark it. Better make my swaps on DK's game then.

I loved FanDuel's email. "Antoine Griezmann was the highest scorer and best value choice to boot. [...] Too bad more of you didn’t pick him - he wasn’t in your top 10 most-owned players"

I almost wish WilliamHill would emails like this for the laugh value. "France beat Germany 2-0, too bad more of you didn't bet on this exact score, fishes".

I saw this article but have never heard of Transferkings, has anybody played on there or ever even heard of them? Seems like such an odd move

http://www.sbcnews.co.uk/esports-fan...facing-roster/

I read about the acquisition as well...first time I had heard of them.

I'm thinking about building my own projection model/algorithm sort of thing and just wondered if anybody has any thoughts. I have very limited coding knowledge so will be starting from scratch.

Has anybody built there own and is it any use (should I bother)?
Has anybody got any pointers, links to articles etc?
Would anybody who has built one like to have a chat with me about it over twitter?

Thanks

I've built my own. You don't need coding knowledge, you just need Excel. Being a member of FPL Scout is good because you can get historic Opta stats going back five years.

It's not that hard to get info on football data analysis, but it's not that easy either. Most of it won't matter for now anyway, you have at least a hundred hours of work just to get the basics using the above.

Thanks for the response

Im reasonably confident with excel, I'm more than willing to put in the time and Im already a member of FantasyFootballScout where I would probably pull the data from. It's just a case of trying to figure out where to start and seeing if there are any articles that can show me how to get started?

How does your work, do you drop the data in each week and then all the formulas do the rest or does it involve a lot of manual intervention every week?

Half the reason I do it is to get better at Excel. It's rare that I don't know how to do something immediately at work, but this gives plenty of opportunities to do something even more complicated.

So I want as little manual entry as possible. I paste in the results and the Opta data, and that's it. However, there're often new names to add due to DK or FPL name-laziness, which I need to add by hand the first time they show up.

I'd start by pulling in the name & club data from FPL or wherever (https://fantasy.premierleague.com/a/...s/total_points), and figure out how you want to link that to the Opta data from FPL Scout. The names will not match. You'll also need to figure out what data you actually want from FPL Scout, as there's a lot of data but most of it is irrelevant for your purposes. I have custom tables on FPL Scout for what I actually need.

Does the FPL data include crosses (with the same definition DK uses)?
And DanB yeah, it's doable, you're just never done with tinkering with the thing I haven't found any model-building related article, just trying to grab tidbits here and there.

Yeh, FPLS uses Opta, which is also what DK uses.

I meant FPL scout, but I guess that's what you meant too, good, thanks. I guess I will subscribe this year then, need to get this numbers somehow.

Im confused now, I thought you were both referring to Fantasy Football Scout - http://www.fantasyfootballscout.co.uk/, are you not?

I am old school, more of a pen and notebook for my homework. For DFS find the game theory aspect more fascinating than projection models/algorithms.

I try to ask why does the crowd persist in making wrong choices despite there being evidence that their choices are wrong. In short, I try to understand the bias inherent in making forecast/picks/projections.

As an example Ronaldo was picked by a majority for all the games in Euro 2016. But there was enough information out there that he was not an optimal play. Portugal coach was much more conservative and insisted on playing Ronaldo as a forward in a 4-4-2 whereas Ronaldo naturally thrives coming from the wing in a 4-3-3. The optimal picks were Portugal defenders who made record interceptions and alongside some clean sheets in crucial matches.

No projection model can explain why the crowd persists with their bias in picking Ronaldo who fantasy wise had a very average tournament relative to his high salary. In DFS, discerning the bias in the crowd and having information to see they are wrong makes the money.

On making models these two in the links below are very good and helpful. Check them out and drop them a line.

https://twitter.com/teouchanalytics

https://t.co/RSjXJHYS6J

https://twitter.com/artemidorus_1

https://t.co/OvURkQd3tF

That's almost exactly what a model does. It says who the best players to pick are based on their history and on their recent numbers, and it doesn't care if you like Ronaldo or not.

You should of course overrule the model if you know something it doesn't - like the coach is conservative. And it isn't very useful for short term teams like in the Euros. But if you want to know which fixtures are best for the first three weeks of the FPL season, or which 7.5 midfielder to pick, or which 3rd tier mids are best in DK, a good model can do very well.

Yes indeed. I suck.

Oh come on Jean don't be so hard on yourself.

Thanks everyone for your thoughts, my intention for my projection model would be to use it alongside other prep trying to build in my own views/weightings and looking at other data.

Im hoping I can just drop in a custom table from FFS and the data from DK each week with some minor alterations for any kinks. I assume this is doable once the spreadsheet is fully set up and robust.

So now I just need to work out what I want it to output, what data I need to get it to that point and then how I set that up. Sounds like I need to work backwards.

If anybody would like to e-mail me there model so I could have a look at one working in real life that would be great

Thanks for the links. I don't doubt that there's interesting stuff to be found there, but browsing through it all it looks like there's, as usual, a lot of noise and little usable facts. One of the factual pages: https://mathematicallysafe.wordpress...-and-fixtures/ essentially says that correlation from game to game has R^2 of 12% while correlation between the previous 6 games and the next one has an R^2 of 16%, ie slightly better. All that is for FPL scoring, but I assume the same behavior would appear for any DFS scoring system. I wish the author would have gone to the next logical step and made the same calculations for the correlation between the previous N games and the next one for all N (he's done it for N=1 and 6) and see which N gives the best "predictive factor" R^2.

I was just reading some of his old posts. https://mathematicallysafe.wordpress...lying-metrics/ seems stuck in overfitting-mode, but some of it's good.

I'd imagine that finding the best rsq for the next n games is going to be stuck in overfitting too. To me it seems more fruitful to look at the predictive power of relative team strength.

There are models for projecting ownership percentages.

Indeed there are models that project ownership and performance. The hardest task is in discerning the right information to go against the ownership or to go along with it. In short, knowing and timing when the crowd is wrong.

There is a few more that have good models and let me share them

https://goalprojection.com/2016/07/1...elling-review/

https://goalprojection.com/projections/adjusted-goals/

https://goalprojection.com/2016/03/2...odel-20032016/

https://mobile.twitter.com/goalprojection

https://twitter.com/fantomen?lang=en-gb

http://blog.stryktipsetisistastund.se/

In case I wasn't clear, I was speaking of using the number of fantasy points scored in the previous N games to predict the next (one) game. And finding out what the best N is (ie is short tem form a better predictor than longer-term / whole season numbers). I don't think there'd be much overfitting there.

For example, and the end of last EPL's season on mondo, Deeney and Ighalo were both similarly priced, with Ighalo the better performer on the season and Deeney the better perfomer over the past 5-10 gameweeks. Given they had the same opponent and assuming the same expected playing time, who was the best play?

https://mathematicallysafe.wordpress...-and-fixtures/ has some similar analysis, and concludes that "opponent form/strength" is almost irrelevant, which looks very wrong. I'd guess that the way he determines opponent difficulty is bad.

Back to your initial question DanB, one of the major decisions you'll have to make (if you get past the tedious countless hours of writing down excel formulas) is what is the best data to base your prediction on. Is is "recent form" type data (like past 6 gameweeks?)? Is it whole season data? Can you even include data from past seasons without having the massive changes in team strength / situation screw everything?

From Garrincha's linked goalprojection.com:
https://goalprojection.com/2015/11/2...d-goals-model/

I'm kind of happy about these. Long term numbers, that I've been using, seem to be better predictors (at the team level at least) than shorter term numbers. I hope the same stands true for individual players too (although presumably less so).

Also, taking into account shots / SoT (and presumably overfitting, as Sciolist would mention ^^) only adds a tiny it to just using goal difference to assess team strength.

(DanB: another biggie, team strength. Presumably you want to work out how many fantasy points each player would average against an average opponent. Then work out opponent's strength (including home/away). Then figure how much to adjust the fantasy output based on opponent's strength. Presumably different scoring categories need to be adjusted differently (for example mondogoal's points per minute not at all; point per goal scored is easy if you use expected goal scored metrics to analyse the opponent's strentgh; other numbers are much harder (e.g. draftkings crosses, or mondo's passes completed). I guess one would need to compute opponent strength for each scoring category, which I haven't done

Yeh everything you have said is exactly along the lines I'm thinking, aim is to get a base level projection working just taking into account that sort of level, team match ups etc, and then refine it through the season and see if I can build in more detailed aspects such as crosses etc.

I think once I get over the learning curve of getting the spreadsheet to work without too much intervention then I can start to adapt it towards something I really want