SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

[muppet77]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) Id keep diff teams in too. Could they be flagged as being duff for post analysis? Could you also generate a second best comp team and a third and a fourth best...? see what I'm getting at? Even better rank order the entire population off random teams and see how they do. Is there a correlation between your rankings and actual score? So you predict expected weekly points? Expected weekly points based on opposition? there was scientific test by the uni of Soton on this matter. They had a comp team that looked at opposition each week and then looked ahead like a chess game to see what was ahead, as subs were finite. Will post a link if I find it.

[muppet77]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) http://www.ecs.soton.ac.uk/news/4028

[muppet77]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) Bombed out??! http://www.squadguru.com

[muppet77]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) http://metro.co.uk/2012/08/16/rooney-or-rvp-in-your-fantasy-football-team-ask-squadguru-538557/

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

Id keep diff teams in too. Could they be flagged as being duff for post analysis?

Is this a reply to my post about teams that might contain injuries, or a question about the teams that never get through, the ones that are rejected because of not meeting all the criteria? If it's the first question, I probably would do that. I'd either label the teams and post them up as failures, or simply state on here the number of "failing" teams.

Could you also generate a second best comp team and a third and a fourth best...? see what I'm getting at?

I could! If I would do it, I would need to establish a set of rules to determine which team is 2nd, 3rd and 4th. For instance, repeat players allowed? Using more than one team, I feel like I would be spreading out (and therefore artificially inflating) my success/failure rate. It would feel a little like cheating (putting multiple teams in, to increase the chances of a winning team).

Even better rank order the entire population off random teams and see how they do. Is there a correlation between your rankings and actual score?

So a league table, ye? Could you explain a little what you mean about correlation between rankings and actual score? By rankings, do you mean the ranking of each team (after the points have been distributed), or a different kind of ranking? And by actual score, do you mean WDL or goal difference in a game or something else?

So you predict expected weekly points? Expected weekly points based on opposition?

Predicting weekly points would be one way of doing it, and now that I've modified the rules so that I can easily compare multiple teams - this would definitely be a viable way of doing it. And now that you mention it, there are really a lot of options here. I could use the player expected points to predict base wins, draws and losses. Although, predicting some of the things on the rules might be tricky:

• number of minutes a player would play (perhaps pointless and would just imagine all players will play 90 minutes and auto-assign 2pts to each team player)
• goals scored by goalkeeper or defender (impossible to predict a GK goal, unless he works on set pieces. As for goalscoring defenders, I suppose more predictable, depends how dependable the stats are)
• cards (teams/players with higher likelihood of being carded are well known I'd say, and you can predict when they are more likely to receive them, especially if you factor in the referee too)
• own goals (unless it was highly predictable - like Koscielny or Ridgewell a few years back, not easily predictable)

there was scientific test by the uni of Soton on this matter. They had a comp team that looked at opposition each week and then looked ahead like a chess game to see what was ahead, as subs were finite. Will post a link if I find it.

Thanks for those links, i'll look through them and see if I can pick up any tips!

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

http://www.ecs.soton.ac.uk/news/4028

Reading this reminded me of an idea I had a few years back, where 24-7 I was literally feeding my PC player data in an effort to analyse and predict player performance, and by extension - team results. Ultimately, I found that it was taking way to long to collect so much data, and then the actual analysis was taking even more time. I like to have time to actually watch matches and I didn't have a team of scientists to help out! I've downloaded the paper that they published (available here), it's only 7 pages long. SquadGuru is a bright name for it too, pity that they haven't posted since December, but there could be a couple of reasons for this. Looking at the team status (on their site), they seem to have stopped posting after the halfway period. Maybe they were offered a contract or something, i've seen that happen sometimes - people put out a good idea that gets snapped up by a company offering big bucks...hopefully i'm wrong and they'll start posting again next season. On their site, they had an average positions (in FF tables) of 50, 48, 55, 54, 48...pretty good. And the bigger project that they are working on sounds pretty awesome too (ORCHID - looks to combine human knowledge and computer systems to solve scientific problems)

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) Right, got the COMPUTER team now, if there are any injuries or suspensions before the first match kicks off (City vs. S'oton 12:45pm) and i'm able to access the internet (i'll be in transit!!) i'll make the changes, otherwise these are the starters. [ATTACH]5857[/ATTACH] The modified rules will be in play this weekend. At the moment, the random team generator is whizzing away, trying to generate about 10,000 random teams (could be an few hours). If all works out, i'll upload the spreadsheet. :ok

[muppet77]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) By predicting points I meant, how does your pc pick the best team? What metic(s) does it use? i guessed it was expected points.

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

By predicting points I meant, how does your pc pick the best team? What metic(s) does it use? i guessed it was expected points.

Ah, I see. Well, with the modifications made, I can't see why I shouldn't be using expected points, and in fact - it would lend itself to easier backtesting. I'm going to switch it out now and see what difference it would have made to last week's team selection. Expected points makes much more sense than what I was using, considering the nature of this project, because it's basically judging player efficiency for the specific purpose of Fantasy Football!

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) For some reason, i'm unable to upload the spreadsheet with 10,001 randomly generated teams in it...I have it, just not able to upload it on the website at the moment. Interesting fact, Berbatov was the most commonly used player, appearing 551 teams across the 10,001 teams. Unlike the other uploads, this is in the xlsb format because otherwise there wouldn't have been enough rows to contain all the teams (well, 10,001 is a big number)

[muppet77]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) What is the distribution of expected points like for all teams?

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

What is the distribution of expected points like for all teams?

As soon as I've incorporated what you suggested a few posts earlier, i'll post up all the team & player expected points :ok

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

For some reason' date= i'm unable to upload the spreadsheet with 10,001 randomly generated teams in it...I have it, just not able to upload it on the website at the moment. Interesting fact, Berbatov was the most commonly used player, appearing 551 teams across the 10,001 teams. Unlike the other uploads, this is in the xlsb format because otherwise there wouldn't have been enough rows to contain all the teams (well, 10,001 is a big number)

Just reading through the thread again, realised that I hadn't posted the 10,001 randomly generated teams after having those initial difficulties. If you check the file (date last modified, etc), you will see that I haven't touched it since 03/04/2014 07:59 (yes, I know that those things can be modified, but I have no reason to *cheat*), so despite the Man City game already kicking off, these teams were already generated in advance of that. I uploaded the file to SendSpace, so anybody who did want to look at it, just click the link. Looking at the file totally isn't necessary though, as after this gameweek has finished, I will publish some sort of league table to show where COMPUTER finished against all those random teams. Remember, the goal (for this week at least), is to get into the 95th percentile, which would have it in the top 500 of teams.

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) the points distribution this week should be interested - I'm only 10 players because I wasn't able to update the team due to travelling (Rooney not playing). Let's see how they do with a -1 player handicap! :D Worth noting that i've now found a method of acquiring all players injuries (not suspensions yet), so in the future, teams(COMPUTER and the randomly generated ones) will not field any injured ineligible players. Doubts and late fitness tests WILL be included, because that is part of the gamble of playing fantasy football!

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

I'm only 10 players because I wasn't able to update...(Rooney not playing)

just for the sake of knowing (personally and for this project), the guy who MAY have stepped in for Rooney could have been Daniel Sturridge. The reason I say may, is because changing one player changes the dynamic of the team, without Rooney, no longer have two United players, and the value of the team changes, so a variety of different changes could have happened to the entire team. With that potential change, the total value of the team would have dropped to just under £10,000,000 million, which could have been distributed around the team differently...but let's not speculate on what could have happened, and just look forward to the 10 men fighting for their lives :p

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) I wasn't able to check the random pts teams because i've been super busy this week (and I'm not sure I can see an end for my impending busy-ness), but I do know how the COMPUTER team did last week. It was a much lower points tally of 30pts, mitigating circumstance perhaps (me not being able to check line-ups before kickoff, so ending up starting with Hazard & Rooney), but that is going to happen sometimes!). It was still interesting to see the guys perform though. Hazard came on and changed the match, Navas had a good second half against Southampton, Coleman and Barry were BADASS against Arsenal!. Coutinho was rather quiet against West Ham, and it was no surprise to see him get subbed at half time. Anyway, I will try and get my team up before Friday evening!

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) It's that time again - team submission. The RANDOM teams will be uploaded shortly, here is the COMPUTER team: [ATTACH]5895[/ATTACH] Should be some good matchups this week, hopefully get closer to the 50pts barrier than 30pts this time round!

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) I've modified the RANDOM team generator, so that now it shouldn't include any injured players. On top of that, it will also only generate teams of players who have played more than 750 minutes, otherwise it would just be populated with benchwarmer. Again, i've had to upload the file to an external site, you can find it here. The format is XLSX. The top three occurring players are Berahino, Shola Ameobi and Peter Crouch, with 1091, 1075 and 1071 appearances between them. Norwich players made up 10% of the total players, while Hull only made up 0.006%.

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) TEAM SUBMISSION!!! Didn't put up a midweek team, here's the team for today. [ATTACH]5905[/ATTACH] I just noticed a few things, ONE the team i posted last week didn't load up on the thread (dammit) and TWO, i didn't update their pts. I'll deal with that shortly

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) [text removed and reposted]

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) Since after the first week, the points system was modified, i've returned to the first team and reassigned points. Week 1 (uploaded 28-03-2014) [ATTACH]5911[/ATTACH] 51 pts (btw - Neville's team from Week 1 got 48pts) Week 2 (uploaded 03-04-2014) [ATTACH]5912[/ATTACH] 27 pts (Rooney was in my squad, didn't play!!!) Week 3 (uploaded 12-04-2014) [ATTACH]5913[/ATTACH] 39 pts Week 4 (this week) [ATTACH]5914[/ATTACH] So, that's 51, 27, 39 - an average of 39pts so far. Until I have completed developing an en-masse pts distributor, i'm not able to compare my teams against the randomly generated ones, but hopefully i'll have it one day!

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) 2nd highest pts total so far, a nice 43pts. Would have been higher if I'd actually had a left-back who played, but that's tough. Demel!!!! And I watched the Spurs game, after seeing that Adebayor was one of the COMPUTER picks, considering he was up against Fulham, i'd expected him to fill his boots...ah well! To be fair, Fulham put in a good shift, and things could have gone differently.

[muppet77]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) Good work. Have you considered following your first team through week by week like a proper FL team rather than changing each time?

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

Good work. Have you considered following your first team through week by week like a proper FL team rather than changing each time?

I'm not too far away from sorting out the pts distributor I was talking about, so at least for next season I'll be able to use that effectively. I can definitely do that (or at least maintain the same team from now until the end of the season and see what happens). There aren't many games left (3/4), so do you think it would be worth keeping the team from next week, and running it parallel to a newly generated one each week? I know that the rules are slightly different in each FF competition, so what rules should be applied in this case? 3 changes per week? 5?

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

Good work. Have you considered following your first team through week by week like a proper FL team rather than changing each time?

Ah, having just reread what you wrote, you mean the first time I submitted from "Week 1"?

[muppet77]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) Yes look how your week 1 team progresses through each week. Keep a table of position compared to the other randoms and points gained each week. You do need to factor in subs somehow. If it were my project I would at this stage focus entirely on building a FL selector that you can run when you want to make a sub. Your reward is a cash prize! pick a comp and use their rules. You can even pick your best ten teams and enter them.

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?)

Yes look how your week 1 team progresses through each week. Keep a table of position compared to the other randoms and points gained each week

Bad thing about generating the random teams (which i'd paused because i wasn't sure if i would get access to the data i needed for long term pts distribution), is that suspended/injured players will almost always sneak their way into the teams. I'll need to find a way to get around that.

You do need to factor in subs somehow. If it were my project I would at this stage focus entirely on building a FL selector that you can run when you want to make a sub

You're right, it's much easier to have it auto-select a team of 11, as far coming up with a line of "thinking" that could have it automatically choose which players to sub-in/sub-out, I guess i'll need to first choose a competition to model. The problem, is that there are certain pts that I will never be able to measure (saves by goalkeeper, for example) without turning the player scraping process (it's already pretty time intensive because there are 700-odd players in the Premier League, and this project requires them all to be given a shot at making the team) into a day/two day long thing. Fortunately, i've managed to sort out an efficient way of the data collection side relating to OGs, Cards, Assists and Goals. I just need to apply my "rules" to those random teams. A problem will be the subs though, and in applying the sub idea to the COMPUTER team (making a fixed number of changes rather than a completely new team each week), i would presume that the random teams should be afforded the same luxury. And *that* would definitely destroy the speed at which 10,000+ teams could be generated (if even possible...ok...it's possible, i just don't want to think about it, it will take forever!) ----------------------------------------------------------------------------------------------------------------- Over the summer, I'm definitely going to be looking at making the whole thing more automated and less hands on (certain steps require me to be monitoring stuff manually). Of course, i'm totally going to be generating player stats for everyone that will be involved in the World Cup (every country) once the team lists are posted up. It will be fascinating to see what the numbers say about the players. I already have data for almost all the top leagues across Europe, I just need to start considering the lesser known leagues and countries too! So if I manage to collect all the data I need, I will post up a FF team for the World Cup too. And that one, I would put on here from the start of the tournament, and follow whatever fixed rules I'd find from some online competition (surely there will be a few to choose from for the World Cup).

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) data collection for this week is probably about an hour away, after that, i'll post up the points that the initial team I selected would have achieved every week since i've been doing this. Granted...not very long, it's been a month, however, and given that they returned the highest pts total so far, so in theory, they might also return the highest average over the period of a month. Anyway, we shall see.

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) How has the first COMPUTER team performed over the past month I hear you ask? I've just had a look, and here are the results: W1 - 51pts Missing Players: No Average Pts: 4.63 Readjusted pts total (to account for missing players): not neccesary...51 Perfect start to be honest, an average pts haul of 4.63 per player, the highest of the four weeks. Understandable, since the squad was selected for that gameweek W2 - 34pts Missing Players: Yes. 2 out - Boruc & Rooney Average Pts: 3.78 Readjusted pts total: 42pts (this is simply 3.78 * 11) Playing with 9 players in this week. You would expect Rooney to score or assist...they were playing against Newcastle, whom they beat 0-4. Maybe a blessing on the other hand, since Southampton shipped 4 goals at City. W3 - 18pts Missing Players: Yes. 3 out. Boruc, Hazard & Rooney Readjusted pts total: 25pts Low pts here, regardless of readjustment, but missing Rooney & Hazard is effectively reducing the attacking capability of the team in a big way, so not unexpected here W4 - 36pts Missing Players: Yes. 3 out. Boruc, Hazard & Rooney Readjusted pts total: 50pts Again, hampered by missing players, but if the readjusted pts total is anything to go by, this could have resulted in the highest return yet. If we look at the average readjusted pts returns (because I believe that is the only fair way to do it considering there were 2 weeks with 3 players missing), there is a respectable 42pts. Still, at this point, me saying 42/51/25 doesn't really mean anything as there is no point of reference...i'm working on that ;)

[lunatism]

Re: SYSTEM TESTING - Fantasy Football (Can you beat a robot?) It seems that injuries have ruled out a few who probably would be included in the team this week (Silva, Navas perhaps), as always, will be interesting to see how these guys fare this week. Hopefully, at least there will be 11 starters this time! :hope [ATTACH]5938[/ATTACH]