Review of Gameweek 24

Fantasy Football Robot had a reasonably successful week in GW24, scoring 40% precision by making 15 predictions of players that would score 6 points or more and being correct in 6 of them.

Goalies: 0 or 1 correct, 0% precision. Predicted Hart. Wrong again.

Defenders: 4 out of 4 correct, 100% precision. FFR correctly identified that Chelsea would keep a clean sheet at Man City. A very rare feat was that all Defenders avoided getting a yellow card, thus keeping each of them the correct side of 6 points. The next most fancied side to keep a clean sheet (with probabilities narrowly less than 50%) was Southampton and indeed, they succeeded as well.

Midfielders: 2 out of 7 correct, 28.6% precision. Cazorla and Adam Johnson correct.

Forwards: 0 out of 3 correct, 0% precision. Suarez, Dzeko and Adebayor all missed the mark.

FFR Utd, the team formed by the highest probability players, scored 51 points. Compare this to 45 for the average human team, 16 (just 16!) for Random Wanderers (a team of randomly chosen players) and 55 for Moneybags City (a team of the most expensive players). FFR Utd’s points came from Chelsea defenders and Adam Johnson.

Gameweek-by-gameweek summaries of fantasy teams available at
http://fantasy-football-robot.appspot.com/teams?gw=24
and performance statistics at
http://fantasy-football-robot.appspot.com/appraise?gw=24

Follow me on Twitter at https://twitter.com/FantFootRobot

Review of Gameweek 22

Predictions for last gameweek from Fantasy Football Robot were again fairly mediocre. A boring success rate of 29.2%. It made 24 predictions of players that would score 6 or more points and was correct in 7 instances

Goalies: 0 or 1 correct, 0% precision. Predicted Hart. Wrong!

Defenders: 4 out of 12 correct, 33.3% precision. FFR correctly identified that Arsenal would keep a clean sheet. FFR was wrong about Sunderland (who conceded twice in the first 40 mins) and Man City.

Midfielders: 3 out of 10 correct, 30% precision. Toure and Navas at Man City correct, and Eriksen at Spurs.

Forwards: 0 out of 1 correct, 0% precision. Suarez ‘only’ scored 5 this week.

FFR Utd, the team formed by the highest probability players, scored 68 points. Compare this to 45 for the average human team, 57 for Random Wanderers (a team of randomly chosen players) and 64 for Moneybags City (a team of the most expensive players). FFR Utd’s points came from Arsenal defenders, Man City midfielders and Adebayor up top.

Gameweek-by-gameweek summaries of fantasy teams available at
http://fantasy-football-robot.appspot.com/teams?gw=22
and performance statistics at
http://fantasy-football-robot.appspot.com/appraise?gw=22

 

Follow me on Twitter at https://twitter.com/FantFootRobot

Review of Gameweek 21

Predictions for last gameweek from Fantasy Football Robot were fairly mediocre. Not bad, just average. It made 17 predictions of players that would score 6 or more points and was correct in 5 instances, precision of 29.4%

Goalies: made no prediction. Top ranked goalie by probability was Tim Howard and he scored 7 points.

Defenders: only 2 out of 13 correct, 15.4% precision. FFR correctly identified that Everton would keep a clean sheet but frustratingly, John Stones picked up a booking to score 5 and Alcaraz did not feature.

Midfielders: 1 out of 2 correct, 50% precision. Hazard correct. Prediction of Walcott was scuppered through injury so if/when I eventually getting round to making FFR aware of injuries and suspensions, precision should increase.

Forwards: 2 out of 2 correct, 100% precision! Luis Suarez and Welbeck both delivered the goods.

FFR Utd, the team formed by the highest probability players, scored 83 points. Compare this to 57 for the average human team, 52 for Random Wanderers (a team of randomly chosen players) and 81 for Moneybags City (a team of the most expensive players). FFR Utd’s points came from picking Everton defence and goalie, Suarez as captain, and Hazard, Welbeck and Adebayor as supporting attackers.

Gameweek-by-gameweek summaries of fantasy teams available at

http://fantasy-football-robot.appspot.com/teams?gw=21
and performance statistics at
http://fantasy-football-robot.appspot.com/appraise?gw=21

Review of Gameweeks 17, 18, 19 & 20

The gameweeks came stupidly fast over Xmas* and I feel no shame admitting I struggled to keep up with my administrative duties. Fantasy Football Robot managed to make timely predictions for each round but I did not manage to write gameweek reviews** so here I offer an abridged review of gameweeks 17, 18, 19 and 20 that were played over the Christmas and New Year period.

In total, FFR predicted 78 instances where it felt a player would score 6 or more points and was correct in 25 cases, precision of 32.1%. This breaks down by position:

Goalies: 1 out of 2 correct, 50% precision
Defenders: 12 out of 44 correct, 27.3% precision
Midfielders: 10 out of 28 correct, 35.7% precision
Forwards: 2 out of 4 correct, 50% precision.

In the four gameweeks, FFR Utd, the team formed by the highest probability players, scored 215 points. Compare this to 176 for the average human team, 164 for Random Wanderers (a team of randomly chosen players) and 277 for Moneybags City (the most expensive players). It is fairly typical that FFR Utd beats the average human team and scores bang in between Randoms and Moneybags.

I will write a proper half-season summary soon and address whether FFR is scoring better than a drunk monkey at a dartboard.

Happy 2014!

Gameweek-by-gameweek summaries of fantasy teams available at
http://fantasy-football-robot.appspot.com/teams?gw=17
http://fantasy-football-robot.appspot.com/teams?gw=18
http://fantasy-football-robot.appspot.com/teams?gw=19
http://fantasy-football-robot.appspot.com/teams?gw=20

and performance statistics at
http://fantasy-football-robot.appspot.com/appraise?gw=17
http://fantasy-football-robot.appspot.com/appraise?gw=18
http://fantasy-football-robot.appspot.com/appraise?gw=19
http://fantasy-football-robot.appspot.com/appraise?gw=20

* do not interpret this as my promoting a winter break. I love Xmas fixtures.
** Robots > Human. Be afraid.

Review of Gameweek 16

This was a fairly average week for Fantasy Football Robot which managed a non-exciting 27% precision in its predictions. Here’s a summary:

Goalies, 0 players correct out of 1, 0% precision. Pantilimon predicted (again)and he conceded three. Boo, hiss.

Defenders, 4 players correct out of 17, 23.5% precision. FFR predicted clean sheets for Chelsea and Crystal Palace (both wrong), Man City (also wrong, vs Arsenal) and Sunderland at West Ham – this one, gladly, was correct.

Midfielders, 2 players correct out 7, 28.6% precision. Fernandinho and Toure correct; Ozil, Lampard, Nasri, Navas and Hazard incorrect. Despite failing to score 6+, most of these incorrect picks scored more than the perfunctory 2 points for turning up.

Forwards, 1 player correct of 1, 100% precision. FFR spotted that Suarez is in good form. Along with everyone else.

Once more, despite ordinary predictive success, the aggregate points scored by the team of the players with the highest probabilities, FFR United, was superb. FFR United scored 87 points, half of which came from having the Uruguayan knasher as captain. In contrast, a side made up by a random 11, Random Wanderers, scored a lacklustre 44 points and Moneybags City, the most expensive 11 players, scored only 85 points. I track the performance of these hypothetical teams each week. Take a look at the links below. I like it when FFR Utd beats Moneybags City; it demonstrates that the knowledge edge offered by prediction algorithms beats having the magic ability to pick a prohibitively expensive team of the most expensive (and best?) players.

As always, this week’s summaries are at:
http://fantasy-football-robot.appspot.com/teams?gw=16
and
http://fantasy-football-robot.appspot.com/appraise?gw=16

Review of Gameweek 15

A forgettable week for Fantasy Football Robot which managed only 15% precision in its predictions. Here’s a summary:

Goalies, No goalie predicted! The highest probability goalie was Mignolet with a 41% chance of scoring 6+ points, but no goalie had a probability that exceeded 50%.

Defenders, 0 players correct out of 11, 0% precision. A disaster. FFR predicted clean sheets for Chelsea (at Stoke), Arsenal (hosting Everton) and West Brom (hosting Norwich). Chelsea laboured to a win but conceded twice; Arsenal conceded late on after good Everton pressure; and West Brom had a shocker at home to a Norwich side in not much form.

Midfielders, 1 player correct out 6, 17% precision. Ozil correct; Lampard, Toure, Ramsey, Gerrard and Hazard wrong. Not a good week for the big names in midfield!

Forwards, 2 players correct of 3, 67% precision. Suarez and Aguero correct again. Rooney was suspended.

Despite the crumby predictive performance, the team of the players with the highest probabilities, FFR United, scored a handsome 55 points (and cost a £119m, see link below). In contrast, a side made up by a random 11, Random Wanderers, scored only 21 (£82m) and Moneybags City, the most expensive 11 players, scored only 46 points (£130m).  I track the performance of these hypothetical teams each week. Take a look at the links below.

FFR Utd regularly beats the random team and has now beaten Moneybags City as often as not, i.e. it is a toss-up between these two teams in a typical week. This suggests that the information provided by the picking algorithm confers an advantage comparable to being able to pick the most expensive 11 each week (a feat beyond most budgets). I will explore this appraisal in coming posts.

As always, this week’s summaries are at:
http://fantasy-football-robot.appspot.com/teams?gw=15
and
http://fantasy-football-robot.appspot.com/appraise?gw=15

Review of Gameweek 14

I’m just catching up on posts after the midweek set of games so here is a brief review of model performance in gameweek 14 (that took part Tue and Wed nights last week). Fantasy Football Robot had an average week with 36% precision in its predictions. Here’s a summary:

Goalies, 0 players correct out of 1, 0% precision. After a few weeks of success with Pantilimon, FFR switched to Mignolet and instantly regretted it. Despite scoring 5 goals, Liverpool could not resist conceding one…

Defenders, 2 players correct out of 8, 25% precision. FFR incorrectly picked clean-sheets for West Brom and Tottenham. However, despite conceding, Kyle Walker and Vlad Chiriches got amongst the points in other ways. Two marks for FFR.

Midfielders, 1 player correct out 2, 50% precision. Ramsey right; Gerrard wrong. FFR has been reliably been picking more than 2 midfielders in previous game weeks, but not this week.

Forwards, 2 players correct of 3, 67% precision. Suarez shot the lights out and Aguero scored too. Rooney missed the mark.

The team of the players with the highest probabilities, FFR United, scored 78 points but cost a prohibitive £121m and was overweight Tottenham players. The comparable sides: a random 11, Random Wanderers scored only 22 (£90m) and Moneybags City, the most expensive 11 players, scored 88 points (£128.8m). It is a fairly common result now that FFR Utd scores between the random team and the expensive team. I shall expand on that theme in coming weeks.

As always, this week’s summaries are at:
http://fantasy-football-robot.appspot.com/appraise?gw=14
and
http://fantasy-football-robot.appspot.com/teams?gw=14

Review of Gameweek 13

Fantasy Football Robot had a really strong week with 55% precision in its predictions. Here’s a summary:

Goalies, 1 player correct out of 1, 100% precision. Once again, Pantilimon delivers the goods.

Defenders, 11 players correct out of 17, 64.7% precision. FFR correctly picked clean-sheets for Everton, West Ham and Man City but was wide of the mark with Chelsea. The precision score would have been higher had a players not missed out (e.g. Leighton Baines and Joey O’Brien).

Midfielders, 4 players correct out 9, 44% precision. FFR correctly picked Gerrard, Ramsey, Nasri, and Toure. It was wrong on Hazard, Lallana and a couple of the Man City entourage. Midfield is the hardest position to predict so I’m really glad to see a week with 44% precision.

Forwards, 1 player correct of 4, 25% precision. Rooney made hay. Suarez and Aguero missed out for once.

The team of the players with the highest probabilities, FFR United, was dominated by Man City players. The starting scored 65 points and was formed by:

Pantilimon
Dimechils | Clichy | Zabaleta
Navas | Hazard | Toure | Nasri | Fernandinho
Aguero © | Suarez

(yes: 9 Man CIty players, and team cost a prohibitive £117.3m)

Looking at the comparable teams, Random Wanderers scored 53 (£84.6m) and Moneybags City, the most expensive players, scored 60 points (£128.7m). FFR Utd beats its comparables again and I like it when that happens.

As always, this week’s summaries are at:
http://fantasy-football-robot.appspot.com/appraise?gw=13
and
http://fantasy-football-robot.appspot.com/teams?gw=13

Review of Gameweek 12

Man City roll in 6, Everton and Liverpool exchange haymakers, all the big names help themselves to goals. It was points galore again in game week 12.  As usual, let’s look at Fantasy Football Robot‘s predictive success:

Goalies, 1 player correct out of 1, 100% precision. The man picked was Pantilimon and Spurs never got near him.

Defenders, 3 players correct out of 11, 27.3% precision. The model picked clean sheets for Arsenal vs Southampton (the right call), Newcastle vs Norwich (a late Norwich goal robbed FFR of 4 correct predictions) and West Ham vs Chelsea (a terrible call).

Midfielders, 1 player correct out 5, 20% precision. FFR picked Lallana, Gerrard, Nasri, Toure and Hazard. Hazard was the correct prediction. Somehow Toure and Nasri managed to hide from the points when their team scored 6 goals. Gerrard scored 5. Lallana was subbed by Poch fairly early on, despite being the most creative outlet Southampton had.

Forwards, 2 players correct of 3, 67% precision. Suarez delivered the goods, as usual. Aguero filled his boots vs Spurs. Unfortunately, Van Persie did not play but I count that as a dodgy prediction all the same.

Overall precision was 35% Not bad. Not great.

Slightly more successful with point totals in FFR United. The starting 11 of

Pantilimon
Williamson | Santon | Mbiwa
Hazard | Toure | Nasri | Gerrard | Lallana
Aguero (c) | Suarez

scored 76 points but cost a hefty £114.2m. Looking at the comparable teams, Random Wanderers scored 34 (£87.2m) and Moneybags City, the most expensive players, scored 75 points (£129.5m). When FFR Utd beats its comparables, I am happy!

As always, this week’s summaries are at:
http://fantasy-football-robot.appspot.com/appraise?gw=12
and
http://fantasy-football-robot.appspot.com/teams?gw=12

Review of Gameweek 11

Gameweeks 9 and 10 were good weeks for Fantasy Football Robot. Gameweek 11 was rather more disappointing. First, a summary of predictive success:

Goalies, the model picked no goalies as likely to score 6+ this week so made no predictions. The upside is, if you make no predictions, you are never wrong…

Defenders, 0 out of 12, 0% precision, a terrible week. The model picked clean sheets for Tottenham, Chelsea and Man City. Going into the weekend, I was fairly pleased with those predictions. They looked eminently sensible and I was quite looking forward to a similar performance to the 64% the defenders model posted last week. Well, there was a stark reminder of the fundamental randomness in sport. Both Spurs and Man City lost games they would have expected to win and Chelsea needed a dubious last gasp penalty to salvage a draw. Suffice to say, none of the defenders kept clean sheets and none scored the requisite 6+.

Midfielders, 2 out 4, 50% precision. In slightly more encouraging news (clasping at straws), the midfielders model scored a couple of good picks. Gerrard and Hazard scored 11 and 13 points respectively. Also, just outside the 50% cutoff was Adam Lallana (model predicted he had a 49% chance of scoring 6+) and he scored a nice 12. The incorrect predictions were Yaya Toure and Ramsey, two players in a decent amount of form that had games they’d sooner forget.

Forwards, 1 of 3, 33% precision. Trusty Suarez scored another hatful of points but somehow Sturridge managed to avoid all the goals and assists in a game when his team scored 4 and conceded 0. Aguero and City missed out too.

Points-wise, the squad formed by the 15 highest probability players, FFR United, scored a modest 52 points, slightly better than the game-wide average of 45. A random benchmark side, Random Wanderers, scored 55 points and the team of the most expensive players, Moneybags City scored only 54 points. My personal score was 49. There does not seem to have been much variability in the scores I monitor this week. There was something for everyone.

As always, this week’s summaries are at:
http://fantasy-football-robot.appspot.com/appraise?gw=11 
and
http://fantasy-football-robot.appspot.com/teams?gw=11