Last week we didn’t do too bad – reaching the top 10K in the Daily Mail League and in the top 500K in the FPL. The SquadGuru selected Lukaku as part of our attacking front with the expectation that the he would be chosen as the main striker for Chelsea. The algorithms naively assumed Lukaku’s goal scoring percentage at West Brom would carry over to the Chelsea (who scored even more goals!), ignoring the fact that Mourinho is still testing his new recruits out and may want to rotate his players around to get a sense of what works best with his new arrivals. Contrast this with Van Persie moving to Manchester United where he definitely carried over his goal scoring percentage to his new team – and that happened because he move to an equally good team and holding an equally strong/certain position in the team. Note that 12% of managers still selected Lukaku, showing that the SquadGuru’s selection was not that absurd.
Our plan is to encode such features of teams into the SquadGuru. In particular, when a player gets transferred to a new team it’s important to encode the updates to our priors based a number of factors:
- The difference between the two teams in the league. Thus if a player goes from a significantly lower-ranked team to a higher ranked team, the prior should be adjusted to factor in the fact there are other good players who already scored for the top team and may compete with the new player for position or goals. Differences between two teams in two different leagues (e.g., Championship v/s Premier League or Liga v/s Premier League) are even harder to understand and capture as they are from a completely different game mix. But definitely something worth looking at. Our star selection of last year, Michu, was a product of simply transferring the prior from the Spanish league to the Premier league… something we didn’t see for other players this season unfortunately.
- The organisation of the new team. It is hard to pin down a measure for how a team is re-organised by the manager and how frequently he does so. Understanding this would enable us to formulate priors that take into account appearance times better. Thus, if the player comes in as the main striker (e.g., Suarez or Van Persie), you carry over the performance of the player.
- The uncertainty at the beginning of the season. Since every team will be checking what works or not with their new summer recruits, it is difficult to predict the appearance of players on the pitch and the behaviour of the players. Anelka leaving this week came out of the blue and would have hurt (some) managers. Similarly for Bale possibly moving to Real Madrid. Hence adding some uncertainty to your beliefs (accumulated over the previous season) is important.
Watch out for the next post (out in an hour) for our Week 2 selections!