search

alan-turing-institute / AIrsenal

Machine learning Fantasy Premier League team
MIT License
295 stars 87 forks source link

Util function to manually set a players point prediction #384

Open Tdarnell opened 3 years ago

Tdarnell commented 3 years ago

Similar to the get_player util I would suggest a set_player option, for instances like Sigurdsson where the FPL API hasn't caught up to events and the prediction still see's him as a top points scorer.

nbarlowATI commented 3 years ago

Hi @Tdarnell this is an interesting idea! There are definitely times when we humans know more about what is likely to happen than we can get from the FPL API! :)

The other competing philosophy though is to try to minimise the "human in the loop" involvement - there is a case for moving towards having the whole thing run autonomously (including making transfers, and decisions about what chips to play, though we're not yet there with the latter).

For me, the ideal situation would be to combine the best of both worlds - if we could come up with a machine-accessible way of knowing that e.g. a player is very unlikely to play a given match, we could then modify the expected minutes accordingly. But I think it's not easy! :). Any thoughts or contributions in this direction would be very welcome!

richardthomasdev commented 3 years ago

There's a few other examples of this being helpful - Liverpool's Nat Phillips is ranked highly for gameweeks 1 to 3 but he's not in training with the squad and the rumours indicate a transfer.

Maybe some sort of 'blacklist' players can be added to manually? Currently for the above case (and with Sigurdsson) I'm manually setting their points in the database to zero to take them out of consideration.

--

Quick edit to agree that I hate the thought of manually interfering with the process!

nbarlowATI commented 3 years ago

Yes, this is very true - the bot loves Nat Phillips! There are a few other players that also get good predictions but that we humans might know better too (I'd be pretty wary of putting Kane in for gw1...) Perhaps one thing we could do is put in a framework / base class for a "MinutesEstimator" - the default implementation would just do what we do now and look at minutes in previous matches, but people could easily add other derived classes that take e.g. a manual veto-list, or something that scrapes expected line-ups from match previews on the web...?

Tdarnell commented 3 years ago

I think a web scraper would be the way to go, but finding the right source will be the hard part.

On Fri, 13 Aug 2021, 14:12 nbarlowATI, @.***> wrote:

Yes, this is very true - the bot loves Nat Phillips! There are a few other players that also get good predictions but that we humans might know better too (I'd be pretty wary of putting Kane in for gw1...) Perhaps one thing we could do is put in a framework / base class for a "MinutesEstimator" - the default implementation would just do what we do now and look at minutes in previous matches, but people could easily add other derived classes that take e.g. a manual veto-list, or something that scrapes expected line-ups from match previews on the web...?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/alan-turing-institute/AIrsenal/issues/384#issuecomment-898448004, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABEM44RWQLGSZDDBVHMOMRDT4UK2ZANCNFSM5B32TN4Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .

richardthomasdev commented 3 years ago

While not directly solving the problem, what about an arg that lets the Teams Selected % be taken into account?

eg. airsenal_make_squad --budget 1000 --min_teams_perc 3

The above would automatically disqualify both Nat and Siggy from contention, and while not a direct solution might be a cool and less-obtrusive addition it it's own right?

Would love to contribute to this project and take a stab at this, if a direction can be agreed!

Tdarnell commented 3 years ago

I do like that approach too, and don't see that as any less AI driven than web scraping human entered information for predicted lineup's!

On Fri, 13 Aug 2021, 14:23 richardthomasdev, @.***> wrote:

While not directly solving the problem, what about an arg that lets the Teams Selected % be taken into account?

eg. airsenal_make_squad --budget 1000 --min_teams_perc 3

The above would automatically disqualify both Nat and Siggy from contention, and while not a direct solution might be a cool and less-obtrusive addition it it's own right?

Would love to contribute to this project and take a stab at this, if a direction can be agreed!

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/alan-turing-institute/AIrsenal/issues/384#issuecomment-898454856, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABEM44WUJIHIL4JPTVNAD6DT4UMEDANCNFSM5B32TN4Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .

nbarlowATI commented 3 years ago

@richardthomasdev that's a very good idea! Would be great if that could then be applied to the regular week-to-week optimization as well as the initial squad selection, and would be double-great if you could contribute this!! :) Feel free to create a new issue to cover that addition (or can just continue here if you prefer), and ask if there's anything we can help with in the implementation!

richardthomasdev commented 3 years ago

@nbarlowATI I've opened a separate issue (https://github.com/alan-turing-institute/AIrsenal/issues/400) in case any further discussion directly related to this happens here.