Plug For The Leagues With The "Similar Players" Multiplayer Feature

[Ben E Lou]

I'm not posting this to toot my own horn, but as a shameless recruiting tool for leagues I run. :D It was mentioned in the Combine Benchmarks thread, but I thought I'd break it out here and highlight the leagues that have it or will have it very soon.

The output is mostly self-explanatory: the code looks at a database table containing the league's draft history and determines the five players in league history with combines most similar to the player just drafted. The code is added to Conscriptor as part of the function that fires when a pick is made, so it automatically outputs this information to the thread in the league forum where the draft is being discussed.

Here's how it looks in action:

2045 Amateur Draft Thread - Page 5 (http://www.younglifenorthdekalb.com/ccfl/forum/showthread.php?t=6537&p=33174&viewfull=1#post33174)

If you want to play in a league that has this feature, here are the ones that either are already using it or will have it starting at the next draft:

CCFL: (six gameday sims per week, low injury setting, super-active and competitive owner base) CCFL Forum (http://www.younglifenorthdekalb.com/ccfl/forum/forum.php)
FOOL: (super-fast league that focuses on offseason stuff, sims entire regular season in one shot) FOF-FOOL Forum (http://www.younglifenorthdekalb.com/fool/forum/forum.php)
GML: (GM-only league. All teams use only "Recommend" game plans.) General Manager League Forum (http://www.fof-gml.com/forum/forum.php)
IHOF: (Standard 3 gameday sims per week league): IHOF Forums - Powered by vBulletin (http://www.fof-ihof.com/forums/index.php)
WOOF: (Standard 3 gameday sims per week league) World Order Of Football Forum (http://www.fof-woof2.com/forum/forum.php)

[WheelsVT]

Hey Ben, awesome feature I must say again. Have you thought about showing this as a draft preview tool for owners to use in preparing for the draft?

I'm also curious if I could just build a HUGE database of players from a generated single player history with ratings and combines and then pull from that as a way to predict statistically how prospective players will likely turn out with standard deviations for individual bars and everything... Perhaps you better GMs already do this.

[gstelmack]

Hey Ben, awesome feature I must say again. Have you thought about showing this as a draft preview tool for owners to use in preparing for the draft?

I've certainly been thinking about how to use something like this for Draft Analyzer, I'd love to have a better prediction algorithm. The concept could be extended to include bars and the like, which might be a must for leagues like WOOF with low combine accuracy.

I'm also curious if I could just build a HUGE database of players from a generated single player history with ratings and combines and then pull from that as a way to predict statistically how prospective players will likely turn out with standard deviations for individual bars and everything... Perhaps you better GMs already do this.

If the combine accuracy matches, something like this is certainly possible. It's one of my motivations behind writing PlayerTracker - I'll be able to build up this database with FOOL as it goes along, and/or build a database from a single player career with a similar combine setting to allow better predictions in DraftAnalyzer.

However, there is often enough randomness to make such predictions problematic, I'm very curious to see how this utility's matches work as the database grows.

[Ben E Lou]

Hey Ben, awesome feature I must say again. Have you thought about showing this as a draft preview tool for owners to use in preparing for the draft?Because of the time the search function takes, it's not really feasible to do it for the whole class as a dynamic php page. I set up a page to report just the 32 players taken in the first round in one league, and that took 2ish minutes to load. At that pace, the entire draft class would take over an hour to load. At that pace, the I suppose that I could run it once and post it, but with 830ish players in a draft class, I'd think it would break the league message board.

I'm also curious if I could just build a HUGE database of players from a generated single player history with ratings and combines and then pull from that as a way to predict statistically how prospective players will likely turn out with standard deviations for individual bars and everything... Perhaps you better GMs already do this.You *could* to some degree, but combines don't really work like that. They can vary pretty widely within the ranges. It's a fun feature, but I wouldn't use it to try to predict too much. For example, check out the five matches for this DT:

2045 Amateur Draft Thread - Page 3 (http://www.younglifenorthdekalb.com/ccfl/forum/showthread.php?t=6537&p=33047&viewfull=1#post33047) *Extremely* close combines, but ratings varied from 30 (useless) to 50 (likely starter).

Or these QBs:

2045 Amateur Draft Thread - Page 5 (http://www.younglifenorthdekalb.com/ccfl/forum/showthread.php?t=6537&p=33149&viewfull=1#post33149)

[WheelsVT]

Both of you guys do amazing work.

Sounds like the making of a stand-alone tool in addition to the faster online feature. If you could run through say the top 150 players in the draft, let it take X minutes to process, and if the database is large enough perhaps the range of overall ratings begins to fit a distribution or maybe the predicted bars are a more statistical fit.

Like if you make a player-database an add-on option to draft analyzer where you click on "similar player search" and you get a pop-up with ranged bars backed with past player data (not to weigh you down with more ideas gstelmack).

We'd start passing around databases at different combine accuracy values... or would it be better to pull from a database built on a 100 combine correlation setting so that the tool predicts statistically without randomness?