FSM Ratings

[schnee]

The FSM needs your help.

The current “Greatest Farktographer Evar/Recent” reports merely count the number of votes and the one with the most is the greatest (evar = all time, recent = last 10 contests).

I find that dissatisfying in that it does not score for place-finishing. I also don’t think it weights participation enough. I’d like to tweak the ratings, but I’m not sure how. Any ideas? Maybe something from sports statistics (I know precious little about those). General guidelines are useful (like “I think we should emphasize Farktographer names that mean ‘snow’ in some Teutonic language”).

As a prototype, I’m assigning 32 “points” for each first place finish, 16 for each 2nd, 8 for third and one down to 1 point for any finish below 5th. Let’s call the sum of those points ‘F’.

I then figure out how many contests someone has entered and divide that by the number of contests that they could have entered (based on their Boobies – I don’t ‘punish’ people for not knowing about farktography). For example, if I my Boobies was 15 contests ago, but I only entered 7 contests since then, I calculate 7/15 as the ‘at-bat’ ratio (= ‘R’).

Then I multiply the total number of votes (= ‘V’) by ‘R’ and add that to F. This way, the absolute value of votes tends to become less important than a place-finishing, unless the farktographer participates a lot and then those factors can balance out.

F + (R * V) = FSM Rating

Ideas? Maybe work a natural logarithm in there or something?

(little of this is on the FSM yet, but its coming)

[Screaming In Digital]

E=mc2

[Elsinore]

Oi, math is definitely not my strong suit, so I’ll have to think through how that might work. Are you wanting to do this in addition to the current Greatest Evar stats? Maybe rename what’s currently the Greatest Evar to Highest Vote Getters or something? Then the Greatest Evar could be the product of the place finishing formula?

One wrinkle that might occur is when the final finishing places change several days or weeks after the contest ends. In that case, the Farktography page will list one person as a winner while the FSM will list someone else. I noticed that happens with the Written Word, where at the time of the contest, Encyclopedia Galactica and I tied for first, but after the contest was called, he got a few more votes.

To this point, you’ve been counting votes that come in for a period of a few weeks after the contest, which then counts towards the various totals and averages, which makes sense. I guess the question comes down to whether it’s more “fair” or makes more sense to assign places based on the dynamic data the FSM has, or whether the places should be assigned when the contest is called. The answer might depend on which statistic you’re running. I guess if you really wanted to go all out you could run two different statistics: one that would show votes and place finishing when the contest is called and another that factors in late votes for a period of time and assigns places based on that extra data. Or even a combined statistic that shows an entry’s place at the end of the contest and any change after the contest was called.

Dunno…think I’m babbling now lol

[Marley]

schnee, I’m going to have to chew on this before I come up with anything useful, but right off the bat I can see I’m going to take a hit. I had entries in “Flower Power,” sometime around June of last year, and then I was gone for 8 or 9 months while my life fell apart. I’ve been submitting fairly regularly over the past four months or so, but if I understand you correctly, that gap will count pretty harshly against me.

I’m not sure why missing a contest would be a bad thing. For example, for “Shadows”, I had a few pictures that met the criteria, but I didn’t think any of them were worth looking at. So I skipped that one. By this measure, it would be better to submit bad pictures and get no votes than to just spare everyone my rejects.

Just my preliminary $.02.

[schnee]

Lots of things:

First, thanks for the feedback. Keep it coming. Don’t be shy. I can probably handle the math; I need ideas.

The ratings will take the place of the Greatest Evar / Greatest Recently reports. They may or may not replace them, but keep reading.

I want the FSM to be a value-add to Fark first, Farktography.net 2nd. That means that the FSM will strive to match the results on Fark. Since Fark allows for votes after Farktography.net “calls” the contest (thanks for doing that, veruca!), we have at least two versions of the truth. Probably three since “striving to match” doesn’t mean “exactly matching”. The FSM stops picking up late votes after 4 weeks; I’m not sure when Fark “archives” a thread – there’s probably a window. Personally, I really like the Farktography.net placings – they give a sense of urgency to the contest (come on Secret Voting Block! Get those votes in!).

I was talking to some co-workers about this. One of the guys was a moto-cross rider and he told me how those scores work. In his league, place-finishing was all that mattered – points were scored according to 1st, 2nd, 3rd, &c. In other leagues, points were also granted based on “fastest lap”, “winning pole position”, and other factors. In his league, not finishing really killed chances at the championship. In the other leagues, not so much. But in the other leagues, riders tended to be more conservative since they could pick up points for other things. Not sure how this applies to Farktography, but it is interesting.

Marley, I hear what you are saying. You’ve already taken a hit by missing those months, why heap insult on injury by using my R ratio. Very good point. “Participation” is built-in to place-scoring and voting. I’ll probably nix the R ratio. Keep in mind, however, that I’m going to do an “Evar” and a “Recent” rating: “Recent” only looks at the past 10 contests. My “fascination” on participation is based on the “Greatest Averages Evar” report – there are some amazing averages from people who don’t participate all that much. I introduced the artificial threshold of 10 contests to keep that report distinct from the one-hit-wonder report, but 10 is pretty arbitrary – I have a vague feeling of uneasiness.

Elsinore, your ideas of multiple scores contains echoes of the conversation that I had with co-workers. One guy had the idea of something equivalent to the Bowl Championship Series in US College Football: multiple rankings that combine into some overall rating (the FCS, if you will). The software engineer in me is very intrigued (oooohhh, pluggable scoring engines – shiny!), but I’m a bit hesitant to really formalize that. 1) It sounds like a lot of work and, well, that’s about it really. Still, I live in the US, home of the BCS and with fresh memories of the 2000 Presidential Election. I love that sort of conflict (seriously, I live in Austin, TX but I’m a USC graduate (which if you don’t follow college football, doesn’t mean much)). Maybe I can come up with something – total votes, place finishing, best averages – maybe they all combine into a rating that we can argue about forever (‘we’ the community, not you and I – we’re cool).

[staplermofo]

Maybe you could weight them based on the percentage of the votes they get in relation to the competition and the size of the competition.

Here’s a practical example:
Let’s say 3 people enter a contest. staplermofo got first place with 150 votes and 2nd and 3rd got 15 and 10 votes respectively. staplermofo would be weighted 3 (# of contestants +1 -position) x10, the untalented schmuck that paled in his glory and came in 2nd would get 2 x1.5, and the loser would get 1 x 1.
That way it would factor in the difference of an unholy whomping, and the size of the competition. It’s horribly crude and ugly, but you see what I’m getting at.

I’d also like know the mean, median and the standard deviation for each voting, and my average deviation from them expressed in both raw and standard deviation form.
I’d also like you to quit your job, shut out your family and work on this full time for no compensation what-so-ever.

I’m not crazy about anything that might encourage people not to participate for fear of ruining their records, though I don’t really know how to avoid it.

[Author]

Posts