*% Chance to Make the Hall of Fame - Like half of a model basically

*% Chance to Make the Hall of Fame - Like half of a model basically - Printable Version

+- [DEV] ISFL Forums (http://dev.sim-football.com/forums)
+-- Forum: Community (http://dev.sim-football.com/forums/forumdisplay.php?fid=5)
+--- Forum: Media (http://dev.sim-football.com/forums/forumdisplay.php?fid=37)
+---- Forum: Graded Articles (http://dev.sim-football.com/forums/forumdisplay.php?fid=38)
+---- Thread: *% Chance to Make the Hall of Fame - Like half of a model basically (/showthread.php?tid=30102)

Pages: 1 2 3 4 5

RE: % Chance to Make the Hall of Fame - Like half of a model basically - Nokazoa - 02-19-2021

I'm in this and I don't hate it

RE: % Chance to Make the Hall of Fame - Like half of a model basically - iStegosauruz - 02-20-2021

(02-19-2021, 08:43 PM)Thor Wrote: Something has to be wrong with TE - how could Heath be -3% with a TEOTY award, a top 10 season for catches, and several pro bowls where specified a lack of those is the issue, when Daniel George has none of those and is a 9% chance? I believe Heath has more stats in about every category than George as well so that has me perplexed haha, are ultimus' perhaps weighted too heavily given that a player can see significant individual success without them?

Like you kinda answered your own internal question though and this is something I addressed repeatedly when talking about the players who ended up in the negatives at every stage of the article. Like multiple times. Negative doesn't imply you have a trash chance to make the Hall of Fame, it implies that there's a glitch in the model I really haven't ironed out.

Lets look at the specific situation with Tight End. When the regression runs and tries to fit the input to the "line" of what makes a Hall of Fame Tight end it only has three people to compare to - Blaze, L'Alto, and DiMirio. They all played (mostly) in a time pre-S22 and an explosion of max earning, vertical threat Tight Ends that serve as true offensive weapons. They also played in a time with just generally less competition for awards and probowl spots. As a result, the model credits them with 12 (DiMirio), 10 (L'Alto), and 8 (Blaze) probowls apiece. Thats more than any other active Tight End has by a margin of, at minimum, 3. So modern Tight Ends are already playing at a discrepancy when the model takes the snapshot in time and tries to compare volume of probowls. Its conceivable that because you're headed into regression in a season you never crack L'Alto's 10 probowls. That DOES NOT mean you won't be a Hall of Famer when all is said and done, it means the model has a hard time comparing you because it has no data points from a modern era where by nature of there just being more competition you aren't likely to hit that peak. Thats the first knock - you start at an innate disadvantage in trying to make up ground. This manifests itself in multiple positions where we just don't have a ton of modern comps yet because the HOF has only been around since S23 and there are only like 50 players across all position groups enshrined in it. Tight Ends are inherently going to have low values until I either weight more heavily by era played in or until more modern Tight Ends eventually make the Hall.

Now the big question is why George sits at 9% chance to your negative value. The issue comes from how the model interprets statistics that qualify for you for a top-10 category. You literally address this in your post when you mention you have more stats than him in every category. You're right! And thats why the model has some issues. It takes your 7th place ranking for catches in a season and goes "Wow, this was an incredible season. But why hasn't he done it again? Why didn't he win more awards? Why doesn't he have an Ultimus?" Then it gets in its own head and spins its wheels and can't come up with a reason and so it tanks your value. It essentially fucks itself over. Even more simply put because two of the TEs in the Hall - Blaze and L'Alto - hardly show up at all on top-10 lists (and neither do on any top-10 season list) but have both won significantly more awards than you have it decides your trash because you've got technically an infinite amount more credit for top-10 seasons but less awards Like it can't rationalize the two issues.

Why doesn't it do this with George? Because with a lack of other inputs he doesn't confuse the model. He's like a 0% player besides one thing that ticks up his value. You technically have a resume much different than most of what it looks at with Tight Ends and it doesn't know how to interpret that so it tanks your eventual final metric because like its trying to tell me "this isn't the same as the others," etc.

Basically thats a lot of mumbo jumbo for like comparatively to other TEs in your era you've been incredible but like it can't compare you to them and so somehow tanks you into the negatives.

So yes, tight end has a problem. I know it does. I've repeatedly said as much. George essentially gets credit above you because he's almost a "null" player. He has one Ultimus which boosts him a decent bit since it gets him 1/3 to the average amount of Ultimus titles a HOF Tight End has currently. There is a big boost to that. He has nothing that the model can't wrap its head around and as such would tank his like "ranking/value" or however you want to put it. You do. Can I fix this? Probably and I'm working on de-linking some of the top-10 categories with awards for different positions and changing up how eras weight a bit.

A much briefer and less technical explanation of "the model fucked itself because its struggling to compare something thats entirely different than it has seen before" was readily available throughout the piece.
[Image: F2fzA6i.png]

But like this is a work in progress. Its not that Ultimus wins are weighted too highly its that like you just have a different resume than any TE before you so its struggling to compare you. This is a total mind fuck to talk about to be fair. Like why do good players get knocked? Cause its a new era and I tried to keep the weighting on eras fairly light for now. Can adjust it more down the road but like I don't want to go too far too early and have @zaynzk see his sorry player on the cusp of getting in.

RE: % Chance to Make the Hall of Fame - Like half of a model basically - iStegosauruz - 02-20-2021

(02-19-2021, 09:59 PM)Matty7478 Wrote:
(02-19-2021, 08:43 PM)Thor Wrote: Something has to be wrong with TE - how could Heath be -3% with a TEOTY award, a top 10 season for catches, and several pro bowls where specified a lack of those is the issue, when Daniel George has none of those and is a 9% chance? I believe Heath has more stats in about every category than George as well so that has me perplexed haha, are ultimus' perhaps weighted too heavily given that a player can see significant individual success without them?

Let’s be -% chance players together Thor, don’t leave me at the bottom like this

Making a second comment for this separate from the one for Thor because although its a similar issue its just cleaner to explain separately. So basically the model takes that "snapshot in time" of what players look like now and try to compare them to current guys in the HOF. In terms of how the like career and season top-10 lists are input for kickers you're a bit below the three HOF kickers in all three but still like fairly close. The issue comes in that comparatively it credits you for one award while the lowest it credits a HOF'er for is 5. So you're close in production but pretty far behind in awards (and probowls too for that matter). This is a new phenomenon that a lot of positions are finding especially post S22 - there is more competition for awards and probowl slots because the positions are so ridiculously saturated. We haven't had any players from this era enter the Hall yet, so there's nothing to compare y'all to that really resembles y'all.

This leads the model to have no clue how to interpret y'all. It can't compare your overall production and so it decides to instead tank you - you don't "fit" the "line" of what its looking for. This is fixable - I have to really heavily weight by eras (or wait until more modern players make the Hall and the "line" normalizes), and thats something thats quite complex and finicky. Until then we've got players who are having great careers - Bowie, Watts, McDairmid, Evans, etc. - who are "new" in terms of how their resumes look and as a result get smacked around a bit by the model because they don't fit with what its expecting to see. Thats why older players fit fairly well - think Saba in WRs, Gump in RBs. They accumulated resumes during (mostly) a different era of the league which allows them to fit the expectation the model has for the Hall. Once more players from this post S22 boom make it in with resumes that more fit this era the model will move towards the midpoint between the two and modern players will shoot up these charts.

RE: % Chance to Make the Hall of Fame - Like half of a model basically - Matty7478 - 02-20-2021

(02-19-2021, 10:20 PM)Thor Wrote:
(02-19-2021, 09:59 PM)Matty7478 Wrote:
(02-19-2021, 08:43 PM)Thor Wrote: Something has to be wrong with TE - how could Heath be -3% with a TEOTY award, a top 10 season for catches, and several pro bowls where specified a lack of those is the issue, when Daniel George has none of those and is a 9% chance? I believe Heath has more stats in about every category than George as well so that has me perplexed haha, are ultimus' perhaps weighted too heavily given that a player can see significant individual success without them?

Let’s be -% chance players together Thor, don’t leave me at the bottom like this

Sorry I can't be seen associating with a -7%er

wtf

RE: % Chance to Make the Hall of Fame - Like half of a model basically - OrbitingDeath - 02-20-2021

My one goal for this league and im at 0% Big Grin

RE: % Chance to Make the Hall of Fame - Like half of a model basically - Opera_Phantom - 02-20-2021

Cue's Pro Bowl as a OL will be a lock for HoF

RE: % Chance to Make the Hall of Fame - Like half of a model basically - CLG Rampage - 02-20-2021

(02-19-2021, 10:18 PM)moonlight Wrote: Imagine not having a 100% chance to make the hall of fame lol

truuuuuu

RE: % Chance to Make the Hall of Fame - Like half of a model basically - hihihi_62 - 02-20-2021

48% Chance?! Hell ya

RE: % Chance to Make the Hall of Fame - Like half of a model basically - Thor - 02-20-2021

(02-20-2021, 12:07 AM)iStegosauruz Wrote:
(02-19-2021, 08:43 PM)Thor Wrote: Something has to be wrong with TE - how could Heath be -3% with a TEOTY award, a top 10 season for catches, and several pro bowls where specified a lack of those is the issue, when Daniel George has none of those and is a 9% chance? I believe Heath has more stats in about every category than George as well so that has me perplexed haha, are ultimus' perhaps weighted too heavily given that a player can see significant individual success without them?

lots of words and explainy stuff

Thanks man, I appreciate the clarification - I wasn't taking shots I just couldn't understand where the difference came in, but that makes a lot more sense now. I look forward to seeing future updates to your model.

RE: % Chance to Make the Hall of Fame - Like half of a model basically - iStegosauruz - 02-20-2021

(02-20-2021, 07:17 PM)Thor Wrote:
(02-20-2021, 12:07 AM)iStegosauruz Wrote:
(02-19-2021, 08:43 PM)Thor Wrote: Something has to be wrong with TE - how could Heath be -3% with a TEOTY award, a top 10 season for catches, and several pro bowls where specified a lack of those is the issue, when Daniel George has none of those and is a 9% chance? I believe Heath has more stats in about every category than George as well so that has me perplexed haha, are ultimus' perhaps weighted too heavily given that a player can see significant individual success without them?

lots of words and explainy stuff

Thanks man, I appreciate the clarification - I wasn't taking shots I just couldn't understand where the difference came in, but that makes a lot more sense now. I look forward to seeing future updates to your model.

Yeah for sure! I really do appreciate people poking at things trying to figure it out because it gives me some direction of what to look into. If I came off as snarky it definitely wasn't my intention, I just wrote that 5AM and being completely coherent and pleasant at 5AM isn't a strongsuit of mine haha.

This is an absolute headache of a puzzle to figure out and I love it. Hopefully gonna keep refining it, just gotta figure out some of the math stuff that I don't understand with it.