Drafting up an all in one

[TeemoTeejay]

Hello Everyone! My first post here but I've lurked a bit before, honestly was super helpful for this little project I started doing this week!

I made an All in one metric, I dont have a name for it yet but its Time Decayed RAPM up to 2 seasons prior (with the decay starting at the first day of the season chosen, if that makes sense, so the current season is weighted at 1.000 all the way), and the box prior has something i made called Synergy POE which is just points over expectation considering play type distribution, and some tracking data like rim points saved which was just FGA LT06 * 2 * Effected DFG%.

Its very much it its early phase, very much in its proof of concept phase so far it took me a day or two to get the data, a day or two to run it, and a day or two to do this write up, so haven't explored playing with different sigmas or decay rates, and the box score prior, especially on defense, still is very much in phase 1, but I just wanted to kind of have a number out there to see if this proof of concept was good, but overall I was happy with the results. I did retrodiction testing and it beat out LEBRON and EPM by a decent amount, I dont know what to call it though

The Luck adjusted RAPM stuff seems a bit weird, I did it for free throws, but on threes i wasnt to sure, I did a really small one there but I've heard it doesn't actually help and it was such a small adjustment i might just take it out. Free throws seem fair though. The OREB and TOV stuff feels like a huge can of worms, but yeah.

i did a write up here that I kinda just posted and alot of it was kinda aimed towards non data people at the start, and everyone here is a data wiz so not super relevant here I think, but kinda went more into my thoughts of the data and some potentially caveats to this approach (Like if the previous year creates too much bias) and how those looked in practice, but generally I think its a solid proof of concept, I wanted to get some opinions here on how it looks/if this seems like a decent thing, keeping in mind its very much in its alpha phase since its basically a first run.

NOTE: Long rambling version, but with some good stuff on ridge and my thoughts on all in ones in general
https://www.teemohoop.com/mamba-or-lepo ... 0One-mm8gk

Note: More editted and professional version:
https://www.teemohoop.com/mamba/Blog%20 ... m8gk-cy9wh


Would love to hear thoughts or any like advice on this stuff, I'm pretty new to doing this kind of stuff! On the SPM especially, I've heard interaction terms and non linear models can suck and was wondering about that, since it did feel that way when I tried it just to see.

[Crow]

Went to check link but I get a privacy warning. I eventually decided to proceed but others may not. Anything you can do to correct privacy certificate and reassure?

Article was too long for me to properly consume and react to at this moment, but hopefully I'll try harder later.

Pete Davidson and other fake player mentions near start were a mild distraction for me.


Any chance you'd try a second summary explanation with just the absolute essentials? The full article is good, especially for a few; but a shorter summary would be a good companion aimed at broader audience, perhaps as a starter to the full one.


Big picture, good wishes with continuing the ambitious / rigorous project and sharing progress.


I will spot check your metric's values against 1 or more others but any interest in a table comparing top 20s or full league with top competing metrics?

First case to catch my eye, your metric has Tatum 15, while Darko puts him at 2.

Wanted to compare Wembanyama but not finding his record in table or by player search.

Any chance you'd add a rank column?

Anunoby, both metrics put in low 30s.


"Data wiz" is a term that depends on standard to the author. Also depends on if you are talking about technical production or intelligent use. Not everyone who reads here are one or both. And both are important.


Looking for better WNBA 1metrics but won't push you beyond this mention on that.

[TeemoTeejay]

Went to check link but I get a privacy warning. I eventually decided to proceed but others may not. Anything you can do to correct privacy certificate and reassure?

Article was too long for me to properly consume and react to at this moment, but hopefully I'll try harder later.

Pete Davidson and other fake player mentions near start were a mild distraction for me.


Any chance you'd try a second summary explanation with just the absolute essentials? The full article is good, especially for a few; but a shorter summary would be a good companion aimed at broader audience, perhaps as a starter to the full one.


Big picture, good wishes with continuing the ambitious / rigorous project and sharing progress.


I will spot check your metric's values against 1 or more others but any interest in a table comparing top 20s or full league with top competing metrics?

First case to catch my eye, your metric has Tatum 15, while Darko puts him at 2.

Wanted to compare Wembanyama but not finding his record in table or by player search.

Any chance you'd add a rank column?

Anunoby, both metrics put in low 30s.


"Data wiz" is a term that depends on standard to the author. Also depends on if you are talking about technical production or intelligent use. Not everyone who reads here are one or both. And both are important.


Looking for better WNBA 1metrics but won't push you beyond this mention on that.

Appreciate the advice! I'll do a rewrite but I updated the link the SSL stuff should be better now.
I made a mistake with the 2024 list of players, so its missing rookies, ill edit that tonight thats a big oopsie by me.
I'm a bit busy right now so not sure about making a table with the ranks and everything, but I'll rewrite and make it a bit more clear! Was a bit suprised with tatum, DPM uses playoff data too though right?

[Crow]

I'd guess so, but not on any immediate proof in description.

Is it same weight as any game or more? I have probably given up asking questions there.

[Crow]

If you go with MAMBA name (reasonable for marketing), is it based on initials or just a name?

Is Nikola or Jokic workable or a contender label?

If not MAMBA, what is your second choice?

[Crow]

I am going to try to go line by line thru article and make whatever comments come to mind.

The effort to explain RAPM is noble but a few sentences might benefit from a bit of tuning.

The noise vs. bias section is good for an audience segment not total aware previously but interested.

Metrics will vary. Anything you want to do to try to explain the differences or uses the differences to create a meta-metric... that hopefully is better?

Are you solely interested in explanation or also interested in prediction?

Your way of doing time decay of data is fine by me but might not handle rookies as sensitively.

Time Decay RAPM seems like the right standard over 1 year or multi-year without time decay.

I am not certain about my opinion of luck adjustments, so light ones may be a good compromise. Performance comparison at different levels of adjustment would seem wise.

"... only thing I changed was for rookies I gave them a -1.5 instead of replacement values (-2.5)..."

I go in probably disagreeing but what was your reason for this heightened starting point against historical data?

Strong initial results compared to other metrics. Congrats. I'll leave any detailed discussion of that potentially to others.

"It will generally struggle to Identify really good players on teams that can be elite in the regular season without them."

Would like to here more discussion of that.

Any thoughts about "Estimated errors" and how to calculate and whether to present and how to describe?

IF you used your metric as a starting point and shifted to presenting tiers, how many tiers would you use?


All things considered you might have done in days what might have taken weeks or months for some predecessors.

Good wishes in future endeavors and share what you want / can.


P.S. On second take, writeup wasn't as daunting.

(Maybe more white space between sections?)


P.P.S. Anything you want / can say about draft modeling?

[TeemoTeejay]

I am going to try to go line by line thru article and make whatever comments come to mind.

The effort to explain RAPM is noble but a few sentences might benefit from a bit of tuning.

The noise vs. bias section is good for an audience segment not total aware previously but interested.

Metrics will vary. Anything you want to do to try to explain the differences or uses the differences to create a meta-metric... that hopefully is better?

Are you solely interested in explanation or also interested in prediction?

Your way of doing time decay of data is fine by me but might not handle rookies as sensitively.

Time Decay RAPM seems like the right standard over 1 year or multi-year without time decay.

I am not certain about my opinion of luck adjustments, so light ones may be a good compromise. Performance comparison at different levels of adjustment would seem wise.

"... only thing I changed was for rookies I gave them a -1.5 instead of replacement values (-2.5)..."

I go in probably disagreeing but what was your reason for this heightened starting point against historical data?

Strong initial results compared to other metrics. Congrats. I'll leave any detailed discussion of that potentially to others.

"It will generally struggle to Identify really good players on teams that can be elite in the regular season without them."

Would like to here more discussion of that.

Any thoughts about "Estimated errors" and how to calculate and whether to present and how to describe?

IF you used your metric as a starting point and shifted to presenting tiers, how many tiers would you use?


All things considered you might have done in days what might have taken weeks or months for some predecessors.

Good wishes in future endeavors and share what you want / can.


P.S. On second take, writeup wasn't as daunting.

(Maybe more white space between sections?)


P.P.S. Anything you want / can say about draft modeling?

Here is the updated link! https://www.teemohoop.com/mamba/Blog%20 ... m8gk-cy9wh
A bit less of a daunting read

On the good team that can handle without them part, I just kind of thought about it in the sense that as all in ones inherently still are impact measurements even with an attempt to make it more like "goodness" with the box score prior (whether or not thats the goal it still does that practically imo), RAPM and as a whole impact itself also depends on how much you are relied on, which isn't always = to how good you are. Things like fit and scarcity of your strengths within that team and stuff, not a strict rule and situational, of course.

I don't really have much in terms of draft modeling, just did Gradient boosting with a ton of predictors that I whittled down and some interaction stuff, in hindsight my portion (its a team) was decent, but I think I would do better now with better ability on that end.

No real reasoning with the rookies in hindsight I was just really overthinking the Wemby and Chet thing, but that was just for testing anyways and I was only interested in the relative results

"Metrics will vary. Anything you want to do to try to explain the differences or uses the differences to create a meta-metric... that hopefully is better?"

Not sure what you mean by this?

Thank you so much for reading it! Really appreciate it, not sure how I can push this haha. Admittedly alot of it is for the resume but I will try to try my hand at factor RAPM and stuff, a bit busy over the next few days and today but I should have some free time to do it. I've always been more of a Film Tracking/Xs and Os guy.

Generally speaking the exact placements at the top vary alot, curry looks like the undisputed king right now but less weight on the priors had bron boost up to 1 there, but everyone else just got funky lol. Kinda touched on it with the bias thing but its just an unfortunate thing there, names have to somewhat make sense in a public sphere for it to be reasonably pushed, although the general accuracy was better with tighter sigmas (I haven't really gone deeper into that either), I would assume with a guy like Bron its still underselling him (although higher on more recent years than like, LEBRON, ironically)

[TeemoTeejay]

If you go with MAMBA name (reasonable for marketing), is it based on initials or just a name?

Is Nikola or Jokic workable or a contender label?

If not MAMBA, what is your second choice?

Originally it was going to be Lepookie (Lol) but I feel that would be far too unserious, but right now I kind of like MAMBA Just because I love me some Kobe, gotta figure out some initials XD

[Crow]

Metric of APM Model w/ Boxscore Adjustments?

[Crow]

Explain the differences between metrics with sensitivity analysis of small changes in individual discrete stats or factors?

Test and find if there is a current / stable best meta-metric? The purpose of metrics. The search for which doesn't have to stop at the end of individual metrics. If you can do better combining metrics (whatever way), combine.

[TeemoTeejay]

Went to check link but I get a privacy warning. I eventually decided to proceed but others may not. Anything you can do to correct privacy certificate and reassure?

Article was too long for me to properly consume and react to at this moment, but hopefully I'll try harder later.

Pete Davidson and other fake player mentions near start were a mild distraction for me.


Any chance you'd try a second summary explanation with just the absolute essentials? The full article is good, especially for a few; but a shorter summary would be a good companion aimed at broader audience, perhaps as a starter to the full one.


Big picture, good wishes with continuing the ambitious / rigorous project and sharing progress.


I will spot check your metric's values against 1 or more others but any interest in a table comparing top 20s or full league with top competing metrics?

First case to catch my eye, your metric has Tatum 15, while Darko puts him at 2.

Wanted to compare Wembanyama but not finding his record in table or by player search.

Any chance you'd add a rank column?

Anunoby, both metrics put in low 30s.


"Data wiz" is a term that depends on standard to the author. Also depends on if you are talking about technical production or intelligent use. Not everyone who reads here are one or both. And both are important.


Looking for better WNBA 1metrics but won't push you beyond this mention on that.

Oh, responding to this part, The Rookies should now be up there. I had thought there was a huge issue with the level of team defense skewing a guy like Wemby's defense down, but he's third on defense behind Draymond and Gobert.

[TeemoTeejay]

Explain the differences between metrics with sensitivity analysis of small changes in individual discrete stats or factors?

Test and find if there is a current / stable best meta-metric? The purpose of metrics. The search for which doesn't have to stop at the end of individual metrics. If you can do better combining metrics (whatever way), combine.

Random question related to this, I know there’s a preseason projection contest, where do you guys get minutes? Thought it might be fun to try that if I speedrun developing the SPM A little bit

[Crow]

Sensitivity analysis to me, here would be take a players stats and add 1 to some stat and see how much change occurs in each metrics overall impact estimate and compare the effective values. If this were done, a blended metric would be easier to build and run.


Minutes, typically folks say they use Kevin Pelton's if / when released or fantasy sites or do it themselves, freestyle judgment or weighted from past 3 years. Not everyone goes to minutes and weighting player impacts by them.

[TeemoTeejay]

Sensitivity analysis to me, here would be take a players stats and add 1 to some stat and see how much change occurs in each metrics overall impact estimate and compare the effective values. If this were done, a blended metric would be easier to build and run.


Minutes, typically folks say they use Kevin Pelton's if / when released or fantasy sites or do it themselves, freestyle judgment or weighted from past 3 years. Not everyone goes to minutes and weighting player impacts by them.

Do you know where to find Kevin Pelton's minutes? I want to edit the SPM and then compare the numbers to the others over the next few days

I think I want to see if it can beat vegas with projections too

[Crow]

He probably hasn't published his article based on them yet.


Occasionally in past, he would post them here.

Could try contacting him at his bluesky feed (or by private message here).


Other options:

Rotoevil.com

https://www.rotowire.com/basketball/projections.php

https://fantasy.espn.com/basketball/players/projections

https://hashtagbasketball.com/fantasy-b ... rojections