Reworked All in One - MAMBA

[TeemoTeejay]

I ended up reworking the All in One, (To rephrase that slightly my computer broke and I had to start it basically from scratch as my backedup files were from the week prior and since this was just a personal fun thing I didn't end up backing it up 1000 times like I usually do D: ) I wasn't sure if posting this as an addition to the previous thread made more sense, or doing it in this type of way, but since I essentially started the entire thing from scratch (somehow took so much longer than it originally did) I put it here instead.

I had realized a pitfall was that with it being so reliant on TD RAPM as a component that while the bias created from previous years at the end of a season would be fine, this metric might not work super well in-season or at least early-mid season

Therefore, I when recreating the metric I wanted to make the Prior better and have the metric retain predictability and accuracy with a higher Decay Rate. Overall, accuracy was essentially the same (Slightly better but close enough that it really wasn't anything of significance, different testing methods could make up the difference there) with a decay rate that was nearly double the original

Testing was a bit different and not nearly as complicated as previously with the replacement values just since the only thing that matters here is the relative accuracy of the metrics, but generally the same results were there, although this time before running the test I took out players not in EPM to make it a more fair comparison (Accuracy is lower on all the metrics tested than previously because I did less than before to accomadate for injured players or rookies, just since I was focused on forward prediction in this test)

I tested the Prior against BoxLebron, and the overall metric against EPM. BoxMamba was about a tossup on defense and significantly better on offense and overall (Offensively it outperformed the entire LEBRON metric actually), while Mamba was better on offense than EPM and a tossup on Defense compared to EPM (Being a bit better testing overall). AD also is no longer glaringly low, and Lebron generally looks better than before. I would say the Prior defensively is not better than EPM's defensive prior which isn't public (nor do i have access to it) just based on the results of the final metrics being so close on defense, although I didn't approach the prior solely trying to make it the most predictive possible.

With the prior seeming promising, I want to explore how a single year version of this does with a bit more emphasis on precision in the Defensive portion of the prior as well as I had gotten versions that tested better but just didn't like the results quite as much. I go more into what I did specifically here, much faster read than the original one.
https://www.teemohoop.com/mamba/mamba-reworked-updated

here is the reworked metric. Not super different overall but for the most part I like most of the changes compared to the original, although Luka is still a tad low (As I took out POE), his Box prior was very good in 2023, not as much in 2022. It's more from defense than it being low on his offense, 2023 at least. : https://timotaij.github.io/LepookTable/

Im pretty sure the github is public so if you want to see how it compared to the old one, just download it straight off the github page:

[Crow]

Default view of 2024 might be slightly more user friendly than 2017. And 2025 when you there.

[TeemoTeejay]

Default view of 2024 might be slightly more user friendly than 2017. And 2025 when you there.

Im probably gonna just present all of this stuff in Rshiny instead its way easier to manage that and way more user friendly

[TeemoTeejay]

Default view of 2024 might be slightly more user friendly than 2017. And 2025 when you there.

https://timothywij123.shinyapps.io/MAMBAData/

^Rshiny results

Will potentially add a "wins added" metric as well from this

[Crow]

That is straightforward, ready.