Read on Twitter
This is now a #BigDataBowl submission thread.
Here's a really neat idea with clean execution https://twitter.com/CharlieGel/status/1347208988367200256
Here's a really neat idea with clean execution https://twitter.com/CharlieGel/status/1347208988367200256
Oooh here's another really cool #BigDataBowl submission.
@mfbanalytics builds a coverage classification model and then looks at the distribution of EPA by coverage.
Cover 0, for example, results in either a very good or very bad play for the defense
https://www.kaggle.com/tejseth/random-forest-for-pass-coverages
@mfbanalytics builds a coverage classification model and then looks at the distribution of EPA by coverage.
Cover 0, for example, results in either a very good or very bad play for the defense
https://www.kaggle.com/tejseth/random-forest-for-pass-coverages
This is a nice #BigDataBowl submission that estimates target probability and uses that to get at individual DBs' deterrence effects
https://twitter.com/nnstats/status/1347271449187393538
https://twitter.com/nnstats/status/1347271449187393538
(there are a million more submissions that look interesting but need to take a break. to be continued)
It's a #BigDataBowl all-star team!
As expected this is extremely well done
https://twitter.com/msubbaiah1/status/1347306980273233921
As expected this is extremely well done
https://twitter.com/msubbaiah1/status/1347306980273233921
(sidebar: enjoying how a sanity check for all of these defender models is "do we find that Stephon Gilmore is in the top 5?")
Okay this one is wild #BigDataBowl
They make a package ( https://hjmbigdatabowl.github.io/bdb2021/reference/) and a shiny app ( https://bdb-2021.herokuapp.com/ )!
https://www.kaggle.com/hughmccreery/hjmo-big-data-bowl-submission
(enjoying how many of these have open-sourced the code, a welcome change after last year)
They make a package ( https://hjmbigdatabowl.github.io/bdb2021/reference/) and a shiny app ( https://bdb-2021.herokuapp.com/ )!
https://www.kaggle.com/hughmccreery/hjmo-big-data-bowl-submission
(enjoying how many of these have open-sourced the code, a welcome change after last year)
!! Ron did a submission! See thread here, really neat stuff https://twitter.com/Stat_Ron/status/1347284805319335947
I think the #BigDataBowl submissions that incorporate estimating target probability at some point are my favorite. This one is extremely well done and has very beautiful viz/tables.
And again we see Stephon Gilmore way up there https://twitter.com/asmae_toumi/status/1346599886234546177
And again we see Stephon Gilmore way up there https://twitter.com/asmae_toumi/status/1346599886234546177
This #BigDataBowl submission has something unique: estimating target suppressing but then connecting it to overall team defensive outcomes and asking how valuable it even is https://twitter.com/EthanCDouglas/status/1347337980671795206
Submission was linked in the top post in the thread but here's the official submission thread, including the video which is incredibly well done https://twitter.com/SuvanshSanjeev/status/1347337446204182528
Ooh this is a really neat #BigDataBowl submission.
Quantify wasted movement by LBs biting on play action --> relate that to expected run % from @nflfastR --> show that more wasted movement results in higher CPOE (bad for defense) https://www.kaggle.com/timlivingston/nfl-data-bowl-submission-taking-the-bait https://twitter.com/J_Blas24/status/1347346307573276680
Quantify wasted movement by LBs biting on play action --> relate that to expected run % from @nflfastR --> show that more wasted movement results in higher CPOE (bad for defense) https://www.kaggle.com/timlivingston/nfl-data-bowl-submission-taking-the-bait https://twitter.com/J_Blas24/status/1347346307573276680
This notebook just crushes everyone else in terms of accurately classifying coverages, so read it if you're interested in doing that https://www.kaggle.com/powerthinking/how-nfl-pass-defenses-can-learn-from-poker-players
(a pause for my personal commentary: in all 3 Big Data Bowls, we've seen that to get the most out of player tracking data for some problems, it needs to be treated like an image recognition problem with some sort of CNN-type approach)
