Nathan Walker
@bbstats.bsky.social
basketball analytics person
data science @ NFL
data science @ NFL
omg (to the latter)
November 13, 2025 at 4:18 PM
omg (to the latter)
Por que no los dos
October 24, 2025 at 2:02 AM
Por que no los dos
(the even shorter version of the story is - on data with a large sample size...like sports...logistic regression almost always has worse calibration plots in my experience)
September 9, 2025 at 2:37 PM
(the even shorter version of the story is - on data with a large sample size...like sports...logistic regression almost always has worse calibration plots in my experience)
here's some non-cherry-picked calibration plots using random forest vs logistic regression. no clear winner but I think RF is slightly better.
colab.research.google.com/drive/1thJ3S...
colab.research.google.com/drive/1thJ3S...
Google Colab
colab.research.google.com
September 9, 2025 at 2:36 PM
here's some non-cherry-picked calibration plots using random forest vs logistic regression. no clear winner but I think RF is slightly better.
colab.research.google.com/drive/1thJ3S...
colab.research.google.com/drive/1thJ3S...
more of the picture they are painting - Logistic Regression's "probability" is usually worse-calibrated than other models.
often times these sorts of "statistical reasons" for using one algo vs another only focus on one aspect of what could make probability calibration better or worse.
often times these sorts of "statistical reasons" for using one algo vs another only focus on one aspect of what could make probability calibration better or worse.
September 9, 2025 at 2:34 PM
more of the picture they are painting - Logistic Regression's "probability" is usually worse-calibrated than other models.
often times these sorts of "statistical reasons" for using one algo vs another only focus on one aspect of what could make probability calibration better or worse.
often times these sorts of "statistical reasons" for using one algo vs another only focus on one aspect of what could make probability calibration better or worse.
Rebounds are pretty height related, threes are a gamble. Overall makes sense though.
August 19, 2025 at 6:31 PM
Rebounds are pretty height related, threes are a gamble. Overall makes sense though.
The data here described in the features is the data they're requesting access to. Nothingburger.
July 17, 2025 at 12:23 PM
The data here described in the features is the data they're requesting access to. Nothingburger.
No involvement! Ask @lucashaupt.bsky.social perhaps?
July 16, 2025 at 10:48 PM
No involvement! Ask @lucashaupt.bsky.social perhaps?