LombardoTron

I got a fantasy football draft coming up Monday Aug 26, 2024, and I need a way to rank players. So I will try out some models in SciKit-Learn that map year K's stats to year K+1's overall performance.

It's an IDP league, so I gotta pay attention to offense and defense. Which, look, this is all the same to me, I'm outsourcing all this to some random forest anyway.

Python environment

Here are the actual pip commands I have run in this project's virtual environment:

$ pip3 install -U scikit-learn

Results:

$ pip freeze
joblib==1.4.2
numpy==2.1.0
scikit-learn==1.5.1
scipy==1.14.1
threadpoolctl==3.5.0

Data

Much thanks to NFLVerse for these CSVs:

• https://github.com/nflverse/nflverse-data/releases/tag/player_stats
• https://github.com/nflverse/nflverse-data/releases/tag/weekly_rosters

I downloaded a few into a folder called data/ that I told git to ignore.

The column translations are available at:

https://nflreadr.nflverse.com/articles/dictionary_player_stats.html

When adding a new player_stats_def_season_20XX.csv file, manually go in and replace the column "games" with "def_games". Similarly, when adding a player_stats_kicking_season csv, replace "games" with "kck_games". This is a convention I am using to merge stats for the same player across offense, defense, and special teams, while still being able to validate that every distinct player is only tabulated once per season.

Rankings

See my use of this repo to rank players for our league draft in SEASON2024.md and SEASON2025.md.