baseball-analytics

Goal: Predict matchup outcome for MLB games from scratch

Model: Neural Network with Pytorch

Features

• Batter 365 days moving average (long-term)
  • Rates per PA for 1B, 2B, 3B, HR, BB, SO, DP, FO, HBP, SF, SH [done]
  • Average wOBA [done]
  • Max exit velocity, Average exit velocity [done]
  • Number of plate appearances [done]
  • Platoon wOBA, Number of platoon plate appearances [done]
• Pitcher 365 days moving average (long-term)
  • Rates per PA for 1B, 2B, 3B, HR, BB, SO, DP, FO, HBP, SF, SH [done]
  • Average wOBA [done]
  • Max exit velocity, Average exit velocity [done]
  • Number of plate appearances [done]
  • Platoon wOBA, Number of platoon plate appearances [done]
• Batter 30 days moving average (short-term)
  • Rates per PA for 1B, 2B, HR, BB, SO [done]
  • Number of plate appearances [done]
  • Average wOBA [done]
• Pitcher 30 days moving average (short-term)
  • Rates per PA for 1B, 2B, HR, BB, SO [done]
  • Number of plate appearances [done]
  • Average wOBA [done]
• Batter and Pitcher head-to-head all-time
  • Rates per PA for 1B, 2B, HR, BB, SO [done]
  • Average wOBA [done]
  • Number of appearances [done]
• Ballpark
  • Park factors for 1B, 2B, 3B, HR, BB [done]
• Game state
  • Outs [done]
  • Innings [done]
  • Net Score [done]
  • 1B Occupied [done]
  • 2B Occupied [done]
  • 3B Occupied [done]
  • top/bottom of inning [done]
  • days since start of season [done]
  • temperature at game start time [not available yet]
• Batter fielding position
  • batter's 9 main positions + DH [TODO]

TODO:

• Check Team abbreviations for every year [done]
• Add logging
• Time the transformers (preparation for below) [done]
• Scale the transformers by parallelizing jobs
  • concurrent futures [done]
  • Spark pipeline on databrick
    • setup Spark locally
    • Build Spark pipeline
• Offline model and feature selection in notebook
• write script to scale to 20 years worth of data [done]
• translate the pipeline running code from notebook to prod [done]