Advertisement

More Related Content

Recently uploaded(20)

Advertisement

Final presentation

  1. The Goal Predict an individuals NHL player's contribution to his team over the life of his contract
  2. The Data • Private stats site (stats.hockeyanalysis.com) • Rows correspond to individual player's stats by season • Data spread across multiple tables – Joined tables using Pandas with name as key – Be careful with excel…
  3. Y’s (EN Goals + EN Assists) / TOI …for the following season
  4. X’s • Basic Feature Selection/Manipulation – Removed redundant/derived – Removed noisey nonsense – Normalized for TOI across seasons
  5. Features • • • • • • Goals, Assists, First Assists iFenwick tFenwick for, tFenwick against lmFenwick for, lmFenwick against, HARO/HARD/HART QoC A few others
  6. More about the data • Advanced stats only go back to 2007 • Aggregated prior 3 seasons of data to predict the following season • Players show up as multiple rows • Y’s in certain row’s become x’s in other rows
  7. Models • • • • • RF  .326 (be sure to set oob_score = True) ElasticNetCV  .359 ElasticNetCV w/PCA  .364 RF w/PCA  .285 SVR(xVal)  .314 (could be improved if I normalized featureset) • SVR(xVal) w/PCA  .322 • GBR w/PCA  .187
  8. Improvements • More data!! – more rows – Less noise variables (especially Y’s) • Choose a ‘longer’ Y • Get age data • Include injury status
Advertisement