/Otto

Code for Kaggle Otto Production Classification Challenge

Primary LanguageRMIT LicenseMIT

Otto

Code for 85th place (out of 3514) in Kaggle Otto Production Classification Challenge (private leaderboard).

Feature engineering (not all are used in final ensemble)

  • Sum of all features for each row
  • Variance of all features for each row
  • Number of filled features for each row
  • Operational features (+, -, *, /) created on top 20 features (does not work all the time)
  • Transforming features with mean-standarization (new feature = original feature - column mean)

Models

  • XGBoost
  • Neural Networks (using Lasagna and H20; only Lasagna model was used for final ensemble)
  • randomForest

Software

  • R 3.1.3

  • R packages:

    • doParallel
    • Caret
    • xgboost
    • party
    • glmnet
    • dplyr

  • Python 2.7

  • Python libraries:

    • Lasagna
    • numpy
    • scipy
    • theano