Top 10 R Packages to be a Kaggle Champion

Kaggle top ranker Xavier Conort shares insights on the “10 R Packages to Win Kaggle Competitions”.

RAcross all major surveys, R has clearly dominated as one of the top programming choices for data scientists. Thus, it is no wonder that knowing the important R packages can be a vital advantage in Kaggle competitions. Xavier Conort (currently Data Scientist at Data Robot) has compiled a list of 10 R packages that played a key role in getting a top 10 ranking in more than 15 Kaggle competitions (including winning a few of them).

Since R is widely being used even outside the data science community (such as by statisticians, actuaries, etc.), this list of top 10 powerful R packages might help you in more ways than you might think.

Here are those 10 packages particularly powerful to build winning solutions:

    Allowing the machine to capture complexity:
  1. gbm [Gradient Boosting Machine]
  2. randomForest [Random Forest]
  3. e1071 [Support Vector Machines]

  4. Taking advantage of high-cardinality categorical or text-data:
  5. glmnet [Lasso and Elastic-Net Regularized Generalized Linear Models]
  6. tau [Text Analysis Utilities]

  7. Making your code more efficient:
  8. Matrix [Sparse and Dense Matrix Classes and Methods]
  9. SOAR [Memory management in R by delayed assignments]
  10. foreach [Foreach looping construct for R]
  11. doMC [Foreach parallel adaptor for the multicore package]
  12. data.table [Extension of data.frame]

Expert Advice for Kaggle Competitions: Use your intuition to help the machine by doing the following:
  • Always compute differences/ratios of features
  • Always consider discarding of features that are "too good"

The complete set of slides for this presentation by Xavier Conort: