Session 14 Final session

• Structure of a predictive paper
  • Introduction
  • Review of previous predictive papers for respective data
  • Methods & data
    • Description of method etc.
  • Empirical results
  • Conclusion
• Term paper deadline
• Pagedown & Bibtex
• API reviews
• Training & test data (RFs)
  • Definition of test error: Resampling methods (2): Cross-validation
  • RF logic: Out-of-Bag (OOB) Error Estimation
    • OOB error used to estimate test error
  • BUT in principle you could take out some data from your labeled training dataset to validate your model in the end.
    • In the lab Lab: Random Forest for text classification you would only use a subset of the training observations (training_data==TRUE) to grow the forest, e.g., 80% and use the other 20% for a final validation