tonglu/r-book

Classifier Development in R

Classifier Development in R

This book is intended to serve as an introduction to production-level classifier development in the R programming language. The sections denoted by * are optional.

1. Introduction

• Setting up your development environment
• A review of R
• Troubleshooting
• Exercises

2. Data Preparation

• Some manual exercises
  • Filtering out values
• Transformations
  • Column transformations
  • Row transformations
  • Multi-column transformations
  • Dataframe transformations
  • Exercises
• Mungebits
  • A simple filter
  • A simple imputer
  • The mungebit data structure
  • The imputer mungebit
  • More advanced mungebits
  • Passing state between training and prediction
  • Exercises
  • *Writing our own mungebit
  • *Testing mungebits
• Mungepieces
  • An imputer with memory
  • Differences between training and prediction
  • Using the parse_mungepiece function to make mungepieces
  • Exercises
• Munging
  • Putting it all together with munge
  • Re-munging against a data.frame
  • *Inspecting stored mungepieces
  • *Debugging the data preparation process

3. Modeling

• Some manual exercises
  • A linear regression
  • A GBM model
• Tundra (turning models into R objects)
  • A linear regression tundra model
  • A GBM tundra model
  • Understanding training parameters
  • Using prediction parameters
  • The philosophy of tundra
  • * Writing our own GLM tundra container
• Stagerunner
  • An example of the full modeling process
  • Introducing stagerunner: parametrizing our modeling process
  • Quick detour: other use cases for stagerunner
  • * Advanced features of stagerunner objects
  • * Debugging stagerunner objects
  • Interactive stagerunners and caching
  • An ensemble of stagerunners
• Syberia
  • Introduction
  • The import stage
  • The data stage
  • The model stage
  • The export stage
  • * The evaluation stage
  • Testing your syberia models
  • Re-factoring complicated models with Ramd

4. Deployment

• Microserver
  • Launching a microserver on EC2
  • Deploying your syberia model to S3
• Validation
  • The philosophy of validation
  • Testing alpha versus beta operations
• Scaling with nginx