Machine Learning with R

R has a wide variety of machine learning (ML) models. While the many ML functions solve similar problems by predicting various outcomes, they use a confusing array of different command styles, making them hard to learn. Fortunately, the caret package provides a standard approach to dozens of ML functions, greatly speeding learning and use. This two-day hands-on workshop starts with ML basics and takes you step-by-step through increasingly complex modeling styles.

Photography by Steve Chastain http://www.stevechastainphotography.com/

Most of our time will be spent working through examples that you may run simultaneously on your computer. You will see both the instructor’s screen and yours, side-by-side, as we run the examples and discuss the output. However, the handouts include each step for people who prefer to just take notes.

This workshop is available at your organization’s site, or via webinars.

The 0n-site version is the most engaging by far, generating much discussion and occasionally veering off briefly to cover topics specific to a particular organization. The instructor presents a topic for around twenty minutes. Then we switch to exercises, which are already open in another tabbed window. The exercises contain hints that show the general structure of the solution; you adapt those hints to get the final solution. The complete solutions are in a third tabbed window, so if you get stuck the answers are a click away. The typical schedule for training on site is located here.

A webinar version is also available. The approach saves travel expenses and is especially useful for organizations with branch offices. It’s offered as two half-day sessions, often with a day or two skipped in between to give participants a chance to do the exercises on their own and catch up on other work. There is time for questions on the lecture topics (live) and the exercises (via email). However, webinar participants are typically much less engaged, and far less discussion takes place.

For further details or to arrange a webinar or site visit, contact the instructor, Bob Muenchen, at muenchen.bob@gmail.com.

Prerequisites

This workshop assumes a basic knowledge of R. Introductory knowledge of statistics is helpful, but not required.

Learning Outcomes

When finished, participants will be able to use R to apply the most popular and effective machine learning models to make predictions and assess the likely accuracy of those predictions.

Presenter

Robert A. Muenchen is the author of R for SAS and SPSS Users and, with Joseph M. Hilbe, R for Stata Users. He is also the creator of r4stats.com, a popular website devoted to analyzing trends in data science software, reviewing such software, and helping people learn the R language. Of the over 750 R blogs on the Internet, Feedspot rates r4stats.com the eleventh most influential.

Bob is an ASA Accredited Professional Statistician™ with 35 years of experience and is currently the manager of OIT Research Computing Support (formerly the Statistical Consulting Center) at the University of Tennessee. He has taught workshops on research computing topics for more than 500 organizations and has presented workshops in partnership with the American Statistical Association, RStudio, DataCamp.com, New Horizons Computer Learning Centers, Revolution Analytics (acquired by Microsoft), and Xerox Learning Services. Bob has written or coauthored over 70 articles published in scientific journals and conference proceedings and has provided guidance on more than 1,000 graduate theses and dissertations.

Bob has served on the advisory boards of SAS Institute, SPSS Inc., StatAce OOD, the Statistical Graphics Corporation, and PC Week Magazine. His suggested improvements have been incorporated into SAS, SPSS, StatAce, JMP, jamovi, BlueSky Statistics, STATGRAPHICS and numerous R packages. His research interests include statistical computing, data graphics and visualization, text analytics, and data mining.

Computer Requirements

On-site training is best done in a computer lab with a projector and, for large rooms, a PA system. The webinar version is delivered to your computer using Zoom (or similar webinar systems if your organization has a preference.)

Course programs, data, and exercises will be sent to you a week before the workshop. The instructions include installing R, which you can download R for free here: http://www.r-project.org/. We will also use RStudio, which you can download for free here: http://RStudio.com. If you already know a different R editor, that’s fine too.

Course Outline

  1. INTRODUCTION

1.1 Topics

1.2 Preparing Your Computer

1.3 Note to System Administrators

1.4 Workshop Files

  1. OVERVIEW OF MACHINE LEARNING

2.1 Definition

2.2 Goals of Machine Learning

2.3 Types of Problems

2.4 Types of Machine Learning

2.5 Statistical Models

2.6 Heuristic Models

2.7 Bagging Models

2.8 Boosting Models

2.9 Data Set Size

2.10 Model Building Steps

2.11 Allocating Data: Train, Validation, & Test

2.12 Types of Splits

2.13 Fitting Issues

2.14 Feature Selection

2.15 Feature Selection Approaches

  1. INTRO TO THE caret PACKAGE

3.1 Formula Method vs. Non-formula

3.2 Each R Package May Predict Differently

3.3 Goals of the caret Package

3.4 caret References

3.5 Types of Models in caret

3.6 Type of Predictions by Model

3.7 Types of Functions in caret

  1. DATA PRE-PROCESSING

4.1 The caret::preProcess Function

4.2 preProcess Tasks

4.3 Pre-processing the titanic Dataset

4.4 Examining Factor Variables

4.5 Setting Value to Model

4.6 Examining Numeric Variables

4.6 Examining Relation to Target

4.7 Creating Some Problems to Diagnose

4.8 Counting Missing Values

4.9 Visualizing Missing Values

4.10 Imputing Missing Values

4.11 Finding High Correlations

4.12 Finding High Correlation Automatically

4.13 Getting Names of Highly Correlated vars

4.14 Finding Variables with Near Zero Variance (NZV)

4.15 Automatically Finding and Fixing Problems

4.16 caret’s Imputation Options

4.17 Fixing All Problems at Once

4.18 What is a preProcess Object?

4.19 Details of What preProcess Will Do

4.20 Box-Cox / Yeo-Johnson Lambda Values

4.21 How to Use preProcess to Fix the Problems

4.22 Effect of Yeo-Johnson Normalization

4.22 Practice Time

  1. PRINCIPAL COMPONENTS ANALYSIS

5.1 Prepare the Workspace

5.2 What PCA Does

5.3 Visualize Variance of Original Variables

5.4 Now Create Principal Components

5.5 Plot the Principal Components

5.6 Interpreting PC Construction

5.7 Practice Time

  1. DUMMY VARIABLES

6.1 Preparing the Workspace

6.2 Dummy Variables Defined

6.3 Creating Dummy Variables Manually

6.4 Creating Dummy Variables Automatically

6.5 Adding Auto-Dummies

6.6 Practice Time

  1. PARTITIONING DATA SETS

7.1 Preparing the Workspace

7.2 Select Rows to Train On

7.3 Create Train and Test Data Frames

  1. FEATURE SELECTION

8.1 Preparing the Workspace

8.2 Example Data Set

8.3 Supervised vs. Unsupervised

8.4 Built-in Feature Selection

8.5 Filter Methods

8.6 Selection By Filtering (SBF)

8.7 Random Forest SBF

8.8 Saving the Variables Chosen by SBF

8.9 Wrapper Methods

8.10 Wrapped Feature Extraction (RFE)

8.11 Recursive Feature Elimination (RFE)

8.12 Impact of Number of Variables

8.13 Saving the Variables Chosen by RFE

8.13 Practice Time

  1. CONTROLLING MODEL TRAINING

9.1 Prepare the Workspace

9.2 The trainControl Function

9.3 Control Arguments

9.4 Measuring Model Quality

9.5 trainControl Example

9.6 Timing Training

9.7 Practice Time

  1. NAIVE BAYES

10.1 Prepare the Workspace

10.2 NB Algorithm

10.3 NB Advantages

10.4 NB Disadvantages

10.5 NB Model Training

10.6 Interpreting NB Model Tables

10.7 Interpreting NB Model Plots

10.8 NB Predictions & Validation

10.9 Save Model for Future Use

10.10 Practice Time

  1. CLASSIFICATION AND REGRESSION TREES

11.1 Prepare the Workspace

11.2 CART & Ctree Algorithms

11.3 CART & Ctree Advantages

11.4 CART & Ctree Disadvantages

11.5 Tree Model Training

11.6 Tree Model Plot using partykit

11.7 Tree Model Plot Using rpart.plot

11.8 Tree Model Converted to Rules

11.9 Tree Variable Importance

11.10 Tree Predictions & Validation

11.11 Practice Time

  1. RANDOM FORESTS

12.1 Prepare the Workspace

12.2 RF Algorithm

12.3 RF Advantages

12.4 RF Disadvantages

12.5 RF Train the Model

12.6 RF Variable Importance

12.7 RF Prediction & Validation

12.8 Save RF Model for Future Use

12.9 Practice Time

  1. GRADIENT BOOSTING MACHINES (gbm)

13.1 Prepare the Workspace

13.2 gbm Algorithm

13.3 gbm Advantages

13.4 gbm Disadvantages

13.5 gbm Model Training

13.6 gbm Variable Importance

13.7 gbm Predictions & Validation

13.8 Practice Time

  1. NEURAL NETWORKS (NN)

14.1 Prepare the Workspace

14.2 Neural Network Algorithm

14.3 Neural Network Advantages

14.4 Neural Network Disadvantages

14.5 Neural Network Training

14.6 Neural Network Variable Importance

14.7 Neural Network Tuning Plot

14.8 Neural Network Prediction & Validation

14.9 Practice Time

  1. SUPPORT VECTOR MACHINES

15.1 Prepare the Workspace

15.2 SVM Algorithm

15.3 SVM Advantages

15.4 SVM Disadvantages

15.5 SVM Training Model

15.6 SVM Variable Importance

15.7 SVM Plots

15.8 SVM Prediction & Validation

15.9 Practice Time

  1. LOGISTIC REGRESSION (LR)

16.1 Prepare the Workspace

16.2 LR Algorithm

16.3 LR Advantages

16.4 LR Disadvantages

16.5 LR Training Model

16.6 LR Variable Importance

16.7 LR Interpretation

16.7 LR Prediction & Validation

16.8 Practice Time

  1. DISCRIMINANT ANALYSIS

17.1 Prepare the Workspace

17.2 LDA/QDA Algorithm

17.3 LDA/QDA Advantages

17.4 LDA/QDA Disadvantages

17.5 LDA Training

17.6 LDA Common Functions

17.7 LDA Interpret Model

17.8 LDA Study Scores Manually

17.9 LDA Prediction & Validation

17.10 Practice Time

  1. LINEAR REGRESSION ANALYSIS

18.1 Prepare the Workspace

18.2 Regression Analysis Algorithm

18.3 Regression Analysis Advantages

18.4 Regression Analysis Disadvantages

18.5 Regression Training

18.6 Regression Diagnostic Plots

18.7 Regression Analysis Model Interpretation

18.8 Linear Regression Prediction & Validation

18.9 Practice Time

  1. ROC CURVES

19.1 Prepare the Workspace

19.2 ROC Curve Defined

19.3 Plotting the ROC Curve

19.4 Extracting Measures from ROC Curves

19.5 Overlaying ROC Curves

19.6 Testing Significance Between ROC Curves

19.7 Practice Time

  1. CLASS IMBALANCE ISSUES

20.1 Prepare the Workspace

20.2 The Problem

20.3 Modeling Solutions

20.4 Changing Cutoff

20.5 Changing Prior Probabilities

20.6 Changing Case Weights

20.7 Sampling Changes

20.8 Synthetic Data

20.9 Practice Time

  1. MODEL TUNING GRIDS

21.1 Prepare the Workspace

21.2 Tuning Parameters

21.3 Examining rpart Parameters

21.4 An Example Tuning Grid

21.5 Training with Parameter Grid

21.6 Predicting with Tuning Grid

21.7 Practice Time

  1. INTERPRETING BLACK-BOX MODELS

22.1 Preparing the Workspace

22.2 Global vs. Local Interpretation

22.3 Creating a Profile Grid

22.4 The Resulting Grid

22.5 Creating a More Interpretable Model

22.6 Generating an Interpretable Model

22.7 Practice Time

  1. CHOOSING A MODEL

23.1 Summary of Above Models

23.2 Choose for Performance

23.3 Choose for Understanding

23.4 Choose for Ease of Deployment & Maintenance

23.5 Choose for Speed

  1. CONCLUSION

24.1 Brief Review

24.2 Feedback

24.3 Future Support

24.4 Question Time

Here is a slideshow of previous workshops.