practical_supervised_machine_learning

Introduction

Resources for a practical on supervised machine learning

Code

The Rmarkdown scripts and R scripts constitute the practical.

cross_validation_practical.Rmd shows the basics of cross-validation.

Ch8-baggboost-lab.Rmd shows the basics of decision trees and random forests (bagging and boosting).

If you have time, you can try code in the additional_code folder:

• caret_rf.Rmd shows the basics of using the caret package in R to build a machine learning pipeline.

Installation

Clone or download this repository.

Then install R and R Studio.

• Install R

  https://www.r-project.org/

• and R Studio

https://www.rstudio.com/products/rstudio/download/preview/

OR

follow the instructions here:

https://cambiotraining.github.io/intro-r/#Setup_instructions

From the command line run the R script installer.R to install all packages

R --no-save < installer.R

OR

run the script installer.R in R Studio.

Exercise

For an exercise, do these problems.

• Exercise 1: Download data from

https://github.com/neelsoumya/teaching_reproducible_science_R/blob/main/metagene_score.csv

and train a classifier to predict yes/no (flag_yes_no). This would be a binary classifier. Do this with cross-validation. Write your own R code to do this task. You can work in groups. Do this in class.

The data is also available here

https://github.com/neelsoumya/practical_supervised_machine_learning/blob/main/metagene_score.csv

• Exercise 2: Download data from

https://archive.ics.uci.edu/dataset/2/adult

and train a classifier to predict if income > 50K or < 50K (binary classifier). Do this with cross-validation. Write your own R code to do this task. You can work in groups. Do this in class.

Challenge: Use cross-validation to select a few important features.

The data is also available in the adult folder here

https://github.com/neelsoumya/practical_supervised_machine_learning/tree/main/adult

Challenge exercise

How would you select the features that go in the logistic regression model?

Think of a brute force approach.

Can you think of a more sophisticated apporach?

For an advanced exercise use the glmnet package in R.

See

https://glmnet.stanford.edu/articles/glmnet.html#logistic-regression-family-binomial

Need more of a challenge? See the additional_code folder and see the caret programs.

• Exercise 3: Download the data from

https://github.com/neelsoumya/practical_supervised_machine_learning/blob/main/diabetes.csv

* Remember to visualize the data and normalize features

* Build a random forest model to predict diabetes outcome (0/1)

* Plot the out of bag (OOB) error as a function of the number of trees

Resources

Free PDF of book and R code

• https://www.statlearning.com/

• https://www.statlearning.com/resources-second-edition

More practical tutorials and R code

• https://cambiotraining.github.io/intro-machine-learning/

Acknowledgements

I thank Dr. Bajuna Salehe for useful discussions and feedback.

All material is take from the following resources:

• https://www.statlearning.com/resources-second-edition

• https://cambiotraining.github.io/intro-machine-learning/

Contact

Soumya Banerjee

sb2333@cam.ac.uk