Learning R - 101

This is just a folder containing small R projects that I have done to help me understand the basics of the language

So... Where did I start?

• R programming course, by Josh Hopkings University is great course to start. It teaches you the basics and give you tools that you can perfect as you gain experience.
• SWIRL can also be a great tool if you are more into learning by doing code.

And where do I go from there?

• R for Data Science is a great beginners book that has pretty much all the basics.
• CRAN is also a great library where all R functions are described.
• Cheat sheets are great when you are working with a specific package that is relatively new to you

What if want to make graphics?

R Graphics Cookbook is a good start, as it will have many of the basics on graphs and commands

And what about maps?

Geocomputation with R has a lot of information on how to make maps

How to personalize your graphs?

This has been my toughest challenge so far. I did found resources to help in the design of the output:

• Top R color palettes has a good overall presentation of different palettes and colors available;
• Color brewer is good place if you are looking for a personalized color layout;
• For maps, tmap, has nice basic layouts;
• The Cran for tmap is also very detailed and good if you are looking for specific personalization

Machine learning with R

R has an extensive biography on machine learning techniques. It also has a variety of packages to help solving all kinds of problems. To learn these techniques, I have been studyng the classic Titanic Problem. The book Introduction to Machine Learning with R by Scott V. Burger is really the place to start. It gives a detailed look on supervised ML techiniques such as regressions, trees, randomForests and neural networks.

I have also discovered that there are an infinity of packages that will often do the same type of analysis, or at least, very similar analysis. The caret Package book has often helped me to visualize what different ML packages can do. Section 7 of the book has a great index of what each package does, and what are the tuning parameters in each case.

By trial and error, I have found the following packages to be useful:

Tree models

• rpart will process data relatively fast and is a flexible model. It is easy to define important parameters using the core function: rpart or the auxiliary rpart.control. The package produces a desision tree, based on weights of information fed to the model, therefore it require a hands on approach, where selecting parameters and pruning the tree is often required to avoid overfitting.
• party processing is not as fast as it is build as a conditional inference tree. Beacuse the tree is already based in tests of significance the results is more consistent - no need for additional steps. Visually it is also a better representation on how trees work.

Random Forest models

• As trees, the same distinction can be made in random forests: randomForest is the equivalent of rpart while party also does random Forests.
• The RRF package provides regularization options, which could help on decreasing noise from samples.

And what is next?

ML is still a large field that I am extremely interested in learning more about. Listing the things that I am planning to do:

• Get a better understanding in the topic of tuning trees and random forests;
• Improving the visualization of trees and forests;
• Start the study of Neural Networks;
• Start studying unsupervised learning techniques.