This is one of the best massive online open courses (MOOC) on machine learning and is taught by Prof. Andrew NG. However, Prof. NG teaches the course along with MATLAB/Octave and the assignments must be done and submitted in MATLAB/Octave. Do you like the course but not the proprietary MATLAB or the sluggish Octave? Or for any reason, would you rather to use the free GNU R to complete the assignments? This project is your answer.
To view the lecture videos, slides and assignments instructions visit the course website and its wiki page. This repository provides the starter code to solve the assignments in R statistical software; the completed assignments are also available beside each exercise file. Simply follow these steps to complete the assignments:
_solution
inside the same directory of the starter code. For example, starter/ex1/computeCost.R
has an associated solution file named starter/ex1/computeCost_solution.R
In order to produce similar results and plots to Octave/Matlab, you should install a few packages:
rgl
package is used to produce the 3D scatter plots and surface plots in the exercises.SnowballC
: portStemmer
function in this package has the same role of the portStemmer.m
.raster
package is used to produce the plot of the bird in exercise 7.jsonlite
and httr
packages are needed for submission.pinv.R
: The ginv
(generalized inverse) function in MASS
package doesn't produce the same result of the Matlab pinv
(pseudo-inverse). I wrote pinv.R
as the modified version of MASS ginv
to produce the same result of the MATLAB pinv
. For more info see this stackoverflow discussion
lbfgsb3_.R
: Certain optimization tasks could only be solved using lbfgsb3
package, yet there are a few bugs in this package. The purpose of lbfgsb3_.R
is to address these bugs; it is used for exercises 4 and 8. Beware that fmincg
or fminunc
optimization function in MATLAB takes one function as input and computes cost and gradient simultaneously. However, cost and gradient functions must be supplied into optim
or lbfgsb3
functions individually.Before starting to code, install the following packages:
install.packages(c('rgl','lbfgsb3','SnowballC','raster','jsonlite', 'httr'))
After completing each assignment, source("submit.r")
and then submit()
in your R console.
I submitted the solutions to Coursera for testing and the scores were 100%. Please report any problem with submission here.
A few screenshots of the plots produced in R:
This project is released under MIT to the extent it is original.