In this post we briefly talk about our experience with the Coursera Machine Learning course from Stanford.
I took part in the (verified) Coursera course during the period March-May 2016. The course is created and presented by Professor Andrew Ng, the material is clear and to the point. The instructor puts a lot of effort in explaining step by step what is going on. Although some calculus background might be needed, the course does not really depend on it.
The course demystifies a lot around the machine learning world and at the basics it is not that difficult: a good understanding of your data and some basic algorithms can get you already very far. It was amazing to see that with very little code and basic hardware you could run algorithms for character recognition and simple recommender systems.
The quizzes are not too difficult, usually 5 questions, you need at least 4 out of 5 correct to pass the course.
You can spend quite some time in the exercises. Most of them are presented as code in Octave/MATLAB you have to complete. Sometimes this can be done in one or two lines, depending on your knowledge of the matter and the understanding of the Octave functions. Also, as most of the code is already given, you have the reflex of just adding the missing parts without looking at the workings of the surrounding code.
There appear to be some issues with Octave on Windows, but I did not really encounter serious problems. The language isn’t that fast but fast enough for this track.
For me this is the weakest part of the course: Octave is not that much used (yet?) in data science compared to R and Python. Sometimes I spent too much time searching for the correct function or syntax.As I currently using R and Python, another syntax was sometimes frustrating.
A very interesting course to follow, very well explained, but if the practical part was in Python or R this would have been a 5-star for me.