Machine Learning Foundations course on Coursera

From raju

The Machine Learning Foundations: A Case Study Approach is a 6 week course offered by the University of Washington on Coursera. It is the first course in the Machine Learning Specialization which consists of 6 courses and a Capstone project. It is taught by Carlos Guestrin and Emily Fox. The course costs $79.

I am interested in machine learning. So I enrolled for the 2015-11-09 - 2015-12-28 slot. I finished the course and scored 100% in all the assignments. The course content, assignments, case studies are interesting, challenging and exciting. I definitely learned a thing or two about machine learning. However, I do not recommend this course to others for the following reasons.

Conflict of Interest:

The course uses a proprietary software called “graphlab create”. It is developed and promoted by a company owned by Carlos Guestrin who is one of the instructors. This software is free for academic purposes, but costs money to use in a commercial environment. As such the programming skills are not readily transferable unless you plan on paying for it. Even if you pay for the licenses, the skills can become obsolete if the company were to go bankrupt in the future.

The instructors say that users can use any software of their choice, but they highly recommend using graphlab create. However IMHO it is very difficult to do the assignments without graphlab create unless you are already an expert in the other alternative.

In the end, you will just learn the concepts in machine learning which one can do by reading a book or watching a youtube video. The idea of a course is to gain some hands-on programming experience as well.

Technical difficulties:

  1. I was unable to use graphlab create on my desktop running Linux. Running the same model multiple times gives different results. So the results were not reproducible. In the end, I was forced to create an Amazon EC2 instance just so I can do this course. The Amazon EC2 instance is free for first year and costs money after that.
  2. The lectures used graphlab create 1.5 which is an older version. The Amazon EC2 instance uses 1.7 which is the latest version. The new version is incompatible with old version making it impossible to reproduce the results shown in lectures. This happened in multiple but not in all the lectures.
  3. Some of the models created in the lectures, assignments use too much memory (~ 4 GB). In such cases, if there are multiple models involved, I had to constantly kill/restart ipython kernels in order to switch between them.


  1. This is an expensive course and costs $79 to get the certificate. It is free if you do not need the certificate.

After investing 6 weeks of time in this introductory course, I was in a dilemma as to whether it is worth finishing the remaining courses, capstone project. In the end I decided against it.

I am currently on the scout for machine learning courses using python and open source software. If you know of any, please shoot me an email [kamaraju at gmail dot com].