Stanford University's Machine Learning course review on Coursera

In this post, I will review one of the most popular online Coursera machine learning courses I've taken.

The Machine learning course can be called one of the flagship courses for Coursera. This course was created in the early days of the service by one of its founders, Andrew Ng.

To quote a brief description from the site:

"Machine learning is the science of making computers act without using explicit deterministic algorithms. In this course you will learn about the most effective machine learning algorithms, get experience in their practical application, and learn how to make them work for you. More importantly, you'll learn not only the theoretical foundations of machine learning, but also the practical know-how required to quickly and effectively apply algorithms to solve new problems."

Indeed, by the end of the course I have developed an understanding of the basic principles and mechanisms of machine learning. Also, what is not less important, I have experience in solving specific machine learning problems in practice.

The course lasts 11 weeks in total. Each week includes 1-2 hours of video lectures, a theory test and a practical assignment on the application of specific machine learning methods. In total, it took me 4-6 hours to go through all the material and complete all the assignments of one week.

Course Strengths:

• Lecturer explains well and clearly, although it is subjective.
• Excellent practical assignments in specific applications. You can choose both Matlab, which is more common in Russia, and free GNU Octave.
• Checking of tests and homework is conveniently organized. Each homework is divided into several logical stages, each of which can be checked one by one. In case of homework, the checking script shows not only the correctness of the proposed solution, but also what is expected.
• All videos have subtitles.
• Conveniently, you can choose your own learning speed. In principle, the course itself is looped, and almost as soon as one stream ends (11 weeks), another one starts. Lectures and assignments can be taken both before and after the deadlines suggested by the creators.

The disadvantages mainly stem from the fact that the course is old and perhaps not always updated promptly:

• The quality of video recordings sometimes leaves much to be desired. Often it is impossible to understand what Andrew is saying without subtitles. In some cases even in subtitles this or that line is missing, because, probably, the author of subtitles himself could not understand the text.
• A couple of times there were errors in the subtitles, which can be confusing when trying to understand the thesis voiced. For example, the meaning of "for cycles" said by the lecturer can be written as "four cycles".
• Unfortunately, starter scripts don't always work right away. I don't remember exactly, but I had to literally replace some constant in every script with another one to make it work. The problem is alleviated by the fact that the solution can be found quite quickly on the official course forums.

Unfortunately, any less "official" certificate can only be obtained for money. Nevertheless, the course is very well known in the relevant circles and can be called somewhat standard.

In general, I recommend it. It is a really good course, after which many things in machine learning become clear.