Machine Learning with Python: from Linear Models to Deep Learning.

The course is pretty disappointing. The total lecture time for each chapter is roughly 20 minutes, mostly consisting of three black-and-white slides with outdated and convoluted explanations of how certain models work. This format makes it difficult to walk away with a solid understanding of most machine learning models.

One highlight of the course is the lectures by an Israeli professor, which are definitely worth watching. Unfortunately, she only teaches about 5-10% of the entire course, leaving much of the content lacking in depth and clarity.

The programming assignments are decent, providing a practical application of some concepts. However, many of the other questions are math-focused (which is fine) but are paired with oddly phrased high-level multiple-choice questions that can be confusing.

Overall, this course is not at the same level as the other three courses in the MicroMasters program and certainly not worth the $300 price tag. For those looking for a more valuable learning experience, I recommend cheaper alternatives on platforms like Coursera, such as the courses by deeplearning.ai or the University of Colorado Boulder.

Not recommended and 100% regret not dropping out

The negative reviews of this course on class central had me on guard for a chaotic learning experience as I began this course. However, the course greatly exceeded those low expectations. This leads me to believe that the course may have undergone some recent improvements in response to student feedback.

I've found it to be a rewarding and challenging path to deeper understanding on a variety of ML topics, but never a cause of excess stress or hopelessly difficult problems. I am a professional data scientist and am familiar with many of the ideas in the course, but find that I've gained a firmer comprehension of the specifics of many techniques through this course.

The course combines mathematical exercises with programming ones. You need to be able to write python code as well as understand multivariate calculus, probability, and basic linear algebra in order to take it. It will also be very helpful to have familiarity with the concepts and techniques of machine learning from a broad, qualitative perspective. That way you'll be motivated to know the details of the algorithms rather than lost as to their relevance.

My sole complaint is that the programming exercises hold your hand to the point that you lose much of the benefit of implementing algorithms yourself. I believe that this is partially due to the fact that the course must limit itself to problems that can be automatically graded, rather than self-directed projects which support this kind of integrated understanding. You will need to experiment and apply the solutions on your own in order to have a complete first principles understanding of the techniques.

Because the course takes you part of the way towards the goal of practical and theoretical comprehension, it is a highly valuable experience well worth your time.

I recommend it without hesitation to anyone who would benefit from understanding how machine learning works from mathematical fundamentals.

A good mid-theoretical course on Machine Learning algorithms

I believe the reasons of many negative reviews for this course is due to a mismatch between students' expectations and content actually delivered.

I agree that it is not the "best" MIT course, in large part I would say due to the platform limits where autograding requires to break down the software projects in too many micro parts, and as other comments have noticed, tie a bit your hands on what you can do in the projects.

This is not a course to learn how to apply ML libraries, not even I would say how to code ML algorithms. Python is here because there is no ML without computational aspects, but it is not really the focus of the course. Its presence in the title attracts the masses but then it disappoints those that come to the course for it. At this point the course would have been better to have used Julia rather than Python to illustrate the ML algorithms.

What the course excels in my view is in giving the intuitions of the algorithms themselves, the theory behind them, but still with a formalism that uses only the minimum level of math - this is NOT a course for mathematicians in the same way it is not a course for programmers.

I arrived to MIT 6.86x as my last course in the MicroMaster program, with completely zero experience in ML, but now I managed to implement (almost) all the algorithms described in my own Machine Learning library - BetaML (including some algorithms not discussed in the course, as decision trees/random forests).

I enjoyed it and, as Community Teaching Assistant during my run of the course, I produced very detailed notes of the course that you will find on GitHub (https://github.com/sylvaticus/MITx_6.86x).

I hope you will enjoy it as much I did. And yes, I liked the videos also of of the Finnish prof.

I finished the course and received a certificate and i can easily say that it is one of the best courses online. People are leaving negative reviews because the course is about 80% independent work, which is perfectly fine since it is a GRADUATE course, not an introduction to calculus. Many people compare it with Probability course of Prof Tsitsiklis, another MIT gem, but they are very different. Probability is dry math and has 20+ lectures dedicated to it, so it starts from zero to advanced with lots of tutorials and exercises, making it easier to study for homework and exams. This course assumes people already have mastered probability and statistics, including linear algebra and python coding, especially Numpy library, so it is sort of a survey of basic principles of ML.

Prof Regina is more sort of organized in her lectures because she covered very specific units and it may be her style as well, while Prof Tommy is more jumpy because his assumption is that the students taking the course already have the background and he is covering so many topics in one lecture. I also think he emphasizes a lot the maths, which is important if you want to understand vs if you just wanna run code. Folks, you need to learn how to work independently, find sources on google, books, youtube, udemy etc to fill your gaps, Grad School is not freshmen years. That's how you learn and understand things.

I'd encourage anyone to take this course, its the best survey on almost everything ML out there, its hard to find such breadth in a single course. After finishing it, you will have the necessary background to go deeper into any ML field very easily.

I have finished several online courses the last 3 years. I can safely say that this is the worst online course I have ever seen. The lectures are very bad and short. There are almost no examples. I guess the logic of the course is to provide some short and vague ideas and then teach you by googling and scanning the discussion forum for hints. Some people might argue that this way of teaching is better because it leads to a better understanding. Personally I think that this should be reserved for certain parts of a course and it should not be the primary teaching method. I am comparing it with the probability course of MIT (both courses are part of MicroMasters in data science). The probability course would require much more time to complete but the video lectures were more than enough to get full grade.

P.S. The main lecturer's accent is very bad and from time-to-time confusing. The TAs do not seem to know much about the class either, sometimes they are posting on the forum asking how to solve a problem....

Tl;DR

This course is bad, start only if you are planning to get the micromasters from MIT.