I almost completed Nanodegree “Machine Learning Engineer” from Udacity and now its time for a first review. There are some positive and negative things to say, i don´t want to sugarcoat. I liked the projects and the reviewers and also some of the videos from the original courses. But first lets look at the things, which i don´t like.
Basics and duplicate content
This course is and “advanced” course with the highest skill level and it contains very basic stuff in the supervised learning chapter (slope, intercept and simple linear functions). So i skipped it to the programming part. The funny thing is, that all the stuff, which was asked in the first project, is now described in the programming videos after the first project (R², and Sum of squared error). I think the reason for this is, that the Nanodegree course is constructed from other Udacity courses (+projects), so there are 61 slides (many of them with short vids) about regression from 3 different lecturers with many intersections. And this continues over the hole course.
So what happens here is, that there is no red line and the projects just anticipate the content of the lessons, because the sequence does not fit. In the end the better way to go is just doing one of these three original courses. All of these three courses do have an focus (supervised, unsupervised and reinforcement learning).
Useless Quizzes and long breaks
I don´t like the quizzes, i don´t think you get something out of it for the most time. Some of them are just calculating an average, sum or some other really basic math. What i also don´t like is that there is always a five seconds delay between the YouTube videos, even if they only last 20 secondes. In the end you always need to press the forward button. Should be configurable.
I don´t think its very useful to give application advises to people in completely different countries. It was not clear for me, that these part of the course is optional (It is not marked as optional). So i tried to get these things done, until a reviewer told me that it is not obligatory.
Projects and reviews
The very best part of the Nanodegree are the projects. I did a course on machine learning at the university and the projects were comparable. Sure the university projects are more demanding but even with the previous knowledge of supervised and unsupervised learning I learned some new things. The reviewers also give some feedback to your answers and its not just copy pasting answers, it valuable feedback with new ideas. You also can ask questions and you will get answers. The only thing I was not happy about, was that sometimes I was told to be a bit more accurate on things, that were clear for my understanding.
Should you subscribe?
Well it depends. Because I already knew the basics of supervised and unsupervised learning, it took not much time for me to get through it and payed only 200$ until now and will not pay more than 400$ after the capstone project. The estimate of the Nanodegree is about 6 months, with 10 hours per week. If you have no idea of machine learning you will maybe need this time, but i don´t think so.
I would compare this course to a single university course. 240 hours workload estimated by Udacity, which is about 8 ECTS points. So the question you should ask yourself is, how much money do you want to pay for one 8 ECTS course? Or do you need a certificate?
To be clear the Nanodegree is not comparable in any way to a bachelor´s degree (210 ECTS) or a master´s degree (300 ECTS), but it is comparable to a single semester project. I think it is worth the money, if you don´t take more than three months for it, as you also can show that you are interested in machine learning to employers.
I payed 200$ for it and i would not suggest to pay more than that.
It´s a good intro to machine learning, but the lectures are too theoretical.