loading...

5 courses in 1 week, a retrospective

drbearhands profile image DrBearhands ・2 min read

This was a bad idea.

While I have an MSc in Artificial Intelligence, the coverage of machine learning techniques was, in my opinion, rather lackluster. There were some topics that got decent coverage, such as Monte Carlo simulation and evolutionary algorithms, but most were grouped together in an introductory course that didn't require too much understanding of the subject matter for a passing grade. Yes I am very salty about this.

So I decided to fill one big gap in my knowledge with a 5 course deep-learning specialization on coursera.

The first week was free!

I had a basic idea of what a neural network works so how hard could it be?
I took a week off. A client I'm working for was also interested in using deep neural networks in a project, as it didn't look like it was going anywhere with hand-crafting. They paid me 2 working days to do the course, which was nice.

Armed with overconfidence, I started my binge learning experience. I felt rather drained after the second day already, but with 2.5 courses down, I was on schedule. I eventually managed to finish them all within the planned time, yesterday at 9 pm.

I hate neural networks now.

That feeling will probably pass in a few days. Nevertheless, I would discourage anybody from doing this. It may have gone better if the assignments were not in python, a language which I consider to have been a mistake, but realistically any course is going to have some element that just drains you, no matter how much interest you have in the subject.

So were the courses themselves any good? Surprisingly, yes, they were a lot better than I had thought. However, the assignments are very simple with respect to the subject matter. You could solve them without understanding what you are doing as they often came down to "write this formula using numpy/tensorflow/keras". The real challenge was finding the right documentation and, again, fighting against the kind of unhelpful error messages that result from python's lack of a type system and useful composition rules. E.g. raise_exception_on_not_ok, some failed assertion 4 layers deeper in the call stack.

Because of aforementioned ease of passing, I would not personally hire anybody based on their ability to get these certificates, even dismissing the possibility of cheating. The courses really are what you make of them, but they do contain a lot of information, so I would recommend them if you're interested in the subject. Just don't rush it.

Discussion

pic
Editor guide