How I Learned About Machine Learning

June 01, 2018

Originally an email to a friend in June 2018.

A friend asked me how I got started with machine learning the other day. I found the train and test nature of machine learning addictive, to say the least, and spent 2016 obsessed with the topic. Stealing any moment I could work on a Kaggle problem or read the latest posts that flooded Hacker News at the time. The response below is what I told my friend.

I started with the two courses mentioned in a Medium article. The Udacity course is overly simplistic. After that, I moved to Andrew Ng’s Coursera course — it’s rough, but I plowed through it.

After that, I went on to use this course by Jeremy Howard, which was the most practical. The right balance between theory (all the linear algebra) and implementation (using Keras).

Python Machine Learning was also a good book. When I have the willpower, I try to avoid screens before bed, so I was using this to supplement before bed. Happy to lend you this one if you’re interested.

I also enjoyed listening to Talking Machines during my year of ML. Partially Derivative was also okay, but too focused on politics.

Some of the Kaggle tutorials are helpful too! :)

I haven’t taken this one, but it’s on my radar: https://developers.google.com/machine-learning/crash-course/

If I could do it again, I would do the Udacity course, then take the google and/or fast.ai course and supplement it with a book and the Andrew Ng class.

The Google course on Android development was well done. However, it focuses directly on TF, while fast.ai uses Keras, which can use TF or Theano as a backend.