I am a huge advocate of online learning. If you have read my posts advising new traders I always suggest going through online courses and specializations that give people the basic tools they need to tackle quantitative trading. Having this basic formation has always seemed critical to me since without knowing basic mathematics, statistics and programming it is fundamentally much more difficult to succeed at automated trading, primarily because it will be hard to use data and understand how ideas need to be tested and how the results of those tests need to be interpreted. Today I want to talk to you about my latest experience with online learning, primarily my experience with the Udacity Machine Learning Nanodegree, which I just finished after three months of intensive work and learning. Note that I am not involved with Udacity in any way – besides being a student – nor are they paying me to talk about them here.
–
–
First let me talk about the reasons why I decided to go through this nanodegree. I had never done a formal course in machine learning – all I had learned was through actual coding experience both in trading and in my PhD and books/articles I had read – so I wanted to go through a class to get a stronger theoretical background and an overview of machine learning methods and tips/tricks that I might be ignoring. When I started looking I found several interesting machine learning offers available online, the two that stood out were the Udacity nanodegree and the coursera machine learning specialization. The coursera specialization is charged all from the start – going for a bit less than 600 USD – while the udacity nanodegree is charged at 200 USD/month and the total charge depends on how long you take to finish the degree. The coursera option offers financial aid if you qualify while the nanodegree offers a 50% cash back option after you finish the nanodegree if you finish within 12 months and spend at least 2 months working on the degree.
The reason why I decided to go for the nanodegree instead of the coursera specialization was mainly because coursera tends to be very focused towards academic development – they are after all based solely on university courses – while the nanodegree is an industry-designed course that contains more practical approaches to the problems. I also liked the fact that the nanodegree was focused around projects at the end of learning sections and that lectures were structured with small quizzes in between instead of homework assignments, something that made the course easier for me to handle while also working my regular full-time independent trader role. This of course does not mean that the coursera specialization is bad – I have in fact recommended it before – but it just means that it didn’t offer the exact focus and methodology I wanted in this case. I have also done a few coursera experiences in the past so I also wanted to try a different learning platform.
–
–
The udacity nanodegree is structured in sections. There are 6 sections that span basic statistics, supervised learning, unsupervised learning, reinforcement learning, deep learning and a final capstone project were you apply this knowledge in a particular problem that you’re personally interested in solving. Each section is comprised of a series of lectures that you should go through – about 1-5 minutes each – with quizzes in between some of these lessons to further help you grasp the content. At the end of each section there is a mandatory small project that you must carry out. For the first several sections the projects are presented in the form of jupyter python notebooks – which makes them easy to follow and complete – while for the reinforcement learning and deep learning projects you are faced with a much more hands-on set of coding challenges.
My experience with the Udacity nanodegree was excellent. I definitely liked the way in which the sections were structured and the way in which the courses flowed. The machine learning courses are based mainly on the Georgia tech courses that are given for their masters degree in machine learning plus some other material that was created for this nanodegree as well as for other udacity courses. There are some times when you might feel disengaged because the contents were not all explicitly created for the nanodegree but the way in which the courses are laid out made sense to me. Some people also find it annoying that there are some sections were the same content is explained by different videos but I found this enriching as it’s often easier to learn material when it’s presented to you in different ways. I can see that they did this particularly with topics that might be difficult to understand if you only received one explanation, so I thought this was a clever way to ensure better learning.
–
–
Of course there were parts that I dreaded. For example there was a module on basic statistics that I absolutely hated going through as I am way past learning how to calculate an average and a standard deviation. You can clearly skip this if you want to but I am a bit obsessed with always going through all course material in online degrees so I simply went through it all and relearned how to do very basic descriptive statistics. This only took a few hours – so it wasn’t as painful – but I certainly think they could just assume people know this stuff. If you are going to assume people know calculus, linear algebra and python programming when doing this nanodegree you must as well assume they know basic descriptive statistics. I believe the people who meet the prerequisites for the degree but do not know how to calculate a mean or a box-and-whiskers plot are few and far between.
On the next post I will continue talking about the machine learning aspects of the course as well as my experience with the reviewers, other students and the capstone projects. If you would like to learn more about machine learning in trading and how you can actually mine trading strategies that constantly adapt please consider joining Asirikuy.com, a website filled with educational videos, trading systems, development and a sound, honest and transparent approach towards automated trading.strategies.
