At the end of April I decided to take a break from Preceden and start using that time to level up my data science skills. I’m about 3 months into that journey now and wanted to share how I’m going about it in case it’s helpful to anyone.
Data science is very broad and depending on who you ask it can mean a lot of different things. Some folks would consider analyzing data in SQL or Excel as data science, but to me that’s never felt quite right. I prefer a definition that leans more on writing code that makes use of statistics, machine learning, natural language processing, and similar fields to analyze data.
Going into this I had a lot of programming and data analysis experience, but hadn’t done much with Python and barely knew what regression meant.
I considered continuing to learn R which I already have some experience in, but I’m not a huge fan of R so decided to start fresh and learn Python instead. Having used Ruby extensively for Preceden and other projects has made learning Python pretty easy though.
DataCamp is an online learning platform to help people learn data science. They have hundreds of interactive courses and tracks for learning R, Python, Excel, SQL, etc. If you have the interest and time, the $300/year they charge for access to all of their courses is nothing compared to the value they provide.
I’ve been making my way through their Machine Learning for Everyone career track which starts off with a basic introduction to Python and quickly dives into statistics, supervised learning, natural language processing, and a lot more.
Each course is a combination of video lectures and interactive coding exercises:
The courses are really well done and I feel like they’re giving me exposure to a broad range of machine learning topics. I wouldn’t say the courses go deep on any particular topic, but they provide great introductions which you can build on outside of DataCamp.
So far I’ve completed 10 out of 37 courses in this career track + 2 additional Python courses that were not in the track but recommended prerequisites for some of the courses that are in the track.
If you pushed through a course it might take 4 hours to complete, but I’m probably spending 10-15 hours on each course (so about 1 course/week). This is because I spend a lot of extra time during and after the course writing documentation for myself and trying to apply the material to real-world data to learn it better.
Every time I stumble across a new function or technique I spend some extra time researching it and documenting it in a public Python Cheat Sheet GitHub repository.
At first I was doing writing notes in markdown files, but have since gotten a little savier and am doing them in iPython Notebook files now. Here’s a recent example of documentation I wrote about analyzing time series.
I usually try to come up with some super simple example demonstrating how each function works which helps me learn it better and serves as an easy reference guide when I need to brush up on it when applying it down the road.
Real World Projects
For each course, I also try to apply the material to some real world data that I have access to, whether it be for Help Scout or Preceden.
For example, after DataCamp’s supervised learning course I spent some time trying to use Help Scout trial data to predict which would convert into customers.
For any projects involving Help Scout data, I usually share a short writeup afterwards in our metrics Slack channel as a way to help educate people on data science terms and techniques:
I’ve also picked up a few books which I’ve found to be excellent resources for learning matrial in more depth.
- Introduction to Machine Learning with Python – a fantastic introduction to supervised and unsupervised learning
- Machine Learning with Python Cookbook – lots of assorted code examples for accomplishing specific tasks
- Feature Engineering for Machine Learning – because as soon as you start trying to do real world machine learning projects you realize how important and tricky preparing your data can be
You can search YouTube for almost any data science topic and find dozens of videos about it. The quality varies, but I’ve found that watching a few on any topic are usually enough to fill in any major gaps in my understanding.
For example, last week I was working through DataCamp’s course on time series analysis and having trouble with a few concepts. A quick search on YouTube for videos on autoregressive models turned up this video which cleared things up for me:
After DataCamp’s course on supervised learning I spent a lot of time trying to apply it to Kaggle’s Titantic Survival data competition.
The public notebooks that other people have shared are fantastic learning resources and in the future I want to spend a lot more time trying these competitions and learning from the work others have done.
At the rate I’m going I should be through DataCamp’s machine learning track before the end of the year which will be a nice milestone in this journey. Along the way I’ll continue trying to apply the material to real world problems and hopefully wind up somewhat competent with these techniques when all is said and done. We shall see!