What Are the Most Common Mistakes Made by Aspiring Data Scientists?
1. “I can’t do X until I know Y.”
So many aspiring Data Scientists don’t even get started because they think they need know EVERYTHING before they can do ANYTHING. There are so many posts on LinkedIn or Quora where calculus, linear algebra, differential equations, data structures and algorithms, Hadoop, MapReduce, and numerous other competencies are listed as prerequisites for doing data science work. While it might be true that you do need a good portion of that background to be hired as Data Scientist, it doesn’t mean you can’t start learning now. The knowledge perquisites for learning pandas in Python is actually quite low, and I think a great way for anyone to get excited about data science.
2. “You need at least a Masters or PhD in <insert technical discipline here> in order to be a “real” Data Scientist.
This always drives me crazy. The endless, and pointless debate about the credentials required to be a “real” Data Scientist instead of a “fake” Data Scientist. The mistake is having an exclusive, rather than an inclusive mindset. Rachel Thomas at FAST.AI has an excellent article on providing a good education in deep learning
If you have this mindset that you need a PhD in order to become a Data Scientist, you are going to spend 4-6 years waiting to do something you know you’ll love. You can learn machine learning, deep learning, data science, etc. all with online resources. Many of the top Data Scientists on Kaggle have no more formal education than a bachelors degree - This high schooler is ranked in the top 100! There is absolutely no reason to wait!
3. “Don’t let the perfect be the enemy of the good.”
Otherwise known as the “80/20 rule”. Oftentimes, 80% of the business value of data science project can be captured with a simple method like a Decision Tree or a Linear regression. Instead of spending countless hours chasing that additional 20% developing more sophisticated methods, you could be working on another project that will create substantially more value than fine-tuning or using the state-of-the-art algorithm for your particular problem.
4. Feature Engineering > Algorithms
Algorithms are (mostly) a commodity. This is why it is so bizarre to me that so many colleges, universities, and online courses spend so much time implementing a Random Forest, Gradient Booster, or Neural Net from scratch. For academics and researchers building novel algorithms this makes a lot of sense. However, most Data Scientists and analytically-minded people need to know (1) when an algorithm is appropriate for a problem, and (2) how to implement that algorithm to solve a problem. Oftentimes, this means doing something like this:
This is a lot easier, and saves a lot of time, than building a Random Forest regressor yourself! If you are working at a startup or corporation, reinventing the wheel is a form of business malpractice. Don’t do it!
