Databases Reference
In-Depth Information
Get a pair!
Learning how to code well is challenging. And it's even harder to go
about it alone. Imagine trying to learn Spanish (or your favorite lan‐
guage) and not being able to practice with another person.
Find a partner or a team who will be willing to pair program and
conduct rigorous code reviews. At Square, every single piece of code
is reviewed by at least another pair of eyes. This is an important prac‐
tice not only for error checking purposes, but also to ensure shared
ownership of the code and a high standard of code quality.
Here's something to try once you find a programming buddy. Identify
a common problem. Set up a workstation with one monitor and two
sets of keyboard and mouse. Think of it as teaming up to solve the
problem: you'll first discuss the overall strategy of solving the problem,
and then move on to actually implementing the solution. The two of
you will take turns being the “driver” or the “observer.” The driver
writes the code, while the observer reviews the code and strategizes
the plan of attack. While the driver is busy typing, the observer should
constantly be asking “do I understand this code?” and “how can this
code be clearer?” When confusion arises, take the time to figure out
the misunderstandings (or even lack of understanding) together. Be
open to learn and to teach. You'll pick up nuggets of knowledge quick‐
ly, from editor shortcuts to coherent code organization.
The driver and observer roles should switch periodically throughout
the day. If done right, you'll feel exhausted after several hours. Practice
to improve pairing endurance.
And when you don't get to pair program, develop the habit to check
in code using git. Learn about git workflows, and give each other
constructive critiques on pull requests. Think of it as peer review in
academia.
Productionizing machine learning models
Here are some of the toughest problems in doing real-world machine
learning:
1. How is a model “productionized”?
2. How are features computed in real time to support these models?
