One aspect of Factual’s culture that I find most rewarding is the commitment to engineers’ technical development. Factual supports learning by encouraging all engineers to adopt new technologies, read research articles, and attend conferences all in order to expand our technical breadth. We also focus on shipping, so we can develop exciting new tools and systems like neutronic. Striking this balance between learning and productivity can be tricky. So how do we achieve it? In the summer, we have a full hackathon, but what about the rest of the year? Some companies’ methodologies advocate for lab days, others have a 10% rule, and at Factual one approach we take is Data Sci-Fris!
Data Sci-Fri is an initiative that gives engineers time on Friday afternoons to explore various data science techniques and methodologies together. As a group we select a topic and then it’s part seminar, part extended hackathon. The first few weeks, the focus is learning. This learning can take many forms including lectures by a guest speaker or team experts, paper discussions, or working through a book. Once we grasp the concepts, we spend the next few weeks working together to apply our new knowledge to a problem (or two) here at Factual with the data we have on hand.
Some projects work, some we quickly realize require more time than a couple Friday afternoons and some… well, new tools can’t solve everything. Recently, we did a series on Bayesian methods, which started with reading and working through the problems in Thinking Bayes. Our Insights team show us how they leverage Bayesian ideas, then we played with Markov chain Monte Carlo (MCMC) and inference with pyMC3. We are applying these techniques to identifying whether a coordinate is likely associated with driving on a road as opposed to visiting a specific place of interest.
In the previous session, on Natural Language Processing (NLP), we had a guest speaker show us how she builds RNNs with Keras in her research to generate stories. With those tools, we tried learning the mappings between our sources’ categories to our internal category taxonomy. While it worked for our English speaking countries, we realized how tricky that automation is when our data contains mixed languages and character sets. No matter the practical outcome, these sessions are a valuable learning experience for the group.
Finally, Data Sci-Fris are also a great time to gather data science minded people, from new grad Software Engineers, to Customer Success Engineers, to Senior Data Scientists, who normally work across many different projects, to tackle problems together. This not only spurs critical thinking and creative new takes on previously siloed problems but also accelerates team building. This collaboration has made a big cultural impact — here are some of my teammates’ thoughts on the program’s effect:
“Data Sci-Fris have been an awesome and inspirational step for everyone to engage in personal technical development”
“I learn oodles and oodles on Data Sci-Fri!”
“Data Sci-Fri is one of the best parts of the week for a lot of people.”
“Without Data Sci-Fri, I would Data Cry-Fri”
If you want to learn new data science techniques and apply them to real problems and real data, we are hiring. View open positions here. Come learn with us!