System1 brings leadership in data science & engineering

Our data science and engineering teams come from organizations that lead the world in technology, such as Google, Amazon, and MIT. Our team produces groundbreaking research, some of which is below.

Using Bandit Algorithms on Changing Reward Rates

One of the problems we have at System1 is updating our estimate of a feature’s performance over time. Even if our initial estimate is correct, the performance of the feature could change at a future point.

View Paper (PDF)
Using Bandit Algorithms on Changing Reward Rates

Data Munging with Pandas

Data munging skills are the keys to the data analysis kingdom. In this talk, John Fries covers fundamental data munging techniques using the Pandas library and presents some data munging brain teasers.

View on YouTube
Data Munging with Pandas

Interactive Data Exploration and Visualization in IPython

Visualization allows for insightful exploration of data. Tamara Knutsen presents the different visualization and plotting Python libraries for interactive data exploration in IPython notebook, covering Matplotlib, Seaborn, mpld3, Bokeh, VisPy and more.

View on YouTube
Interactive Data Exploration and Visualization in IPython

Going Bayesian in Ad Tech (Two Case Studies)

In online advertising, our data is fast and sparse, which can undermine traditional analyses. This session presents two applications of Bayesian methods: the first case is detecting changes in click-through-rate (CTR) in a series of web traffic from end users; the second case is estimating market depth using series of ad network data. In both cases, going Bayesian allowed us to use lower-level data directly to answer high-level strategic questions.

View on YouTube
Going Bayesian in Ad Tech (Two Case Studies)

Scaling Postgres With Some Help from Redis

As we push our servers to support more users, databases too must scale. Scaling Postgres vertically and horizontally works well for most data, but what if you really need to perform 100k writes per second? By using an in-memory data structure server called Redis along with Postgres, we will discuss playing to the strengths of each database to help scale and simplify your operations.

View on YouTube
Scaling Postgres With Some Help from Redis

Extracting Revenue Impact from Business Metrics

When I joined System1, I wanted to make an impact as quickly as possible. The first change I pushed into our system was designed to optimize our click-thru rate (CTR) model. I was hoping to see a clear uptick in user engagement, but instead I saw something similar to the graph below (Fig. 1)

View Paper (PDF)
Extracting Revenue Impact from Business Metrics