At Two Sigma Ventures, we take pride in our ongoing commitment to helping portfolio companies by leveraging the network and resources that come with Two Sigma’s two decades of experience and brand that is synonymous with data & software expertise. Two Sigma has been building a world class technical organization for 20+ years, and today calls 200+ PhDs and 1000+ data scientists and engineers employees, with expertise ranging from distributed computing, computer vision, natural language processing, simulation and much more.
One of the ways we harness the prowess of Two Sigma’s team to assist our portfolio companies is through hackathons. The first hackathon took place a few years ago where over the course of just a few days, Two Sigma data scientists were able to help WHOOP* gain insights that led to a 30% increase in the accuracy of their machine learning (ML) algorithms that turn heart rate variability (HRV) data into insights about strain and recovery of the body. Last year, we ran a hackathon that enabled our portfolio company Nelo,* a Mexico City based fintech company, to make data-driven decisions around the personalization of their app while optimizing for profitability.
This fall, we partnered with TSV portfolio company Radar.* Radar, co-founded by Nick Patrick and Coby Berman and based in NYC, provides location infrastructure for mobile and web applications. To date, they have secured $85 million in funding (including an $8M Series A financing in 2019, which TSV led). Radar’s platform powers location-based app experiences for startups and enterprises across hundreds of millions of devices worldwide. From logistics companies using Radar for real-time tracking to retail companies improving in-store app experiences, Radar aims to play a crucial role in their success.
The Challenge: Improving ETA Predictions
Radar’s Chief Technology Officer, Tim Julien, presented a challenge to the Two Sigma Ventures team: to enhance their algorithm that generates Estimated Time of Arrival (ETA) estimates. Radar employs a static predictive model that considers “first-order” factors such as the route to the destination and speed limits. The objective was to improve this model by incorporating additional features that the Radar team hadn’t yet been able to explore.
Over four days, a group of 17 Two Sigma data scientists and engineers came together to tackle Radar’s data challenge. Many of the hackathon participants had spent time in relevant areas, such as geospatial data engineering, in their current or a previous role. They were divided into teams, each tasked with creating a predictive model that could generate more accurate ETAs. The performance of these algorithms was evaluated based on how much they reduced the difference between Radar’s initial estimate and the actual trip length.
Innovations That Made the Difference
Each team approached the project with unique strategies, such as integrating 3rd-party location-based data including school proximity to the destination and population density metrics, experimenting with neural networks, and fitting to various loss functions that better balanced the improvement to short and long trips. The winning team distinctly enriched the data by incorporating nearest city and county-level population density, deriving new time features like timezone, hour of the day, and minute of the hour, and utilizing two distinct geographical hierarchical hashing techniques. These diverse approaches contributed to significant innovations in addressing the problem.
The Outcome: A Significant Improvement
At the end of the hackathon, each team presented its findings and technical approaches to Radar’s data science team. The combined efforts of this talented cohort compressed what could easily have taken months into a mere four-day span, and the mix of these new, diverse perspectives proved invaluable. The Two Sigma data scientists and engineers succeeded in enhancing Radar’s predictive model for ETAs, reducing the total error of the predictions for this dataset by 24%.
To share thoughts from Tim (CTO, Radar):
“We’ve always had a ton of respect for the Two Sigma Engineering team, but we were blown away by the focus and curiosity they brought to this hackathon. One of Radar’s core values is Raise The Bar, and the Two Sigma team certainly did that. They pushed our thinking in new ways we hadn’t considered and the results speak for themselves.”
At Two Sigma Ventures, we aim to support visionary founders who are at the forefront of the most significant generational shift: the transformative impact of data and software on our world. These innovative founders and their companies, like Radar, will almost always generate substantial amounts of data, providing us the potential to add real value to their businesses through Two Sigma brainpower in software engineering, data science, quantitative modeling, machine learning and more. We believe that together, we can solve problems and seize opportunities in this new age of data.