Hear from CIOs, CTOs, and other C-level and senior execs on data and AI strategies at the Future of Work Summit this January 12, 2022. Learn more
Lyft has spent a lot of time building tools that help it simulate the results of machine learning algorithms, and the company plans to make these tools more broadly available in the near future.
The company’s simulators are a key part of its machine learning development process, according to Erbil Karaman, the head of product for Lyft’s marketplace team. When Lyft first starts testing machine learning models, they get run through simulators to figure out their early impact.
“That does model emulation and looks at what are called hyper parameters and some of the outcomes that we care about and tries to tell us early signals for what this model can do differently from other models,” Karaman said during a panel at VentureBeat’s MB 2017 conference in San Francisco.
Machine learning touches several key parts of the Lyft experience. The company’s routing algorithm, pricing algorithm, driver-matching algorithm, and other systems all rely on machine learning models to one degree or another. Making the tools available to other organizations could help accelerate other companies’ development of machine learning models.
Testing and iterating on machine learning models can be time-consuming, especially once companies try to deploy them in production systems to gather better data. Lyft uses its simulation tools as a way to pre-test models, because it has a limited ability to run experiments on the live version of its service.
The ride-hailing company is currently in the process of making those internal systems ready for public scrutiny, Karaman said.
“We’ll start with some academia outreach in the next coming months, because we think this will be very relevant, this simulator to understand early signals, or at least outliers,” he said.
Lyft’s move is part of a greater trend among tech companies to open-source their internal tools for performing machine learning work. Microsoft, Google, Facebook, Amazon and others have all made substantial investments in releasing key parts of their machine learning stacks to the public through open source software.
Lyft has already released other tools to the public as open source projects, including an edge and service proxy called Envoy.