Comet, provider of the leading MLOps platform for machine learning (ML) teams from startup to enterprise, today announced a bold new product: Kangas. Open sourced to democratize large scale visual dataset exploration and analysis for the computer vision and machine learning community, Kangas helps users understand and debug their data in a new and highly intuitive way. With Kangas, visualizations are generated in real time; enabling ML practitioners to group, sort, filter, query and interpret their structured and unstructured data to derive meaningful information and accelerate model development.
Data scientists often need to analyze large scale datasets both during the data preparation stage and model training, which can be overwhelming and time-consuming, especially when working on large scale datasets. Kangas makes it possible to intuitively explore, debug and analyze data in real time to quickly gain insights, leading to better, faster decisions. With Kangas, users are able to transform datasets of any scale into clear visualizations.
“A key component of data-centric Machine Learning is being able to understand how your training data impacts model results and where your model predictions are wrong. “Kangas accomplishes both of these goals and dramatically improves the experience for ML practitioners.”
Gideon Mendels, CEO and co-founder of Comet
Putting Large Scale Machine Learning Dataset Analysis at Your Fingertips
Developed with the unique needs of ML practitioners in mind, Kangas is a scalable, dynamic and interoperable tool that allows for the discovery of patterns buried deep within oceans of datasets. With Kangas, data scientists can query their large-scale datasets in a manner that is natural to their problem, allowing them to interact and engage with their data in novel ways.
Noteworthy benefits of Kangas include:
Unparalleled Scalability: Kangas was developed to handle large datasets with high performance.
Purpose Built: Computer Vision/ML concepts like scoring, bounding boxes and more are supported out-of-the-box, and statistics/charts are generated automatically.
Support for Different Forms of Media: Kangas is not limited to traditional text queries. It also supports images, videos and more.
Interoperability: Kangas can run in a notebook, as a standalone local app or even deployed as a web app. It ingests data in a simple format that makes it easy to work with whatever tooling data scientists already use.
Open Source: Kangas is 100% open source and is built by and for the ML community.
Kangas was designed for the entire community, to be embraced by students, researchers and the enterprise. As individuals and teams work to further their ML initiatives, they will be able to leverage the full benefits of Kangas. Being open source, all are able to contribute and further enhance it as well.
“Interoperability and flexibility are inherent in Comet’s value proposition, and Comet aims to expand on that value through open source contributions,” added Mendels. “Kangas is a continuation of all of our efforts, and we couldn’t wait to get its capabilities into the hands of as many data scientists, data engineers and ML engineers as possible. We believe by open sourcing it, Comet can help teams get the most out of their ML projects in ways that have not been possible previously.”
Kangas is available as an open source package for any type of use case. It will be available under Apache License 2 and is open to contributions from community members.
About Comet
Comet provides an MLOps platform that data scientists and machine learning teams use to manage, optimize, and accelerate the development process across the entire ML lifecycle, from training runs to monitoring models in production. Comet’s platform is trusted by over 150 enterprise customers including Affirm, Cepsa, Etsy, Uber and Zappos. Individuals and academic teams use Comet’s platform to advance research in their fields of study. Founded in 2017, Comet is headquartered in New York, NY with a remote workforce in nine countries on four continents. Comet is free to individuals and academic teams. Startup, team, and enterprise licensing is also available.