Listen to this story
Last week, one of the leading MLOps platforms ‘Comet’ introduced its new product—Kangas—the first open-source tool for democratising real-time data exploration and analysis for the ML community and computer vision.
Click here to try out Kangas.
Founded in 2017, the New-York headquartered company was established by former Googler Gideon Mendels. Comet provides data scientists and ML teams an MLOps platform to optimise, manage, and accelerate development processes—from training quick runs to monitoring models—across the ML lifecycle. The platform caters to over 150+ enterprise customers such as Cepsa, Affirm, Uber, Zappos and Etsy. Moreover, Comet is free to academic teams and individuals.
As industries have witnessed rapid digitisation—around 97.2% of the companies claim the adoption of data science and analytics in some shape—which has resulted in a growing interest among investors and technopreneurs.
Moreover, a Deloitte report states that the MLOps market is expected to reach USD 4 billion by 2025. Well-known MLOps players include Qwak, ZenML, Comet, Domino Data Lab, Weights and Biases, among others.
So, how is Kangas any different?
Comet teams said that the open-source data exploration tool, Kangas, is based on a cutting-edge technology to help users understand and debug their data in a highly intuitive way.
With the help of Kangas, visualisations are generated in real time. This enables every machine learning practitioner to sort, filter, group, query and interpret structured and unstructured data. This would further let the users derive meaningful information and accelerate better model development.
Comet CEO Gideon Mendels said that the key component of data-centric machine learning is to understand how the user’s training data impacts the results of their models—when model predictions are wrong. He further elaborated that the new tool would successfully accomplish these goals to improve the overall experience.
Considering the unique needs of ML practitioners, Kangas is a scalable and interoperable tool that enables users to discover patterns that are buried deep within the clusters of datasets. Data scientists will be able to query their large-scale datasets in a natural manner, allowing them to interact with the data in novel ways.
Some of the noteworthy benefits of Kangas are unparalleled scalability to handle large datasets with very high performance, purpose built with ML concepts such as scoring, bounding boxes, and auto-generated statistics. Kangas supports different forms of media—not limited to traditional text queries—such as images, and video. Moreover, it can also run as a standalone local app, in a notebook, or can even be deployed as a web app.