Comet.ml is known for helping developers and data scientists in overseeing, comparing and optimizing their machine learning models. The New York-based firm launched its product recently upon culminating the TechStars-driven Amazon Alexa Accelerator program as well as raising a $2.3M seed round.
Trilogy Equity Partners in collaboration with Two Sigma Ventures, Fathom Capital, TechStars Ventures, Founders Co-op and angel investors led the exercise.
The new service by Comet.ml offers you a dashboard that combines the code of your ML experiments and their outcomes. It also enables you to improve your models through tweaking your experiments’ hyperparameters.
In fact, Comet tracks the outcomes and avails a graph of your results as you continue training your machine-learning model. Furthermore, it not only tracks your code changes but also imports them in a bid to help you compare the varying aspects of the different versions of your experiments.
Irrespective of whether they utilize Scikit Learn, Keras API, Pytorch, TensorFlow or simply write a Java code, developers can easily incorporate Comet into their ML frameworks.
When starting, developers only need to integrate the CometML tracking code to all their applications before running their experiments.
Since the service is agnostic regarding where you train your models, you can allow the rest of the team to access your results.
In essence, this means that data scientists can stick with their current development tools and workflow. Aside from that, they have a new tool that provides them insights into the excellent performance of their experiments.
According to Gideon Mendels, the chief executive officer and co-founder of CometML, the company realized that machine-learning teams look much similar to how software teams looked ten or fifteen years back.
Whereas software teams currently have version control and tools such as GitHub for sharing their code, ML teams still share their code and data via email.
Mendels added that the core problem is not discipline but rather the state of tooling. Although current tools such as GitHub are an excellent choice for software engineering, they do not fully meet the needs of machine learning teams. This situation is true even though code serves as a core component of machine learning.
When closing its beta, Gideon Mendels said that Comet signed up nearly 500 data scientists including some from the leading top tire technology companies in the world. As a testament to their expertise, they have created approximately 6,000 models on the platform.
Looking forward, the team at CometML intends to provide developers with additional tools that would help them in building more accurate and better models.
However, Gideon Mendels, the company’s chief executive officer noted that the company has to get this initial building block ready before continuing with its plans.
Currently, CometML is available to all developers who may want to test it on their applications. In addition, they have two choices including the free tier that can handle unlimited public projects. On the other hand, developers or teams can use paid tiers that are similar to tools like GitHub if they desire to preserve the privacy of their project.