A business looking to make use of machine learningOpens a New Window.(ML) needs more than smart devices and reams of data. At its core, ML revolves around two hemispheres: ML models and algorithms on one side and appropriately curated data sets on the other. While both require expertise to create, the former just got a significant boost via Comet.mlOpens a New Window., a service launched earlier this month with tools to allow data scientists and developers to track code and share their ML models more efficiently. The company says it’s answering what it sees as an increased need for more effective and usable ML tools. The service is part of a growing field of convenient services that seek to let more people access, use, and learn about ML.
The GitHub Connection
Despite being less than one month old, describing Comet.ml as “the GitHub of ML” may not be inappropriate. If you’re unfamiliar with GitHubOpens a New Window., it’s a repository hosting service where developers store and share their code. In projects with multiple developers working on the same codebase, repositories such as GitHub play a critical code in organizing workflows and maintaining version control. While the concept of a code repository isn’t new, GitHub opened up a whole new world to the development community by creating a user interface (UI) that went beyond arcane, project-oriented coding capabilities and added an intuitive UI as well as social tools that allow GitHub to talk to users and even communities. Whether you wanted your code reviewed by other developers, find new and interesting applications, or were just curious about what the world’s top engineersOpens a New Window. were working on, GitHub has become one of the most popular places to catch up on what the development community is doing.
With that kind of resume, wanting to be the GitHub of anything seems extremely ambitious, but the founders of Comet.ml are confident. Comet.ml works in a similar way to the popular GitHub service. Simply make a free account on the Comet.ml website, pick your preferred ML library (Comet.ml currently supports Java, PytorchOpens a New Window., TensorFlowOpens a New Window., and several more of the most popular libraries), and you can get up and running building and testing ML models almost instantly—and likely more easily than you’ve been able to do up to this point. This is because Comet.ml also tracks all changes that a team makes to a repository on the website. It offers automated model optimization and you can even integrate your Comet.ml work with GitHub for larger projects.
GitHub also hosts ML models but Comet.ml is designed with the unique needs of ML in mind. Through a type of algorithm known as BayesianOpens a New Window. “Hyperparameter optimizationOpens a New Window.,” the service will tweak your models by changing the hyperparameters of your experiments. If you’re a true data geek, then there’s a more thorough explanation of this on the company’s website. Tweaking models manually can take an incredibly long time. If this algorithm works as well as Comet.ml says it does, then it could definitely get the attention of the data science community. Just like GitHub, one account with publically available repositories is completely free of charge, with private repositories starting at $49 per user per month.
The Need for Something Simpler
Gideon Mendels, co-founder and CEO of Comet.ml, is something of an ML veteran. He has worked in research for Columbia University and at GoogleOpens a New Window.. Throughout his career, he has struggled to find an effective way to test and share ML models.
“I previously worked at a company called GroupWize , and we had about 15 machine learning models in production,” said Mendels. “It was just impossible to keep track of all the changes in them. So, we actually started building Comet internally as a homebrew solution for our pain.”
From there, Mendels and other team members decided to focus on building out Comet.ml on its own. To Mendels, the value of Comet.ml isn’t just the fact that ML models can be stored in the cloud; it’s about making it easier to experiment with that code. Mendels was also quick to dismiss the notion that his service is trying to compete with GitHub. After all, it integrates with the service and users can sign up with their GitHub log-in credentials. For Mendels, it’s really about answering a growing wave of data democratization with better functionality.
“It connects to a bigger point of how a lot of companies are starting to do ML and data science,” Mendels said. “With GitHub, you can store code, but with ML, code is just one piece of the puzzle. What data was used to fit in that code?” Mendels says that the automated tweaking features will help Comet.ml stand apart on its own.
Machine Learning Playgrounds
Comet.ml is just one of several offerings that aim to change the way we interact with ML. Microsoft, which has been very aggressive in the space, launched Azure NotebooksOpens a New Window. a few years ago. Although the company presents it as more of an educational tool than Comet.ml, it is also designed to let you play around with ML models in the cloud.
There is also a whole wave of ML marketplaces available that offer complete, ready-to-go models for both small to midsize businesses (SMBs) and enterprises alike. AlgorithmiaOpens a New Window. is an Opens a New Window.artificial intelligenceOpens a New Window. (AI) marketplace that offers, among other things, ML models that you can buy and use in your own apps via an application programming interface (API) call. Don’t have the skill or time to build a sentence-parsing model? Then use Parsey McParsefaceOpens a New Window. for the low priceOpens a New Window. of $28.54 for 10,000 API calls. Less creatively named models on the marketplace include those for facial recognition algorithms, spectral clustering for geographic data, and text extraction.
If you’re not a data scientist, then you may be thinking that these services are not applicable to you and your organization. But businesses of all sizes are announcing unprecedented support and utilization of AI solutions, and ML is an important part of that. These implementations are spanning the gamut from broad, sweeping projects down to those so targeted that you’re surprised to find ML is part of the recipe.
As an example of a targeted project, WineSteinOpens a New Window. is a digital sommelier service that uses ML models to pair wine with different kinds of food. Broader implementation examples span financial technology (fintech) , healthcare tech, and even chatbotsOpens a New Window. where AI and ML have already changed the way most every business approaches customer service and Opens a New Window.helpdeskOpens a New Window.operations. The user base for AI and ML is growing fast and will leave no business untouched, which makes the future a bright place for up-and-comers such as Comet.ml.