April 20, 2022
This article, written by Angelica Lo Duca, first appeared on Heartbeat. For two months now I have…
When you are building and training ML models, tracking, reproducibility, and collaboration can be challenging. And when you’re doing deep learning experiment management, you can multiply that challenge.
It’s pain being experienced by a growing number of data scientists who are building ML: In a survey of more than 500 machine learning practitioners, respondents cited challenges related to people, process, and tools that make it difficult to track model training runs and results, collaborate with colleagues, and iterate fast during the complex process of developing machine learning. This friction can cause delays in ML development that slow down or halt model deployment to production.
Comet’s latest product update aims to reduce that friction so you can get to the good part: Building better ML models faster.
As your team and your machine learning operations scale, it becomes more challenging to track and version datasets and to link dataset versions to the models they were used to create. You may want to analyze more granular aspects of the data, such as comparing different subsets of data, or perhaps explore how your data from a year ago is different from your data today, and how the data differs by model.
Comet Artifacts Lineage provides an interactive way for you and your team to explore and compare datasets so you can better understand and visualize datasets you are using for machine learning. It also allows you to collaborate and report on your ML projects for easier reproducibility and, ultimately, more confidence in the performance of your models in production.
Artifacts Lineage is available within the different versions of your Artifacts. To view the lineage of a dataset, select Artifacts, choose a dataset, and select Lineage. You will be able to view your Artifact assets, metadata, and lineage. Lineage makes it easier to reproduce models by allowing you to visualize, in an interactive way, how your model was created, how your data was used, and everything in between. Explore the entire experiment graph to find the right dataset and the right model.
In your free Comet account, you can access a demo project and see Artifacts Lineage in action.