Show HN: Extension for VS Code to track ML experiments
marketplace.visualstudio.com
marketplace.visualstudio.com
I used to work at Google where they have an internal tool called mldash. Was never able to find an equivalent user experience outside Google until now.
To be honest, I rarely need versioning for my datasets, and I had sort of put DVC into a bucket in my mind called "dataset versioning". For this reason, I have not really been a DVC user until now.
From what I see in this post, though, you guys are building a more cohesive experience around the common tasks that someone has to do in the course of training, deploying, and evaluating a machine learning model. Experiment tracking is a great place to start. Kudos.
Today, VS Code is the fastest growing IDE among data scientists. That’s why we decided to bring DVC’s functionality natively to VS Code without any infrastructure overhead. We spent the past year building the DVC extension and I’m glad to present it to the community today.
In a nutshell, what DVC provides right from VS Code: Experiment bookkeeping (an alternative to Tensorboard or MLFlow) that automatically saves metrics, graphs, hyperparameters and any other metadata. The feature allows you to compare the experiments and share with your teammates.
- Experiment reproducibility which allows you to pick any past experiment and use it as a foundation for a new set of experiments.
- Data management allows you to manage datasets, files, and models with data living in your favorite cloud storage: S3, Azure Blob, GCS, NFS, etc.
You can read here about the other tools that we’ve released: https://news.ycombinator.com/item?id=23759332
Watch this video to learn more about the extension https://youtu.be/LHi3SWGD9nc
If you use VS Code, please give it a try and let me know what you think. Any feedback is greatly appreciated https://marketplace.visualstudio.com/items?itemName=Iterativ...