Skip to content

Experiments (DVC Integration)

DataChain Studio provides comprehensive ML experiment tracking through DVC integration, allowing you to track, compare, and manage your machine learning experiments with Git-based versioning.

Overview

DataChain Studio integrates with DVC (Data Version Control) to provide a powerful web-based interface for managing your ML experiments. By connecting your Git repositories, you can visualize experiment results, compare different runs, and collaborate with your teamβ€”all without leaving your browser.

Key Features

Project Management

  • Create projects by connecting to GitHub, GitLab, or Bitbucket repositories
  • Configure project settings including project directory, data remotes, and column tracking
  • Support for monorepos with multiple ML projects in different sub-directories
  • Share projects with your team or make them publicly accessible

Experiment Tracking

Visualization and Comparison - Visualize and compare experiments using plots, images, and trend charts - Display plots generated by DVCLive including AUC curves, loss functions, and confusion matrices - Compare up to seven experiments side by side - Generate trend charts to see how metrics changed over time - Export project data to CSV for external analysis

Collaboration - Create teams with multiple collaborators - Share projects within a team or publicly on the web - Track experiments from different branches and commits - Review and manage experiments through pull/merge requests

Getting Started

  1. Create a project - Connect your Git repository to DataChain Studio
  2. Configure your project - Set up data remotes, credentials, and tracking preferences
  3. Run experiments - Start tracking your ML experiments with DVC and DVCLive
  4. Explore and visualize - Analyze your results in the project table and plots
  5. Share your work - Collaborate with your team or share publicly