Marin comes with a data browser that makes it easy to view datasets (in various formats) and experiments produced by the executor. It is basically a file browser that handles popular file formats like jsonl and parquet.
- Basic installation
- Run an experiment, either First Experiment or Executor 101.
The data browser uses configuration files to specify what paths are accessible:
conf/local.conf- For browsing local files (e.g.,../local_store)conf/gcp.conf- For browsing GCP Storage buckets (requires authentication)conf/docker.conf- For Docker deployment
Install dependencies:
cd data_browser
uv venv
uv pip install -e .
npm installNote: If you get ModuleNotFoundError when running the server, ensure dependencies are installed or run via uv:
DEV=true uv run python server.py --config conf/local.confThe data browser consists of two components that need to run simultaneously:
- Backend server (Flask) - Handles file access and API endpoints
- Frontend server (React) - Provides the web interface
Run both servers in separate terminals:
Terminal 1: Backend Server
cd data_browser
DEV=true uv run python server.py --config conf/local.confTerminal 2: Frontend Server
cd data_browser
npm startAccess: http://localhost:3000
If you only need to test the API endpoints or the React server won't start:
cd data_browser
DEV=true uv run python server.py --config conf/local.confAccess: http://localhost:5000/api/view?path=../local_store
root_paths:
- ../local_storeroot_paths:
- gs://marin-us-central2
- gs://marin-us-west4
# ... other buckets
blocked_paths: # Optional: paths to block access to
- gs://marin-us-central2/private-data/
max_lines: 100
max_size: 10000000Note: GCP configuration requires valid Google Cloud credentials (service account or gcloud auth).
If you get "Connection refused" errors, the React dev server may not be running properly. You can still:
- Use API directly: Access
http://localhost:5000/api/view?path=YOUR_PATH - Check React server: Ensure
npm startis running without errors - Port conflicts: Check if port 3000 is available
- For GCP buckets: Ensure you have valid Google Cloud credentials
GET /api/config- Returns server configurationGET /api/view?path=PATH&offset=0&count=5- Browse files and directoriesGET /api/download?path=PATH- Download files