Ground → your workflow
Use a Ground export in Python or a notebook
Continue from Ground’s completed export into Jupyter, Colab or VS Code without generating another dataset.
Start from your dataset’s export controls. Guides are optional; you do not need to visit this page to download your data.
From Ground to your destination
- Generate your data once. From the dataset page, choose Open in Colab, or Export options → Download notebook. These actions reuse the most recent completed export and remain available after a reload.
- Open in Colab asks Google for drive.file authorization on first use, creates a private notebook in your own Drive and opens it in Colab. The export loads automatically: choose Runtime → Run all for quality checks, charts and a baseline. Google may ask you to choose an account again. No Ground API key is needed for this notebook. Cloud data links expire after seven days; small local exports are embedded (up to 2 MB). Uploaded datasets still require your original real file for held-back evaluation.
- For your own notebook, install the Python client from this release’s source checkout into its kernel environment. Public registry distribution is pending. Python 3.9+ is required; SDK use needs a plan with API-key access.
- Use an API key from Account settings. Keep it in the environment or enter it with getpass; do not save it in a notebook cell. Copy the export-specific snippet from Ground and run it.
- A single table becomes a DataFrame; a ZIP of related tables becomes a dictionary of DataFrames. Check counts, identifiers and nulls before using it. The generated evaluation notebook currently supports single tables only.
Optional: use the SDK or CLI
Run this from the Ground source checkout for this release. Package distribution is pending. Keep credentials outside code and notebooks.
python -m pip install -e './sdk/python[pandas]'
Code workflow
import os
from getpass import getpass
from ground import Ground
os.environ["GROUND_URL"] = "https://useground.dev"
os.environ["GROUND_API_KEY"] = getpass("Ground API key: ")
job = Ground().job("YOUR_EXPORT_ID")
data = job.to_pandas()
if isinstance(data, dict):
print({name: len(frame) for name, frame in data.items()})
else:
print(data.head())
print(data.dtypes)Verify the result
The notebook must execute, load the selected export and report counts. For learned data, follow its real held-out evaluation and source-file instructions. Designed data supports synthetic-only exploration; scores on it do not establish real-world utility. The Python client uses the saved schema to preserve text identifiers and nullable types. CSV empty cells become null; use JSONL/Parquet to preserve empty-string distinctions.
When something needs attention
401: check the key and host. 403: check API entitlement. Export not ready: poll ground status EXPORT_ID; do not generate again. Missing pandas/pyarrow: install the extra in the active notebook kernel. Expired notebook link: download a fresh notebook from Ground. Missing real source: use your retained original; do not silently substitute synthetic data for the holdout.