Installation¶
EVALSIG ships as a single PyPI package. The core needs only NumPy, SciPy, and PyArrow.
Requirements¶
- Python 3.10 or newer
- macOS, Linux, or Windows
- About 80 MB of disk for the wheel and its dependencies
Install from source¶
The package is not yet published to PyPI. Install the source checkout in editable mode:
git clone https://github.com/vtensor/evalsig.git
cd evalsig
python -m venv .venv
source .venv/bin/activate
pip install -e .
That gives you everything needed to compare runs, gate releases, write to a local Parquet store, and use the pytest plugin.
The repo also includes a research validation script:
It runs four small Monte Carlo experiments that confirm the library does what the design doc promises (see Methodology for the full story).
Optional extras¶
| Extra | When to install | Command |
|---|---|---|
docs |
Building the documentation site locally | pip install -e ".[docs]" |
dev |
Running the test suite and linters | pip install -e ".[dev]" |
braintrust |
Publishing comparison results to Braintrust | pip install -e ".[braintrust]" |
Verify the install¶
After installing, the CLI should be on your $PATH:
And the Python import should succeed without any optional dependencies:
If either fails, double-check that you're using the same Python interpreter
in your shell and your editor. A common gotcha is having pip install into a
different environment than the one python runs.
Next steps¶
- Quickstart: the 30-second walkthrough.
- Your first comparison: a longer, hand-held tutorial.
- Configuration: every knob the library exposes.