Skip to content

pspso

PSPSO

PSPSO 1.0 is a local hyperparameter-optimization and experiment framework for tabular machine learning. It provides a typed notebook API, PSO, grid and random search, durable experiment tracking, a CLI, and a FastAPI + React dashboard.

Major update in progress

PSPSO 1.0 adds a new dashboard, CLI, experiment service, data preparation workflow, and reproducible result tracking. It remains under active development until the updated package is published to PyPI.

See PSPSO in Action

The overview follows a Breast Cancer classification experiment from local installation and data setup through five-fold evaluation, model search, and the saved result. It uses the current development interface; details may change as the major update progresses.

The current dashboard is designed for local/private experimentation:

  • choose a built-in, imported, or versioned CSV dataset;
  • select a task, metric, estimator, fixed training parameters, and tunable hyperparameters;
  • validate the configuration before a run starts;
  • stream live training and validation progress from the backend to the frontend;
  • inspect trials, failures, best parameters, and final results.

Version 1.0 has one modern contract across Python, CLI, REST, workers, and the dashboard. Start with the notebook helper for direct exploration or use tracked runs when history and artifacts matter.

First Commands

Install and start PSPSO from PyPI with:

uv tool install pspso
pspso-dashboard

python -m pip install --upgrade pspso
pspso-dashboard

The uv command installs PSPSO as a PyPI tool; it does not create a new project. Use uv add pspso only when importing the Python API from an existing uv project. Open the dashboard, or run pspso --help. The package includes the interface. Contributors can use source setup to work from the repository.

Where To Go Next