psiphy — Package for Statistical Inference of Physics

A Python package for cosmological parameter inference, providing tools for simulation-based inference (SBI), MCMC sampling, Fisher forecasting, and posterior diagnostics.

What's inside

Quick example:

from psiphy.mcmc import MHSampler
from psiphy.utils.prior import UniformPrior
from psiphy.toy_models import NoisyLine

line = NoisyLine()
y_obs = line.simulate([-0.8, 5.0])
prior = UniformPrior({'slope': (-3, 3), 'intercept': (0, 10)})

sampler = MHSampler(lambda t: prior.log_prob(t) + line.log_prob(t, y_obs), n_walkers=32,
                    param_names=prior.names)
sampler.run(prior.sample(32), n_steps=2000)
sampler.summary()

Tutorials

Notebooks in examples/, also rendered in the documentation:

• MCMC sampling and priors
• Nested sampling and Bayesian evidence
• Importance sampling and SMC (includes a calibration check over many data sets)
• Likelihood-free inference: rejection ABC, BOLFI, LFIRE
• Fisher forecasting
• Plotting diagnostics
• Toy models

Installation

From PyPI:

pip install psiphy

Optional dependencies are grouped as extras: mcmc (emcee, dynesty, nautilus), sbi (sbi, torch), hmc (JAX/blackjax, for the planned HMC sampler):

pip install "psiphy[mcmc]"

The latest development version from GitHub:

pip install git+https://github.com/sambit-giri/psiphy.git

For a local editable install (recommended for development):

git clone https://github.com/sambit-giri/psiphy.git
cd psiphy
pip install -e ".[dev]"

Running the tests

The test suite uses pytest. After installing with the dev extras:

pytest tests/ -v

Tests cover every module, checking samplers against analytic posteriors and evidences where these are known. New tests are added alongside each new module.

Contributing and feedback

Bug reports, feature requests, and questions are welcome — please open an issue on the GitHub issue tracker. Pull requests are also encouraged.