jesterTOV.inference.samplers.eos_reweighting.resample_eos_posterior

jesterTOV.inference.samplers.eos_reweighting.resample_eos_posterior#

resample_eos_posterior(
masses,
lambdas,
radii,
posterior_weights,
n_samples=None,
seed=0,
)[source]#

Resample a tabulated EOS set into equal-weight posterior samples.

This is a small, standalone utility that turns the discrete weighted posterior produced by EOSReweightingSampler.sample() (one weight per input EOS curve) into an equal-weight set of posterior draws, by drawing curves with replacement according to posterior_weights. It only needs the gridded curves and weights, so it works equally well on a fresh set of EOS curves that were reweighted outside of EOSReweightingSampler (e.g. with a custom likelihood), as long as masses/lambdas/radii share a common mass grid — see EOSReweightingSampler.load_and_grid() for a helper that builds such a grid from a list of NPZ files.

Parameters:
  • masses (Float[Array, 'N L'] | Float[ndarray, 'N L']) – Common-grid EOS curves of shape [N, L], e.g. as returned by EOSReweightingSampler.load_and_grid().

  • lambdas (Float[Array, 'N L'] | Float[ndarray, 'N L']) – Common-grid EOS curves of shape [N, L], e.g. as returned by EOSReweightingSampler.load_and_grid().

  • radii (Float[Array, 'N L'] | Float[ndarray, 'N L']) – Common-grid EOS curves of shape [N, L], e.g. as returned by EOSReweightingSampler.load_and_grid().

  • posterior_weights (Float[Array, 'N'] | Float[ndarray, 'N']) – Normalised posterior weight per EOS (sums to 1), e.g. sample_output.metadata["evidence"]["posterior_weights"].

  • n_samples (int | None) – Number of samples to draw. None (default) uses the Kish effective sample size \(N_\mathrm{eff} = 1/\sum_a (w^{(a)})^2\), rounded to the nearest integer.

  • seed (int) – Seed for the NumPy random generator used to draw the resample.

Return type:

dict[str, ndarray]

Returns:

dict –

  • eos_index : index into the input arrays for each draw, shape [n_samples]

  • masses_EOS, Lambdas_EOS, radii_EOS : resampled curves, shape [n_samples, L]