learn.probabilistic.init_likelihood

learn.probabilistic.init_likelihood#

init_likelihood(energy_fn_template, nbrs_init, energy_scale=1.0, force_scale=1.0, virial_scale=1.0, virial_fn=None, distribution=<function logpdf>)[source]#

Returns the likelihood function for Bayesian potential optimization based on a force-matching formulation.

The scales of the likelihood components are normalized to be on same scale as energy_params.

Parameters:

  • energy_fn_template – Energy function template

  • nbrs_init – Neighbor list to be used for initialization

  • energy_scale – Prior scale of energy data.

  • force_scale – Prior scale of force components.

  • virial_scale – Prior scale of virial components.

  • virial_fn – Virial function compatible ẃith virial data type, e.g. initialized via force_matching.init_virial_fn.

  • distribution – Likelihood distribution. Defaults to a Gaussian logpdf, but any jax.scipy logpdf with the same signature can be provided.

Contents