How to use PosteriorParameters in sbi

How to use PosteriorParameters in sbi#

The PosteriorParameters classes in SBI are dataclasses used to initialize parameters for different types of posterior algorithms, when using the build_posterior method. This guide shows how to configure and use PosteriorParameters in your workflow.

sbi currently implements the following PosteriorParameters subclasses. Each subclass of PosteriorParameters holds parameters for a specific posterior sampling method.

  • DirectPosteriorParameters: used with DirectPosterior

  • ImportanceSamplingPosteriorParameters: used with ImportanceSamplingPosterior

  • MCMCPosteriorParameters: used with MCMCPosterior

  • RejectionPosteriorParameters: used with RejectionPosterior

  • VectorFieldPosteriorParameters: : used with VectorFieldPosterior

  • VIPosteriorParameters: used with VIPosterior

Usage#

The example below uses MCMCPosteriorParameters for passing MCMC options to a posterior sampling via MCMC. You can follow a similar approach as the example below with your sbi workflow when using the other PosteriorParameters types.

import torch

from sbi.inference import NRE
from sbi.utils.torchutils import BoxUniform


def simulator(theta):
    return 1.0 + theta + torch.randn(theta.shape, device=theta.device) * 0.1

num_dim = 3
prior = BoxUniform(low=-2 * torch.ones(num_dim), high=2 * torch.ones(num_dim))
theta = prior.sample((300,))
x = simulator(theta)

inference = NRE(prior=prior)
inference.append_simulations(theta, x)
inference.train();

Create an instance of MCMCPosteriorParameters with desired values to configure MCMC sampling.

from sbi.inference.posteriors import MCMCPosteriorParameters

params = MCMCPosteriorParameters(
    method="slice_np_vectorized",
    thin=1,
    warmup_steps=100,
    num_chains=4,
    init_strategy="sir",
    init_strategy_parameters={"num_candidate_samples": 1000},
    num_workers=2,
    mp_context="spawn"
)

Now pass the instance to build_posterior() using the posterior_parameters argument.

inference.build_posterior(posterior_parameters=params)

Modifying Parameters with .with_param()#

You can create a new copy with updated fields without mutating the original.

new_params = params.with_param(num_chains=8, warmup_steps=200)

inference.build_posterior(posterior_parameters=new_params)
Navigation