aspire.samplers.mcmc

Attributes

logger

Classes

MCMCSampler	Base class for MCMC samplers.
NumpyMCMCSampler	MCMCSampler that maps samples and log probabilities to NumPy arrays for
Emcee	MCMCSampler that maps samples and log probabilities to NumPy arrays for
MiniPCN	Base class for MCMC samplers.
ParallelTemperedMCMCSampler	Wrapper for Parallel Tempered MCMC Samplers

Module Contents

aspire.samplers.mcmc.logger

class aspire.samplers.mcmc.MCMCSampler(log_likelihood, log_prior, dims, prior_flow, xp, dtype=None, parameters=None, preconditioning_transform=None, rng=None)

Bases: aspire.samplers.base.Sampler

Base class for MCMC samplers.

chain_checkpoint_path = 'checkpoint'

Path within checkpoint file to save MCMC chain checkpoints.

The default is “checkpoint”.

chain_dataset_name = 'mcmc_chain'

Name of chain entry within checkpoint file to save MCMC checkpoints.

rng

draw_initial_samples(n_samples)

Draw initial samples from the prior flow.

Parameters:

n_samples (int) – The number of samples to draw.

Returns:

The drawn samples, with log probabilities, log prior, and log likelihood.

Return type:

Samples

log_prob(z)

Compute the log probability of the samples.

Input samples are in the transformed space.

default_mcmc_chain_file_checkpoint_callback(file_path)

Return a callback that saves MCMC checkpoints as native HDF5 samples.

Parameters:

file_path (str | None)

Return type:

Callable[[dict], None]

checkpoint_mcmc_chain(samples, iteration=None, checkpoint_callback=None, checkpoint_every=None, checkpoint_file_path=None)

Save an MCMC chain checkpoint.

Parameters:

• samples (aspire.samples.Samples)

• iteration (int | None)

• checkpoint_callback (Callable[[dict], None] | None)

• checkpoint_every (int | None)

• checkpoint_file_path (str | None)

Return type:

None

class aspire.samplers.mcmc.NumpyMCMCSampler(log_likelihood, log_prior, dims, prior_flow, xp, dtype=None, parameters=None, preconditioning_transform=None, rng=None)

Bases: MCMCSampler

MCMCSampler that maps samples and log probabilities to NumPy arrays for compatibility with numpy-only samplers

log_prob(z)

Compute the log probability of the samples.

Input samples are in the transformed space.

class aspire.samplers.mcmc.Emcee(log_likelihood, log_prior, dims, prior_flow, xp, dtype=None, parameters=None, preconditioning_transform=None, rng=None)

Bases: NumpyMCMCSampler

MCMCSampler that maps samples and log probabilities to NumPy arrays for compatibility with numpy-only samplers

sample(n_samples=None, nwalkers=None, nsteps=500, rng=None, discard=0, checkpoint_callback=None, checkpoint_every=None, checkpoint_file_path=None, **kwargs)

Parameters:

• n_samples (int)

• nwalkers (int)

• nsteps (int)

• checkpoint_callback (Callable[[dict], None] | None)

• checkpoint_every (int | None)

• checkpoint_file_path (str | None)

Return type:

aspire.samples.Samples

class aspire.samplers.mcmc.MiniPCN(log_likelihood, log_prior, dims, prior_flow, xp, dtype=None, parameters=None, preconditioning_transform=None, rng=None)

Bases: MCMCSampler

Base class for MCMC samplers.

sample(n_samples=None, n_walkers=None, rng=None, target_acceptance_rate=0.234, n_steps=100, thin=1, burnin=0, last_step_only=False, step_fn='tpcn', checkpoint_callback=None, checkpoint_every=None, checkpoint_file_path=None)

Parameters:

• n_samples (int | None)

• n_walkers (int | None)

• checkpoint_callback (Callable[[dict], None] | None)

• checkpoint_every (int | None)

• checkpoint_file_path (str | None)

class aspire.samplers.mcmc.ParallelTemperedMCMCSampler(log_likelihood, log_prior, dims, prior_flow, xp, dtype=None, parameters=None, preconditioning_transform=None, rng=None)

Bases: MCMCSampler

Wrapper for Parallel Tempered MCMC Samplers

log_likelihood_wrapper(z)

Wrapper for log-likelihood that takes array inputs.

log_prior_wrapper(z)

Wrapper for log-prior that takes array inputs.