nessai.proposal.flowproposal.truncation

nessai.proposal.flowproposal.truncation#

Truncation rules for flow proposals.

Classes#

`BaseTruncationRule`	Base class for truncation rules.
`LatentRadiusTruncation`	Filter latent samples using a radial threshold.
`MinLogQTruncation`	Truncate samples using the minimum live-point log q.
`LikelihoodThresholdTruncation`	Truncate samples using the current likelihood threshold.
`TruncationScheme`	Apply an ordered set of truncation rules.

Functions#

`get_deprecated_latent_radius_arguments`(→ list[str])	Return deprecated latent-radius arguments that were explicitly set.
`get_deprecated_latent_radius_kwargs`(→ dict)	Build sparse latent-radius kwargs from deprecated proposal arguments.
`normalise_truncation_methods`(→ list[str])	Normalise truncation-method input into an ordered unique list.
`should_enable_latent_radius`(→ bool)	Check if latent-radius truncation should be enabled from kwargs.
`build_truncation_methods`(→ list[str])	Build the effective truncation-method list from legacy and new inputs.
`apply_default_truncation_config`(methods[, ...])	Apply canonical default truncation configuration.
`normalise_truncation_kwargs`([truncation_method, ...])	Normalise truncation kwargs into the canonical method-keyed form.
`get_truncation_rule_class`(name)	Get the truncation rule class for a configured method name.

Module Contents#

nessai.proposal.flowproposal.truncation.get_deprecated_latent_radius_arguments(**kwargs) → list[str]#: Return deprecated latent-radius arguments that were explicitly set.

nessai.proposal.flowproposal.truncation.get_deprecated_latent_radius_kwargs(**kwargs) → dict#: Build sparse latent-radius kwargs from deprecated proposal arguments.

nessai.proposal.flowproposal.truncation.normalise_truncation_methods(truncation_method=None, truncation_methods=None) → list[str]#: Normalise truncation-method input into an ordered unique list.

nessai.proposal.flowproposal.truncation.should_enable_latent_radius(latent_radius_kwargs=None) → bool#: Check if latent-radius truncation should be enabled from kwargs.

nessai.proposal.flowproposal.truncation.build_truncation_methods(truncation_method=None, truncation_methods=None, truncate_log_q=False, enforce_likelihood_threshold=False, latent_radius_kwargs=None, default_latent_radius: bool = False) → list[str]#: Build the effective truncation-method list from legacy and new inputs.

nessai.proposal.flowproposal.truncation.apply_default_truncation_config(methods, truncation_kwargs=None, *, default_latent_radius: bool = False)#: Apply canonical default truncation configuration.

nessai.proposal.flowproposal.truncation.normalise_truncation_kwargs(truncation_method=None, truncation_methods=None, truncation_kwargs=None)#: Normalise truncation kwargs into the canonical method-keyed form.

class nessai.proposal.flowproposal.truncation.BaseTruncationRule#

Base class for truncation rules.

property requires_log_likelihood: bool#: Indicate if the rule needs log-likelihood values.

configure(proposal) → None#: Apply any proposal-level configuration needed by the rule.

prepare(proposal, worst_point, radius=None)#: Prepare per-population data for the rule.

apply_latent(proposal, z)#: Apply truncation in latent space before the inverse pass.

apply_after_backward(proposal, x, log_q, z)#: Apply truncation after the inverse pass and rescaling.

apply_after_likelihood(proposal, x, log_q, z)#: Apply truncation after likelihood evaluation.

reset() → None#: Reset transient state.

class nessai.proposal.flowproposal.truncation.LatentRadiusTruncation(radius_mode: str | None = None, fixed_radius: float | bool = False, min_radius: float | bool = False, max_radius: float | bool = 50.0, compute_radius_with_all: bool = False, constant_volume_mode: bool = False, volume_fraction: float = 0.95, fuzz: float = 1.0, expansion_fraction: float | None = 4.0)#

Bases: BaseTruncationRule

Filter latent samples using a radial threshold.

to_kwargs() → dict#: Return keyword arguments that reconstruct the rule.

configure(proposal) → None#: Apply any proposal-level configuration needed by the rule.

prepare(proposal, worst_point, radius=None)#: Prepare per-population data for the rule.

apply_latent(proposal, z)#: Apply truncation in latent space before the inverse pass.

class nessai.proposal.flowproposal.truncation.MinLogQTruncation#

Bases: BaseTruncationRule

Truncate samples using the minimum live-point log q.

prepare(proposal, worst_point, radius=None)#: Prepare per-population data for the rule.

apply_after_backward(proposal, x, log_q, z)#: Apply truncation after the inverse pass and rescaling.

class nessai.proposal.flowproposal.truncation.LikelihoodThresholdTruncation#

Bases: BaseTruncationRule

Truncate samples using the current likelihood threshold.

property requires_log_likelihood: bool#: Indicate if the rule needs log-likelihood values.

prepare(proposal, worst_point, radius=None)#: Prepare per-population data for the rule.

apply_after_likelihood(proposal, x, log_q, z)#: Apply truncation after likelihood evaluation.

nessai.proposal.flowproposal.truncation.get_truncation_rule_class(name: str)#: Get the truncation rule class for a configured method name.

class nessai.proposal.flowproposal.truncation.TruncationScheme(rules: list[BaseTruncationRule] | None = None)#: Apply an ordered set of truncation rules.