sequince.flows#

Flow interfaces used by Sequince transforms and proposals.

Classes#

NormalizingFlow

Minimal interface for normalizing flows used by Sequince.

Functions#

`flow_asarray`(flow, array)	Cast `array` to the backend array type expected by `flow`.
`flow_state_dict`(flow)	Return persistent flow state when the adapter exposes it.
`load_flow_state_dict`(flow, state)	Restore persistent flow state when supported by the adapter.
`clone_flow`(flow)	Return a fresh copy of `flow`.

Module Contents#

class sequince.flows.NormalizingFlow[source]#

Bases: Protocol

Minimal interface for normalizing flows used by Sequince.

xp: Any#

fit(particles, context=None, *, rng=None, **kwargs)[source]#

Parameters:

particles (sequince.typing.Array)
context (sequince.steps.context.StepContext | None)
rng (Any)
kwargs (Any)

Return type:

NormalizingFlow

forward(particles, context=None)[source]#

Map particles into a transformed space and return log-Jacobians.

Parameters:

particles (sequince.typing.Array)
context (sequince.steps.context.StepContext | None)

Return type:

tuple[sequince.typing.Array, sequince.typing.Array]

inverse(particles, context=None)[source]#

Undo the transform applied by forward().

Parameters:

particles (sequince.typing.Array)
context (sequince.steps.context.StepContext | None)

Return type:

tuple[sequince.typing.Array, sequince.typing.Array]

log_prob(particles, context=None)[source]#

Evaluate the flow log-density in the original particle space.

Parameters:

particles (sequince.typing.Array)
context (sequince.steps.context.StepContext | None)

Return type:

sequince.typing.Array

sample(num_samples=1, context=None, *, rng=None)[source]#

Draw samples from the flow distribution.

Parameters:

num_samples (int)
context (sequince.steps.context.StepContext | None)
rng (Any)

Return type:

sequince.typing.Array

sequince.flows.flow_asarray(flow, array)[source]#

Cast array to the backend array type expected by flow.

Parameters:

flow (NormalizingFlow)
array (Any)

sequince.flows.flow_state_dict(flow)[source]#

Return persistent flow state when the adapter exposes it.

Parameters:: flow (NormalizingFlow)
Return type:: dict[str, Any]

sequince.flows.load_flow_state_dict(flow, state)[source]#

Restore persistent flow state when supported by the adapter.

Parameters:

flow (NormalizingFlow)
state (collections.abc.Mapping[str, Any])

Return type:

None

sequince.flows.clone_flow(flow)[source]#

Return a fresh copy of flow.

External flow adapters can expose _init_args to control how the clone is reconstructed; otherwise deepcopy is used as a fallback.

Parameters:: flow (NormalizingFlow)
Return type:: NormalizingFlow