sequince_density.flows.base

Contents

sequince_density.flows.base#

Shared base class for normalizing-flow adapters.

Classes#

Base class for backend flow adapters.

Module Contents#

class sequince_density.flows.base.Flow(dims, device=None, particle_transform=None)#

Base class for backend flow adapters.

This mirrors the Aspire flow surface closely enough for future migration, while remaining compatible with sequince.transforms.FlowTransform.

Periodic particle transforms, including components of composite transforms, are unsupported because these adapters do not sum over periodic images.

Parameters:

dims (int)
device (Any)
particle_transform (sequince.transforms.ParticleTransform | None)

xp: types.ModuleType | None = None#

dims#

device = None#

property particle_transform: sequince.transforms.ParticleTransform#

Coordinate transform used for sampling and density evaluation.

Return type:: sequince.transforms.ParticleTransform

fit_history: dict[str, list[float]]#

abstractmethod fit(samples, context=None)#

abstractmethod forward(samples, context=None)#

abstractmethod inverse(samples, context=None)#

abstractmethod log_prob(samples, context=None)#

abstractmethod sample(n_samples, context=None, *, rng=None)#

Parameters:: n_samples (int)

abstractmethod sample_and_log_prob(n_samples, context=None, *, rng=None)#

Parameters:: n_samples (int)

fit_particle_transform(x, context=None, *, rng=None)#

forward_particle_transform(x, context=None, *, rng=None)#

inverse_particle_transform(x, context=None, *, auxiliary=None, rng=None)#