Mixture¶

class compressed_kde.Mixture¶

Bases: pybind11_object

Mixture class for (compressed) kernel density estimation.

Parameters

space (Space) – Description of mixture space.
threshold (scalar) – Compression threshold

Attributes Summary

`kernel_bandwidths`	Bandwidths of all components.
`kernel_locations`	Locations of all components.
`kernel_scale_factors`	Scale factors of all components.
`ncomponents`	Number of components in (compressed) density.
`space`	Mixture space.
`sum_of_nsamples`	Number of samples that were added to the density.
`sum_of_weights`	Sum of weights of all samples that were added to the density.
`threshold`	Compression threshold.
`weights`	Weights of all components.

Methods Summary

`add`(samples)	Add samples to the mixture.
`clear`	Remove all components and reset mixture.
`evaluate`	evaulate(args,*kwargs) -> array
`from_yaml`(str)	Create mixture from YAML.
`load_from_hdf5`(filename, path)	Load mixture from hdf5 file.
`load_from_yaml`(path)	Load mixture from file.
`marginal`(args, *kwargs)	Evaluate marginal at samples for select dimensions.
`merge`(samples, random)	Merge samples into the mixture.
`partial`	Partially evaluate mixture at samples for select dimensions.
`partialize`(args, *kwargs)	Partially evaluate mixture at grid points for select dimensions.
`save_to_hdf5`(filename, flags, path)	Save mixture to hdf5 file.
`save_to_yaml`(path)	Save mixture to YAML file.
`to_yaml`()	Represent mixture as YAML.

Attributes Documentation

kernel_bandwidths¶: Bandwidths of all components.

kernel_locations¶: Locations of all components.

kernel_scale_factors¶: Scale factors of all components.

ncomponents¶: Number of components in (compressed) density.

space¶: Mixture space.

sum_of_nsamples¶: Number of samples that were added to the density.

sum_of_weights¶: Sum of weights of all samples that were added to the density.

threshold¶: Compression threshold.

weights¶: Weights of all components.

Methods Documentation

add(samples) → None¶

Add samples to the mixture.

New mixture components are added at the sample location with the default kernel bandwidth. No merging with existing components is performed.

Parameters: samples ((n,ndim) array) – Array of samples

clear()¶: Remove all components and reset mixture.

evaluate()¶

evaulate(*args,**kwargs) -> array

Evaluate mixture at samples.

evaluate(samples)¶
evaluate(grid)

Parameters

samples ((n,ndim) array) – Array of samples
grid (Grid) – grid specification

Returns

Evaluated probabilities at sample locations

Return type

ndarray

static from_yaml(str) → Mixture¶

Create mixture from YAML.

Parameters: string (string) – YAML string space representation
Return type: Mixture

static load_from_hdf5(filename, path) → Mixture¶

Load mixture from hdf5 file.

Parameters

filename (string) – path to hdf5 file
path (string) – path inside hdf5 file

Return type

Mixture

static load_from_yaml(path) → Mixture¶

Load mixture from file.

Parameters: path (string) – path to YAML file
Return type: Mixture

marginal(*args, **kwargs) → array¶

Evaluate marginal at samples for select dimensions.

marginal(samples, selection)¶
marginal(grid)

Parameters

samples ((n,nselect) array) – Array of samples
selection ((ndim,) boolean array) – Selected dimensions for marginal evaluation.
grid (Grid) – Grid on subspace of mixture

Returns

Marginal probabilities.

Return type

ndarray

merge(samples, random) → None¶

Merge samples into the mixture.

New mixture components are added at the sample location with the default kernel bandwidth and merged with existing components if the mahalanobis distance is below the threshold.

Parameters

samples ((n,ndim) array) – Array of samples
random (bool) – Whether the samples will be randomized before merging.

partial()¶

Partially evaluate mixture at samples for select dimensions.

Parameters

samples ((n,nselect) array) – Array of samples
selection ((ndim,) boolean array) – Selected dimensions for partial evaluation.

Returns

Partial log probabilities

Return type

(ncomp,nsamples) array

partialize(*args, **kwargs) → PartialMixture¶

Partially evaluate mixture at grid points for select dimensions.

partialize(grid)¶
partialize(samples, selection)

Parameters

grid (Grid) – Grid on subspace of mixture
samples ((n,nselect) array) – Array of samples
selection ((ndim,) boolean array) – Selected dimensions for partial evaluation.

Return type

PartialMixture

save_to_hdf5(filename, flags, path) → None¶

Save mixture to hdf5 file.

Parameters

filename (str) – path to hdf5 file
flags (int) – flags for file creation
path (str) – path inside hdf5 file

save_to_yaml(path) → None¶

Save mixture to YAML file.

Parameters: path (string) – path tho YAML file

to_yaml() → str¶

Represent mixture as YAML.

Return type: string