GaussianMixtureModel

class jolideco.priors.GaussianMixtureModel(means, covariances, weights, precisions_cholesky, meta=None)

Bases: Module

Gaussian mixture model

means

Means

Type:

~torch.Tensor

covariances

Covariances

Type:

~torch.Tensor

weights

Weights

Type:

~torch.Tensor

precisions_cholesky

Precision matrices

Type:

~torch.Tensor

meta

Meta data

Type:

GaussianMixtureModelMeta

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Attributes Summary

covariance_det	Covariance determinant
covariances_numpy	Covariances (~numpy.ndarray)
eigen_images	Eigen images
log_det_cholesky	Precision matrices pytorch
log_det_cholesky_numpy	Compute the log-det of the cholesky decomposition of matrices
log_weights	Log weights (~numpy.ndarray)
log_weights_numpy	Weights (~numpy.ndarray)
means_numpy	Means (~numpy.ndarray)
means_precisions_cholesky	Precision matrices pytorch
n_components	Number of features
n_features	Number of features
patch_shape	Patch shape (tuple)
pixel_weights	Pixel weights
pixel_weights_numpy	Pixel weights
precisions_cholesky_numpy	Precisions Cholesky (~numpy.ndarray)
weights_numpy	Weights (~numpy.ndarray)

Methods Summary

estimate_log_prob(x)	Compute log likelihood for given feature vector
estimate_log_prob_numpy(x)	Compute log likelihood for given feature vector
from_dict(data)	Create from dict
from_numpy(means, covariances, weights[, meta])	Gaussian mixture model
from_registry(name, **kwargs)	Create GMM from registry
from_sklearn_gmm(gmm)	Create from sklearn GMM
is_equal(other)
kl_divergence(other)	Compute KL divergence with respect to another GMM"
plot_eigen_images([ncols, figsize])	Plot images
plot_mean_images([ncols, figsize])	Plot mean images
read(filename[, format])	Read from matlab file
symmetric_kl_divergence(other)	Symmetric KL divergence
to_dict()	To dict

Attributes Documentation

covariance_det

Covariance determinant

covariances_numpy

Covariances (~numpy.ndarray)

eigen_images

Eigen images

log_det_cholesky

Precision matrices pytorch

log_det_cholesky_numpy

Compute the log-det of the cholesky decomposition of matrices

log_weights

Log weights (~numpy.ndarray)

log_weights_numpy

Weights (~numpy.ndarray)

means_numpy

Means (~numpy.ndarray)

means_precisions_cholesky

Precision matrices pytorch

n_components

Number of features

n_features

Number of features

patch_shape

Patch shape (tuple)

pixel_weights

Pixel weights

pixel_weights_numpy

Pixel weights

precisions_cholesky_numpy

Precisions Cholesky (~numpy.ndarray)

weights_numpy

Weights (~numpy.ndarray)

Methods Documentation

estimate_log_prob(x)

Compute log likelihood for given feature vector

estimate_log_prob_numpy(x)

Compute log likelihood for given feature vector

classmethod from_dict(data)

Create from dict

Parameters:

data (dict) – Data dictionary

Returns:

gmm – Gaussian mixture model

Return type:

~GaussianMixtureModel

classmethod from_numpy(means, covariances, weights, meta=None)

Gaussian mixture model

Parameters:

• means (~numpy.ndarray) – Means

• covariances (~numpy.ndarray) – Covariances

• weights (~numpy.ndarray) – Weights

• meta (GaussianMixtureModelMeta) – Meta data

Returns:

gmm – Gaussian mixture model.

Return type:

GaussianMixtureModel

classmethod from_registry(name, **kwargs)

Create GMM from registry

Parameters:

name (str) – Name of the registered GMM.

Returns:

gmm – Gaussian mixture model.

Return type:

GaussianMixtureModel

classmethod from_sklearn_gmm(gmm)

Create from sklearn GMM

is_equal(other)

kl_divergence(other)

Compute KL divergence with respect to another GMM”

See https://mr-easy.github.io/2020-04-16-kl-divergence-between-2-gaussian-distributions/

Parameters:

other (~GaussianMixtureModel) – Other GMM

Returns:

value – KL divergence

Return type:

float

plot_eigen_images(ncols=20, figsize=None)

Plot images

plot_mean_images(ncols=20, figsize=None)

Plot mean images

classmethod read(filename, format='epll-matlab', **kwargs)

Read from matlab file

Parameters:

• filename (str or Path) – Filename

• format ({"epll-matlab", "epll-matlab-16x16", "table"}) – Format

• **kwargs (dict) – Keyword arguments passed to GaussianMixtureModel

Returns:

gmm – Gaussian mixture model.

Return type:

GaussianMixtureModel

symmetric_kl_divergence(other)

Symmetric KL divergence

to_dict()

To dict