figaro.mixture module¶

class figaro.mixture.DPGMM(bounds, prior_pars=None, alpha0=1.0, probit=True, n_reassignments=0.0)[source]¶

Bases: density

Class to infer a distribution given a set of samples.

Parameters:

bounds (iterable) – boundaries of the rectangle over which the distribution is defined. It should be in the format [[xmin, xmax],[ymin, ymax],…]
prior_pars (iterable) – NIW prior parameters (k, L, nu, mu)
alpha0 (double) – initial guess for concentration parameter
probit (bool) – whether to use the probit transformation or not
n_reassignments (int) – number of reassignments. Default is not to reassign.

Returns:

instance of DPGMM class

Return type:

DPGMM

add_new_point(x, *args, **kwargs)¶

build_mixture(make_comp=True)[source]¶

Instances a mixture class representing the inferred distribution

Parameters:: make_comp (bool) – whether to instantiate the scipy.stats.multivariate_normal components or not
Returns:: the inferred distribution
Return type:: mixture

density_from_samples(samples, make_comp=True)[source]¶

Reconstruct the probability density from a set of samples.

Parameters:

samples (iterable) – samples set
make_comp (bool) – whether to instantiate the scipy.stats.multivariate_normal components or not

Returns:

the inferred mixture

Return type:

mixture

initialise(prior_pars=None)[source]¶

Initialise the mixture to initial conditions.

Parameters:: prior_pars (iterable) – NIW prior parameters (k, L, nu, mu). If None, old parameters are kept

class figaro.mixture.HDPGMM(bounds, alpha0=1.0, prior_pars=None, MC_draws=None, probit=True, n_reassignments=0.0, selection_function=None, injection_pdf=None, total_injections=None)[source]¶

Bases: DPGMM

Class to infer a distribution given a set of observations (each being a set of samples). Child of DPGMM class

Parameters:

bounds (iterable) – boundaries of the rectangle over which the distribution is defined. It should be in the format [[xmin, xmax],[ymin, ymax],…]
prior_pars (iterable) – IW parameters
alpha0 (double) – initial guess for concentration parameter
MC_draws (double) – number of MC draws for integral
probit (bool) – whether to use the probit transformation or not
n_reassignments (int) – number of reassignments. Default is not to reassign.
selection_function (callable) – selection function approximant or samples
injection_pdf (np.ndarray) – pdf of injected samples (for selection function inclusion)

Returns:

instance of HDPGMM class

Return type:

HDPGMM

add_new_point(ev)[source]¶

Update the probability density reconstruction adding a new sample

Parameters:: ev (iterable) – set of single-event draws from a DPGMM inference

build_mixture(make_comp=True)[source]¶

Instances a mixture class representing the inferred distribution

Parameters:: make_comp (bool) – whether to instantiate the scipy.stats.multivariate_normal components or not
Returns:: the inferred distribution
Return type:: mixture

compute_alpha_factor()[source]¶

Compute the integral int pdet(theta)p(theta|lambda) dtheta conditioned on the specific DPGMM parameters

Returns:: value of the integral
Return type:: double

density_from_samples(events, make_comp=True)[source]¶

Reconstruct the probability density from a set of samples.

Parameters:

samples (iterable) – set of single-event draws from DPGMM
make_comp (bool) – whether to instantiate the scipy.stats.multivariate_normal components or not

Returns:

the inferred mixture

Return type:

mixture

initialise()[source]¶: Initialise the mixture to initial conditions

class figaro.mixture.density[source]¶

Bases: object

Class to initialise a common set of methods for mixture models. Not to be used.

cdf(x)[source]¶

fast_logpdf(x)[source]¶

Fast logpdf evaluation using FIGARO implementation of log_norm (JIT) rather than Numpy’s. WARNING: it is meant to be used with MCMC samplers, therefore accepts only one point at a time.

Parameters:: x (np.ndarray) – point to evaluate the mixture at
Returns:: mixture.pdf(x)
Return type:: np.ndarray

fast_pdf(x)[source]¶

Fast pdf evaluation using FIGARO implementation of log_norm (JIT) rather than Numpy’s. WARNING: it is meant to be used with MCMC samplers, therefore accepts only one point at a time.

Parameters:: x (np.ndarray) – point to evaluate the mixture at
Returns:: mixture.pdf(x)
Return type:: np.ndarray

gradient(x)[source]¶

Gradient of the mixture.

Parameters:: x (np.ndarray) – point to evaluate the gradient at
Returns:: gradient
Return type:: np.ndarray

log_gradient(x)[source]¶

Logarithmic gradient of the mixture.

Parameters:: x (np.ndarray) – point to evaluate the gradient at
Returns:: logarithmic gradient
Return type:: np.ndarray

logcdf(x)[source]¶

logpdf(x)[source]¶

pdf(x)[source]¶

rvs(*args, **kwargs)¶

class figaro.mixture.mixture(means, covs, w, bounds, dim, n_cl, n_pts, alpha=1.0, probit=True, log_w=None, make_comp=True, alpha_factor=1.0)[source]¶

Bases: density

Class to store a single draw from DPGMM/(H)DPGMM. Methods inherited from density class

Parameters:

means (iterable) – component means
covs (iterable) – component covariances
w (np.ndarray) – component weights
bounds (np.ndarray) – bounds of probit transformation
dim (int) – number of dimensions
n_cl (int) – number of clusters in the mixture
n_pts (int) – number of points used to infer the mixture
alpha (double) – concentration parameter
probit (bool) – whether to use the probit transformation or not
log_w (np.ndarray) – component log weights
make_comp (bool) – make component objects
alpha_factor (double) – evaluated int pdet(theta)p(theta|lambda) conditioned on the mixture parameters

Returns:

instance of mixture class

Return type:

mixture

condition(vals, dims, norm=True, filter=True, tol=0.001)[source]¶

Mixture conditioned on specific values of a subset of parameters.

Parameters:

vals (iterable) – value(s) to condition on
dims (int or list of int) – dimension(s) associated with given vals (starting from 0)
norm (bool) – whether to normalize the distribution or not
filter (bool) – filter the components with weight < tol
tol (double) – tolerance on the sum of the weights

Returns:

conditioned mixture

Return type:

figaro.mixture.mixture

marginalise(axis=-1)[source]¶

Marginalise out one or more dimensions from the mixture.

Parameters:: axis (int or list of int) – axis to marginalise on. Default: last
Returns:: marginalised mixture
Return type:: figaro.mixture.mixture