starred.optim package

Submodules

starred.optim.inference_base module

class starred.optim.inference_base.FisherCovariance(parameter_class, differentiable_class, diagonal_only=False)

Bases: object

Parameters:

diagonal_only – bool, if you are confident your parameters are not correlated: you can save space by calculating the diagonal of the hessian only.

compute_fisher_information(recompute=False)

Parameters:

recompute – bool, redo? default False

property covariance_matrix

draw_samples(num_samples=10000, seed=None)

Draw samples from the multivariate Gaussian distribution defined by the best fit values and the covariance

Parameters:

• num_samples – int, number of samples to draw

• seed – int, seed for the random number generator

Returns:

array of shape (num_samples, num_parameters)

fisher2covar(fisher_matrix, inversion='full')

property fisher_matrix

get_kwargs_sigma()

Return the standard deviation of the marginalized distribution for each parameter. This corresponds to the square roots of the diagonal coefficient of the covariance matrix.

static split_matrix(matrix, num_before, num_after)

class starred.optim.inference_base.InferenceBase(loss_class, param_class)

Bases: object

Class that defines wraps the loss function, and computes first and second order derivatives.

Parameters:

• loss_class – Loss instance

• param_class – Parameters instance

gradient(args)

Returns the gradient (first derivative) of the loss function.

hessian(args)

Returns the Hessian (second derivative) of the loss function.

hessian_vec_prod(args, vec)

Hessian-vector product.

log_likelihood(args)

Log Likelihood function to be maximised. :param args: list or array of paramers

loss(args)

Loss function to be minimized. :param args: list or array of paramers

property parameters

Returns the parameters.

value_and_gradient(args)

Returns both the value and the gradient (first derivative) of the loss function.

starred.optim.optimization module

class starred.optim.optimization.Optimizer(loss_class, param_class, method='BFGS')

Bases: InferenceBase

Class that handles optimization tasks, i.e., finding best-fit point estimates of parameters It currently handles a subset of scipy.optimize.minimize routines, using first and second order derivatives when required.

compute_fisher_matrix()

property loss_history

Returns the loss history.

minimize(restart_from_init=True, **kwargs_optimiser)

Minimizes the loss function and returns the best fit.

property param_history

Returns the parameter history.

starred.optim.sampling module

class starred.optim.sampling.Sampler(loss_class, param_class, sampler='emcee')

Bases: InferenceBase

prior_draw(key)

Uniform prior

sample(init_params, **kwargs_sampler)

Minimizes the loss function and returns the best fit.

transform(x)

transform x back to the original parameter. Useful if some parameter are sampled in log space for example

Module contents

This subpackage contains optimization and sampling functions