bayespy.inference.VB

class bayespy.inference.VB(*nodes, tol=1e-05, autosave_filename=None, autosave_iterations=0, use_logging=False, user_data=None, callback=None)[source]

Variational Bayesian (VB) inference engine

Parameters:

  • nodes (nodes) – Nodes that form the model. Must include all at least all stochastic nodes of the model.

  • tol (double, optional) – Convergence criterion. Tolerance for the relative change in the VB lower bound.

  • autosave_filename (string, optional) – Filename for automatic saving

  • autosave_iterations (int, optional) – Iteration interval between each automatic saving

  • callback (callable, optional) – Function which is called after each update iteration step

__init__(*nodes, tol=1e-05, autosave_filename=None, autosave_iterations=0, use_logging=False, user_data=None, callback=None)[source]

Methods

__init__(*nodes[, tol, autosave_filename, ...])

add(x1, x2[, scale])

Add two vectors (in parameter format)

compute_lowerbound([ignore_masked])

compute_lowerbound_terms(*nodes)

dot(x1, x2)

Computes dot products of given vectors (in parameter format)

get_gradients(*nodes[, euclidian])

Computes gradients (both Riemannian and normal)

get_iteration_by_nodes()

get_parameters(*nodes)

Get parameters of the nodes

gradient_step(*nodes[, scale])

Update nodes by taking a gradient ascent step

has_converged([tol])

load(*nodes[, filename, nodes_only])

load_user_data(filename)

loglikelihood_lowerbound()

optimize(*nodes[, maxiter, verbose, method, ...])

Optimize nodes using Riemannian conjugate gradient

pattern_search(*nodes[, collapsed, maxiter])

Perform simple pattern search [4].

plot(*nodes, **kwargs)

Plot the distribution of the given nodes (or all nodes)

plot_iteration_by_nodes([axes, diff])

Plot the cost function per node during the iteration.

save(*nodes[, filename])

set_annealing(annealing)

Set deterministic annealing from range (0, 1].

set_autosave(filename[, iterations, nodes])

set_callback(callback)

set_parameters(x, *nodes)

Set parameters of the nodes

update(*nodes[, repeat, plot, tol, verbose, ...])

use_logging(use)

Attributes

ignore_bound_checks