torchtree.variational.chi

Classes

CUBO	Class representing the \(\chi\)-upper bound (CUBO) objective [1].

Module Contents

class torchtree.variational.chi.CUBO(id_: torchtree.typing.ID, q: torchtree.distributions.distributions.DistributionModel, p: torchtree.core.model.CallableModel, samples: torch.Size, n: torch.Tensor)

Bases: torchtree.core.model.CallableModel

Class representing the \(\chi\)-upper bound (CUBO) objective [1].

Parameters:

• id (str or None) – unique identifier of object.

• q (DistributionModel) – variational distribution.

• p (CallableModel) – joint distribution.

• samples (torch.Size) – number of samples to form estimator.

• n (torch.Tensor) – order of \(\chi\)-divergence.

[1] (1,2)

Adji Bousso Dieng, Dustin Tran, Rajesh Ranganath, John Paisley, David Blei. Variational Inference via \(\chi\) Upper Bound Minimization

q

p

n

samples

handle_parameter_changed(variable, index, event)

classmethod from_json(data, dic) → CUBO

Abstract method to create object from a dictionary.

Parameters:

• data (dict[str, Any]) – dictionary representation of a torchtree object.

• dic (dict[str, Any]) – dictionary containing other torchtree objects keyed by their ID.

Returns:

torchtree object.

Return type:

Any