pyvfg 6.2.0

No project description provided

pyvfg

This package declares and defines a class VFG that represents a Verses Factor Graph.

What is a VFG?

VFGs, or Verses Factor Graphs, are a data structure that represents a probabilistic model. They are used to represent the relationships between variables in a model, and can be used to perform inference and learning. This is a generic structure that can be used to represent a variety of models, including Bayesian networks, Markov random fields, and partially-observable Markov decision processes (POMDPs).

Versioning

VFG is versioned from 0.2.0 to 0.5.0. These are, in general, backwards-compatible -- calling pyvfg.vfg_upgrade() on a 0.2.0 VFG will produce a 0.5.0 VFG. However, one exception os <= 0.4.0 POMDPs to 0.5.0 POMDPs. Please see below for how to upgrade the POMDPs.

Upgrading POMDPs from 0.4.0 to 0.5.0

VFG 0.5.0 introduces numeric validation for factor values. This means that POMDPs that use "categorical" for their reward factor will fail validation. As such, the model will need to be updated.

To upgrade a POMDP from 0.4.0 to 0.5.0, you will need to change the reward factor from "categorical" to "logits".

Model Description

Currently supported model types are Bayesian Networks (BNs), Markov Random Fields (MRFs), and Partially-Observable Markov Decision Processes (POMDPs).

Version

Determines how the model will be parsed, for backwards compatability when using durable storage. Will be output as 0.5.0.

Variables

Variables are the nodes in the VFG. They represent the states, actions, and observations in the model.

Variable Role

Role	Model Type	Description
null	BN, MRF, POMDP	a "default" variable without a role.
control_state	POMDP	The state of the system. This is the variable that is being controlled.
latent	BN, MRF	A variable known to be present in the system, but cannot be observed.

Factors

Factors represent the relations between nodes.

Possible Distribution Types

Distribution	Model type	Description	Example Field
categorical	BN, MRF, POMDP	A categorical distribution over a set of variables, for joint probability. Joint probability of a single variable is simply the probability of that variable.	position with role initial_state_prior
categorical_conditional	BN, MRF, POMDP	A categorical distribution conditioned on a set of variables. ["A", "B", "C"] means P(A|B,C).	observation|position
logits	POMDP	A distribution as the input to a softmax, usually for intermediate or reward states.	position with role preference
potential	MRF	A non-normalized probability distribution.	position with role potential

Counts

The counts are raw, observed counts for input variables, and are used to scale for continuous learning. This must be kept in sync with values (which can be done by setting the values field to the normalized counts).

Values

The values are scaled, probabilistic values, and are used for inference. This must be kept in sync with counts, if counts are present for the same factor.

Factor Role

Role	Model Type	Description
null	BN, MRF, POMDP	A "default" factor without a role.
transition	MRF, POMDP	The transition factor. This is a factor that represents the transition probabilities between states.
reward	POMDP	The reward factor. This is a factor that represents the reward probabilities.
initial_state_prior	POMDP	The initial state prior factor. This is a factor that represents the initial state probabilities.
preference	POMDP	The preference factor. This is a factor that represents the preference probabilities.
belief	MRF	The potential factor. This is a factor that represents the potential probabilities.
observation	POMDP	The observation factor. This is a factor that represents the observation probabilities.

Metadata

Stores information about the model, as distinct from the graph. These are user-defined and will be parroted, without affecting output.

Model Version

The version of the model, not the VFG. User-defined. No versioning scheme is imposed.

Model type

Informational only. One of "bayesian_network", "markov_random_field", "pomdp", or the generic "factor_graph". This is not used in any parsers; model type is used implicitly.

Description

Free-form text field describing information and the purpose of the model.

Visualization Metadata

Will be parroted, and never parsed. May be removed in the future.