rbnet.autoencoder.AutoencoderNonTermVar

class rbnet.autoencoder.AutoencoderNonTermVar(dim, chart_type='TMap', *args, **kwargs)[source]

Bases: NonTermVar

A point-wise continuous non-terminal variable of dimensionality dim. Distributions over these variables can be thought of as Dirac deltas (the limit of infinitely narrow Gaussians), represented by a location (a specific variable value) and weight (in the case of mixtures or inside probabilities).

Parameters:

  • cardinality – cardinality

  • chart_type – type of chart to use (“dict” or “TMap”)

Public Methods:

__init__(dim[, chart_type])

A point-wise continuous non-terminal variable of dimensionality dim.

get_chart(n, *args, **kwargs)

Initialise a chart for sequence of length n.

mixture(components[, weights, dim])

Approximate a mixture by its weighted average.
Inherited from NonTermVar

__init__(*args, **kwargs)

get_chart(*args, **kwargs)

Return a parse chart to store this variable type in.

mixture(*args, **kwargs)

Compute a mixture over this variable type.

Private Data Attributes:

_abc_impl

Inherited from NonTermVar

_abc_impl

Inherited from ABC

_abc_impl

get_chart(n, *args, **kwargs)[source]

Initialise a chart for sequence of length n. :type n: :param n: length of the sequence :return: chart

mixture(components, weights=None, dim=0)[source]

Approximate a mixture by its weighted average. The new weight is the sum of mixture weights. Mixture weights are provided as part of the components; additional weights provided as weights are multiplied on the weights provided in components.

Parameters:

  • components – array-like with pairs of (values, weights) mixture components along dim

  • weights – [optional] weights of the mixture components; must be compatible (broadcastable) to weights in components

  • dim – integer or tuple of integers indicating the dimensions of components along which to sum to compute the mixture

Returns:

distribution corresponding to the mixture