gpflow.probability_distributions#

Classes#

gpflow.probability_distributions.DiagonalGaussian#

class gpflow.probability_distributions.DiagonalGaussian(mu, cov)[source]#

Bases: ProbabilityDistribution

Parameters:

  • mu (Union[ndarray[Any, Any], Tensor, Variable, Parameter]) –

  • cov (Union[ndarray[Any, Any], Tensor, Variable, Parameter]) –

property shape: Tuple[int | None, ...] | None#

Return the shape of this distribution.

Shape should be some variation of [N, D], where:

  • N is the number of data points.

  • D is the number of input dimensions.

gpflow.probability_distributions.Gaussian#

class gpflow.probability_distributions.Gaussian(mu, cov)[source]#

Bases: ProbabilityDistribution

Parameters:

  • mu (Union[ndarray[Any, Any], Tensor, Variable, Parameter]) –

  • cov (Union[ndarray[Any, Any], Tensor, Variable, Parameter]) –

property shape: Tuple[int | None, ...] | None#

Return the shape of this distribution.

Shape should be some variation of [N, D], where:

  • N is the number of data points.

  • D is the number of input dimensions.

gpflow.probability_distributions.MarkovGaussian#

class gpflow.probability_distributions.MarkovGaussian(mu, cov)[source]#

Bases: ProbabilityDistribution

Gaussian distribution with Markov structure. Only covariances and covariances between t and t+1 need to be parameterised. We use the solution proposed by Carl Rasmussen, i.e. to represent Var[x_t] = cov[x_t, :, :] * cov[x_t, :, :].T Cov[x_t, x_{t+1}] = cov[t, :, :] * cov[t+1, :, :]

Parameters:

  • mu (Union[ndarray[Any, Any], Tensor, Variable, Parameter]) –

  • cov (Union[ndarray[Any, Any], Tensor, Variable, Parameter]) –

property shape: Tuple[int | None, ...] | None#

Return the shape of this distribution.

Shape should be some variation of [N, D], where:

  • N is the number of data points.

  • D is the number of input dimensions.

gpflow.probability_distributions.ProbabilityDistribution#

class gpflow.probability_distributions.ProbabilityDistribution[source]#

Bases: ABC

This is the base class for a probability distributions, over which we take the expectations in the expectations framework.

abstract property shape: Tuple[int | None, ...] | None#

Return the shape of this distribution.

Shape should be some variation of [N, D], where:

  • N is the number of data points.

  • D is the number of input dimensions.

Functions#

gpflow.probability_distributions.get_probability_distribution_shape#

gpflow.probability_distributions.get_probability_distribution_shape(shaped, context)[source]#
Parameters:
Return type:

Optional[Tuple[Optional[int], ...]]