gpflow.likelihoods¶

gpflow.likelihoods.Bernoulli¶

class gpflow.likelihoods.Bernoulli(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

gpflow.likelihoods.Beta¶

class gpflow.likelihoods.Beta(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

This uses a reparameterisation of the Beta density. We have the mean of the Beta distribution given by the transformed process:

m = invlink(f)

and a scale parameter. The familiar α, β parameters are given by

m = α / (α + β) scale = α + β

so:: α = scale * m β = scale * (1-m)

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

gpflow.likelihoods.Exponential¶

class gpflow.likelihoods.Exponential(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

gpflow.likelihoods.Gamma¶

class gpflow.likelihoods.Gamma(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

Use the transformed GP to give the scale (inverse rate) of the Gamma

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

gpflow.likelihoods.Gaussian¶

class gpflow.likelihoods.Gaussian(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

The Gaussian likelihood is appropriate where uncertainties associated with the data are believed to follow a normal distribution, with constant variance.

Very small uncertainties can lead to numerical instability during the optimization process. A lower bound of 1e-6 is therefore imposed on the likelihood variance by default.

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

__init__(variance=1.0, variance_lower_bound=1e-06, **kwargs)[source]¶

Parameters

variance – The noise variance; must be greater than variance_lower_bound.
variance_lower_bound – The lower (exclusive) bound of variance.
kwargs – Keyword arguments forwarded to ScalarLikelihood.

gpflow.likelihoods.GaussianMC¶

class gpflow.likelihoods.GaussianMC(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.MonteCarloLikelihood, gpflow.likelihoods.scalar_continuous.Gaussian

Stochastic version of Gaussian likelihood for demonstration purposes only.

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

gpflow.likelihoods.HeteroskedasticTFPConditional¶

class gpflow.likelihoods.HeteroskedasticTFPConditional(distribution_class=tensorflow_probability.distributions.Normal, scale_transform=None, **kwargs)[source]¶

Bases: gpflow.likelihoods.multilatent.MultiLatentTFPConditional

Heteroskedastic Likelihood where the conditional distribution is given by a TensorFlow Probability Distribution. The loc and scale of the distribution are given by a two-dimensional multi-output GP.

Attributes

parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

Parameters

distribution_class (Type[Distribution]) –
scale_transform (Optional[Bijector]) –

__init__(distribution_class=tensorflow_probability.distributions.Normal, scale_transform=None, **kwargs)[source]¶

Parameters

distribution_class (Type[Distribution]) – distribution class parameterized by loc and scale as first and second argument, respectively.
scale_transform (Optional[Bijector]) – callable/bijector applied to the latent function modelling the scale to ensure its positivity. Typically, tf.exp or tf.softplus, but can be any function f: R -> R^+. Defaults to exp if not explicitly specified.

gpflow.likelihoods.Likelihood¶

class gpflow.likelihoods.Likelihood(latent_dim, observation_dim)[source]¶

Bases: gpflow.base.Module

Attributes

parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

Parameters

latent_dim (int) –
observation_dim (int) –

__init__(latent_dim, observation_dim)[source]¶

A base class for likelihoods, which specifies an observation model connecting the latent functions (‘F’) to the data (‘Y’).

All of the members of this class are expected to obey some shape conventions, as specified by latent_dim and observation_dim.

If we’re operating on an array of function values ‘F’, then the last dimension represents multiple functions (preceding dimensions could represent different data points, or different random samples, for example). Similarly, the last dimension of Y represents a single data point. We check that the dimensions are as this object expects.

The return shapes of all functions in this class is the broadcasted shape of the arguments, excluding the last dimension of each argument.

Parameters

latent_dim (int) – the dimension of the vector F of latent functions for a single data point
observation_dim (int) – the dimension of the observation vector Y for a single data point

_check_data_dims(Y)[source]¶

Ensure that a tensor of data Y has observation_dim as right-most dimension.

Parameters: Y – observation Tensor, with shape […, observation_dim]

_check_last_dims_valid(F, Y)[source]¶

Assert that the dimensions of the latent functions F and the data Y are compatible.

Parameters

F – function evaluation Tensor, with shape […, latent_dim]
Y – observation Tensor, with shape […, observation_dim]

_check_latent_dims(F)[source]¶

Ensure that a tensor of latent functions F has latent_dim as right-most dimension.

Parameters: F – function evaluation Tensor, with shape […, latent_dim]

_check_return_shape(result, F, Y)[source]¶

Check that the shape of a computed statistic of the data is the broadcasted shape from F and Y.

Parameters

result – result Tensor, with shape […]
F – function evaluation Tensor, with shape […, latent_dim]
Y – observation Tensor, with shape […, observation_dim]

conditional_mean(F)[source]¶

The conditional mean of Y|F: [E[Y₁|F], …, E[Yₖ|F]] where K = observation_dim

Parameters: F – function evaluation Tensor, with shape […, latent_dim]
Returns: mean […, observation_dim]

conditional_variance(F)[source]¶

The conditional marginal variance of Y|F: [var(Y₁|F), …, var(Yₖ|F)] where K = observation_dim

Parameters: F – function evaluation Tensor, with shape […, latent_dim]
Returns: variance […, observation_dim]

log_prob(F, Y)[source]¶

The log probability density log p(Y|F)

Parameters

F – function evaluation Tensor, with shape […, latent_dim]
Y – observation Tensor, with shape […, observation_dim]:

Returns

log pdf, with shape […]

predict_density(Fmu, Fvar, Y)[source]¶: Deprecated: see predict_log_density

predict_log_density(Fmu, Fvar, Y)[source]¶

Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,

i.e. if: q(F) = N(Fmu, Fvar)

and this object represents

p(y|F)

then this method computes the predictive density

log ∫ p(y=Y|F)q(F) df

Parameters

Fmu – mean function evaluation Tensor, with shape […, latent_dim]
Fvar – variance of function evaluation Tensor, with shape […, latent_dim]
Y – observation Tensor, with shape […, observation_dim]:

Returns

log predictive density, with shape […]

predict_mean_and_var(Fmu, Fvar)[source]¶

Given a Normal distribution for the latent function, return the mean and marginal variance of Y,

i.e. if: q(f) = N(Fmu, Fvar)

and this object represents

p(y|f)

then this method computes the predictive mean

∫∫ y p(y|f)q(f) df dy

and the predictive variance

∫∫ y² p(y|f)q(f) df dy - [ ∫∫ y p(y|f)q(f) df dy ]²

Parameters

Fmu – mean function evaluation Tensor, with shape […, latent_dim]
Fvar – variance of function evaluation Tensor, with shape […, latent_dim]

Returns

mean and variance, both with shape […, observation_dim]

variational_expectations(Fmu, Fvar, Y)[source]¶

Compute the expected log density of the data, given a Gaussian distribution for the function values,

i.e. if: q(f) = N(Fmu, Fvar)

and this object represents

p(y|f)

then this method computes

∫ log(p(y=Y|f)) q(f) df.

This only works if the broadcasting dimension of the statistics of q(f) (mean and variance) are broadcastable with that of the data Y.

Parameters

Fmu – mean function evaluation Tensor, with shape […, latent_dim]
Fvar – variance of function evaluation Tensor, with shape […, latent_dim]
Y – observation Tensor, with shape […, observation_dim]:

Returns

expected log density of the data given q(F), with shape […]

gpflow.likelihoods.MonteCarloLikelihood¶

class gpflow.likelihoods.MonteCarloLikelihood(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.Likelihood

Attributes

parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

_predict_log_density(Fmu, Fvar, Y, epsilon=None)[source]¶

Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y.

i.e. if: q(f) = N(Fmu, Fvar)

and this object represents

p(y|f)

then this method computes the predictive density

log ∫ p(y=Y|f)q(f) df

Here, we implement a default Monte Carlo routine.

_predict_mean_and_var(Fmu, Fvar, epsilon=None)[source]¶

Given a Normal distribution for the latent function, return the mean of Y

if: q(f) = N(Fmu, Fvar)

and this object represents

p(y|f)

then this method computes the predictive mean

∫∫ y p(y|f)q(f) df dy

and the predictive variance

∫∫ y² p(y|f)q(f) df dy - [ ∫∫ y p(y|f)q(f) df dy ]²

Here, we implement a default Monte Carlo routine.

_variational_expectations(Fmu, Fvar, Y, epsilon=None)[source]¶

Compute the expected log density of the data, given a Gaussian distribution for the function values.

if: q(f) = N(Fmu, Fvar) - Fmu: [N, D] Fvar: [N, D]

and this object represents

p(y|f) - Y: [N, 1]

then this method computes

∫ (log p(y|f)) q(f) df.

Here, we implement a default Monte Carlo quadrature routine.

gpflow.likelihoods.MultiClass¶

class gpflow.likelihoods.MultiClass(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.Likelihood

Attributes

parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

__init__(num_classes, invlink=None, **kwargs)[source]¶

A likelihood for multi-way classification. Currently the only valid choice of inverse-link function (invlink) is an instance of RobustMax.

For most problems, the stochastic Softmax likelihood may be more appropriate (note that you then cannot use Scipy optimizer).

gpflow.likelihoods.MultiLatentLikelihood¶

class gpflow.likelihoods.MultiLatentLikelihood(latent_dim, **kwargs)[source]¶

Bases: gpflow.likelihoods.base.QuadratureLikelihood

A Likelihood which assumes that a single dimensional observation is driven by multiple latent GPs.

Note that this implementation does not allow for taking into account covariance between outputs.

Attributes

parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

Parameters: latent_dim (int) –

gpflow.likelihoods.MultiLatentTFPConditional¶

class gpflow.likelihoods.MultiLatentTFPConditional(latent_dim, conditional_distribution, **kwargs)[source]¶

Bases: gpflow.likelihoods.multilatent.MultiLatentLikelihood

MultiLatent likelihood where the conditional distribution is given by a TensorFlow Probability Distribution.

Attributes

parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

Parameters

latent_dim (int) –
conditional_distribution (Callable[…, Distribution]) –

__init__(latent_dim, conditional_distribution, **kwargs)[source]¶

Parameters

latent_dim (int) – number of arguments to the conditional_distribution callable
conditional_distribution (Callable[…, Distribution]) – function from Fs to a tfp Distribution, where Fs has shape […, latent_dim]

_conditional_mean(Fs)[source]¶

The conditional marginal mean of Y|F: [E(Y₁|F)]

Parameters: Fs – function evaluation Tensor, with shape […, latent_dim]
Returns: mean […, 1]

_conditional_variance(Fs)[source]¶

The conditional marginal variance of Y|F: [Var(Y₁|F)]

Parameters: Fs – function evaluation Tensor, with shape […, latent_dim]
Returns: variance […, 1]

_log_prob(Fs, Y)[source]¶

The log probability density log p(Y|F)

Parameters

F – function evaluation Tensor, with shape […, latent_dim]
Y – observation Tensor, with shape […, 1]:

Return type

Tensor

Returns

log pdf, with shape […]

gpflow.likelihoods.Ordinal¶

class gpflow.likelihoods.Ordinal(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

A likelihood for doing ordinal regression.

The data are integer values from 0 to k, and the user must specify (k-1) ‘bin edges’ which define the points at which the labels switch. Let the bin edges be [a₀, a₁, … aₖ₋₁], then the likelihood is

p(Y=0|F) = ɸ((a₀ - F) / σ) p(Y=1|F) = ɸ((a₁ - F) / σ) - ɸ((a₀ - F) / σ) p(Y=2|F) = ɸ((a₂ - F) / σ) - ɸ((a₁ - F) / σ) … p(Y=K|F) = 1 - ɸ((aₖ₋₁ - F) / σ)

where ɸ is the cumulative density function of a Gaussian (the inverse probit function) and σ is a parameter to be learned. A reference is:

@article{chu2005gaussian,: title={Gaussian processes for ordinal regression}, author={Chu, Wei and Ghahramani, Zoubin}, journal={Journal of Machine Learning Research}, volume={6}, number={Jul}, pages={1019–1041}, year={2005}

}

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

__init__(bin_edges, **kwargs)[source]¶: bin_edges is a numpy array specifying at which function value the output label should switch. If the possible Y values are 0…K, then the size of bin_edges should be (K-1).

_make_phi(F)[source]¶

A helper function for making predictions. Constructs a probability matrix where each row output the probability of the corresponding label, and the rows match the entries of F.

Note that a matrix of F values is flattened.

gpflow.likelihoods.Poisson¶

class gpflow.likelihoods.Poisson(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

Poisson likelihood for use with count data, where the rate is given by the (transformed) GP.

let g(.) be the inverse-link function, then this likelihood represents

p(yᵢ | fᵢ) = Poisson(yᵢ | g(fᵢ) * binsize)

Note:binsize For use in a Log Gaussian Cox process (doubly stochastic model) where the rate function of an inhomogeneous Poisson process is given by a GP. The intractable likelihood can be approximated via a Riemann sum (with bins of size ‘binsize’) and using this Poisson likelihood.

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

gpflow.likelihoods.QuadratureLikelihood¶

class gpflow.likelihoods.QuadratureLikelihood(latent_dim, observation_dim, *, quadrature=None)[source]¶

Bases: gpflow.likelihoods.base.Likelihood

Attributes

parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

Parameters

latent_dim (int) –
observation_dim (int) –
quadrature (Optional[GaussianQuadrature]) –

_predict_log_density(Fmu, Fvar, Y)[source]¶: Here, we implement a default Gauss-Hermite quadrature routine, but some likelihoods (Gaussian, Poisson) will implement specific cases. :param Fmu: mean function evaluation Tensor, with shape […, latent_dim] :param Fvar: variance of function evaluation Tensor, with shape […, latent_dim] :param Y: observation Tensor, with shape […, observation_dim]: :returns: log predictive density, with shape […]

_predict_mean_and_var(Fmu, Fvar)[source]¶

Here, we implement a default Gauss-Hermite quadrature routine, but some likelihoods (e.g. Gaussian) will implement specific cases.

Parameters

Fmu – mean function evaluation Tensor, with shape […, latent_dim]
Fvar – variance of function evaluation Tensor, with shape […, latent_dim]

Returns

mean and variance of Y, both with shape […, observation_dim]

_quadrature_log_prob(F, Y)[source]¶

Returns the appropriate log prob integrand for quadrature.

Quadrature expects f(X), here logp(F), to return shape [N_quad_points] + batch_shape + [d’]. Here d’=1, but log_prob() only returns [N_quad_points] + batch_shape, so we add an extra dimension.

Also see _quadrature_reduction.

_quadrature_reduction(quadrature_result)[source]¶

Converts the quadrature integral appropriately.

The return shape of quadrature is batch_shape + [d’]. Here, d’=1, but we want predict_log_density and variational_expectations to return just batch_shape, so we squeeze out the extra dimension.

Also see _quadrature_log_prob.

_variational_expectations(Fmu, Fvar, Y)[source]¶: Here, we implement a default Gauss-Hermite quadrature routine, but some likelihoods (Gaussian, Poisson) will implement specific cases. :param Fmu: mean function evaluation Tensor, with shape […, latent_dim] :param Fvar: variance of function evaluation Tensor, with shape […, latent_dim] :param Y: observation Tensor, with shape […, observation_dim]: :returns: variational expectations, with shape […]

gpflow.likelihoods.RobustMax¶

class gpflow.likelihoods.RobustMax(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.base.Module

This class represent a multi-class inverse-link function. Given a vector f=[f_1, f_2, … f_k], the result of the mapping is

y = [y_1 … y_k]

with

y_i = (1-epsilon) i == argmax(f): epsilon/(k-1) otherwise

where k is the number of classes.

Attributes

eps_k1
parameters
trainable_parameters

Methods

__call__(F)

Call self as a function.

prob_is_largest
safe_sqrt

__init__(num_classes, epsilon=0.001, **kwargs)[source]¶: epsilon represents the fraction of ‘errors’ in the labels of the dataset. This may be a hard parameter to optimize, so by default it is set un-trainable, at a small value.

gpflow.likelihoods.ScalarLikelihood¶

class gpflow.likelihoods.ScalarLikelihood(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.QuadratureLikelihood

A likelihood class that helps with scalar likelihood functions: likelihoods where each scalar latent function is associated with a single scalar observation variable.

If there are multiple latent functions, then there must be a corresponding number of data: we check for this.

The Likelihood class contains methods to compute marginal statistics of functions of the latents and the data ϕ(y,f):

variational_expectations: ϕ(y,f) = log p(y|f)

predict_log_density: ϕ(y,f) = p(y|f)

Those statistics are computed after having first marginalized the latent processes f under a multivariate normal distribution q(f) that is fully factorized.

Some univariate integrals can be done by quadrature: we implement quadrature routines for 1D integrals in this class, though they may be overwritten by inheriting classes where those integrals are available in closed form.

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

_check_last_dims_valid(F, Y)[source]¶: Assert that the dimensions of the latent functions and the data are compatible :param F: function evaluation Tensor, with shape […, latent_dim] :param Y: observation Tensor, with shape […, latent_dim]

_log_prob(F, Y)[source]¶: Compute log p(Y|F), where by convention we sum out the last axis as it represented independent latent functions and observations. :param F: function evaluation Tensor, with shape […, latent_dim] :param Y: observation Tensor, with shape […, latent_dim]

_quadrature_log_prob(F, Y)[source]¶

Returns the appropriate log prob integrand for quadrature.

Quadrature expects f(X), here logp(F), to return shape [N_quad_points] + batch_shape + [d’]. Here d’ corresponds to the last dimension of both F and Y, and _scalar_log_prob simply broadcasts over this.

Also see _quadrature_reduction.

_quadrature_reduction(quadrature_result)[source]¶

Converts the quadrature integral appropriately.

The return shape of quadrature is batch_shape + [d’]. Here, d’ corresponds to the last dimension of both F and Y, and we want to sum over the observations to obtain the overall predict_log_density or variational_expectations.

Also see _quadrature_log_prob.

gpflow.likelihoods.Softmax¶

class gpflow.likelihoods.Softmax(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.MonteCarloLikelihood

The soft-max multi-class likelihood. It can only provide a stochastic Monte-Carlo estimate of the variational expectations term, but this added variance tends to be small compared to that due to mini-batching (when using the SVGP model).

Attributes

parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

gpflow.likelihoods.StudentT¶

class gpflow.likelihoods.StudentT(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

__init__(scale=1.0, df=3.0, **kwargs)[source]¶

Parameters

float (df) – scale parameter
float – degrees of freedom

gpflow.likelihoods.SwitchedLikelihood¶

class gpflow.likelihoods.SwitchedLikelihood(*args: Any, **kwargs: Any)[source]¶

Bases: gpflow.likelihoods.base.ScalarLikelihood

Attributes

num_gauss_hermite_points
parameters
trainable_parameters

Methods

`__call__`(args, *kwargs)	Call self as a function.
`conditional_mean`(F)	The conditional mean of Y\|F: [E[Y₁\|F], …, E[Yₖ\|F]] where K = observation_dim
`conditional_variance`(F)	The conditional marginal variance of Y\|F: [var(Y₁\|F), …, var(Yₖ\|F)] where K = observation_dim
`log_prob`(F, Y)	The log probability density log p(Y\|F)
`predict_density`(Fmu, Fvar, Y)	Deprecated: see predict_log_density
`predict_log_density`(Fmu, Fvar, Y)	Given a Normal distribution for the latent function, and a datum Y, compute the log predictive density of Y,
`predict_mean_and_var`(Fmu, Fvar)	Given a Normal distribution for the latent function, return the mean and marginal variance of Y,
`variational_expectations`(Fmu, Fvar, Y)	Compute the expected log density of the data, given a Gaussian distribution for the function values,

__init__(likelihood_list, **kwargs)[source]¶: In this likelihood, we assume at extra column of Y, which contains integers that specify a likelihood from the list of likelihoods.

_partition_and_stitch(args, func_name)[source]¶

args is a list of tensors, to be passed to self.likelihoods.<func_name>

args[-1] is the ‘Y’ argument, which contains the indexes to self.likelihoods.

This function splits up the args using dynamic_partition, calls the relevant function on the likelihoods, and re-combines the result.

gpflow.likelihoods.utils¶

gpflow.likelihoods.utils