tfprobability

GLM families

Runs multiple Fisher scoring steps

Runs multiple Fisher scoring steps

Runs one Fisher scoring step

Runs one Fisher Scoring step

Blockwise Initializer

Installs TensorFlow Probability

Masked Autoencoder for Distribution Estimation

An autoregressive normalizing flow layer, given a layer_autoregressive.

A OneHotCategorical mixture Keras layer from k * (1 + d) params.

1D convolution layer (e.g. temporal convolution) with Flipout

1D convolution layer (e.g. temporal convolution).

2D convolution layer (e.g. spatial convolution over images) with Flipout

2D convolution layer (e.g. spatial convolution over images)

3D convolution layer (e.g. spatial convolution over volumes) with Flipout

3D convolution layer (e.g. spatial convolution over volumes)

Densely-connected layer class with Flipout estimator.

Densely-connected layer class with local reparameterization estimator.

Densely-connected layer class with reparameterization estimator.

Dense Variational Layer

Keras layer enabling plumbing TFP distributions through Keras models

An Independent-Bernoulli Keras layer from prod(event_shape) params

An independent Logistic Keras layer.

An independent Normal Keras layer.

An independent Poisson Keras layer.

Pass-through layer that adds a KL divergence penalty to the model loss

Regularizer that adds a KL divergence penalty to the model loss

A mixture distribution Keras layer, with independent logistic components.

A mixture distribution Keras layer, with independent normal components.

A mixture (same-family) Keras layer.

A d-variate Multivariate Normal TriL Keras layer from d+d*(d+1)/ 2 params

A d-variate OneHotCategorical Keras layer from d params.

Variable Layer

A Variational Gaussian Process Layer.

Adapts the inner kernel's step_size based on log_accept_prob.

Estimate a lower bound on effective sample size for each independent chain.

Runs one step of Hamiltonian Monte Carlo.

Runs one step of Metropolis-adjusted Langevin algorithm.

Runs one step of the Metropolis-Hastings algorithm.

Runs one step of the No U-Turn Sampler

Gelman and Rubin (1992)'s potential scale reduction for chain convergence.

Runs one step of the RWM algorithm with symmetric proposal.

Runs one step of the Replica Exchange Monte Carlo

Runs annealed importance sampling (AIS) to estimate normalizing constants.

Implements Markov chain Monte Carlo via repeated TransitionKernel steps.

Returns a sample from the dim dimensional Halton sequence.

Adapts the inner kernel's step_size based on log_accept_prob.

Runs one step of the slice sampler using a hit and run approach

Applies a bijector to the MCMC's state space

Runs one step of Uncalibrated Hamiltonian Monte Carlo

Runs one step of Uncalibrated Langevin discretized diffusion.

Generate proposal for the Random Walk Metropolis algorithm.

number of params needed to create a CategoricalMixtureOfOneHotCategorical distribution

number of params needed to create an IndependentBernoulli distribution

number of params needed to create an IndependentLogistic distribution

number of params needed to create an IndependentNormal distribution

number of params needed to create an IndependentPoisson distribution

number of params needed to create a MixtureLogistic distribution

number of params needed to create a MixtureNormal distribution

number of params needed to create a MixtureSameFamily distribution

number of params needed to create a MultivariateNormalTriL distribution

number of params needed to create a OneHotCategorical distribution

Objects exported from other packages

A state space model representing a sum of component state space models.

Formal representation of an autoregressive model.

State space model for an autoregressive process.

Build a variational posterior that factors over model parameters.

Build a loss function for variational inference in STS models.

Seasonal state space model with effects constrained to sum to zero.

Decompose an observed time series into contributions from each component.

Decompose a forecast distribution into contributions from each component.

Formal representation of a dynamic linear regression model.

State space model for a dynamic linear regression from provided covariates.

Draw posterior samples using Hamiltonian Monte Carlo (HMC)

Construct predictive distribution over future observations

Formal representation of a linear regression from provided covariates.

Formal representation of a local level model

State space model for a local level

Formal representation of a local linear trend model

State space model for a local linear trend

Compute one-step-ahead predictive distributions for all timesteps

Initialize from a uniform [-2, 2] distribution in unconstrained space.

Formal representation of a seasonal effect model.

State space model for a seasonal effect.

Formal representation of a semi-local linear trend model.

State space model for a semi-local linear trend.

Formal representation of a smooth seasonal effect model

State space model for a smooth seasonal effect

Formal representation of a sparse linear regression.

Sum of structural time series components.

ComputesY = g(X) = Abs(X), element-wise

Affine bijector

ComputesY = g(X; shift, scale) = scale @ X + shift

AffineScalar bijector

ComputesY = g(X) s.t. X = g^-1(Y) = (Y - mean(Y)) / std(Y)

Bijector which applies a list of bijectors to blocks of a Tensor

Bijector which applies a sequence of bijectors

Computesg(X) = X @ X.T where X is lower-triangular, positive-diagonal matrix

Maps the Cholesky factor of M to the Cholesky factor of M^{-1}

Maps unconstrained reals to Cholesky-space correlation matrices.

Computes the cumulative sum of a tensor along a specified axis.

ComputesY = g(X) = DCT(X), where DCT type is indicated by the type arg

ComputesY=g(X)=exp(X)

ComputesY = g(X) = exp(X) - 1

Implements a continuous normalizing flow X->Y defined via an ODE.

Transforms unconstrained vectors to TriL matrices with positive diagonal

Transforms vectors to triangular

Returns the forward Bijector evaluation, i.e., X = g(Y).

Returns the result of the forward evaluation of the log determinant of the Jacobian

Compute Y = g(X) = 1 - exp(-c * (exp(rate * X) - 1), the Gompertz CDF.

ComputesY = g(X) = exp(-exp(-(X - loc) / scale))

Compute Y = g(X) = exp(-exp(-(X - loc) / scale)), the Gumbel CDF.

ComputesY = g(X) = X

Bijector constructed from custom functions

Returns the inverse Bijector evaluation, i.e., X = g^{-1}(Y).

Returns the result of the inverse evaluation of the log determinant of the Jacobian

Bijector which inverts another Bijector

Bijector which applies a Stick Breaking procedure.

ComputesY = g(X) = (1 - (1 - X)**(1 / b))**(1 / a), with X in [0, 1]

ComputesY = g(X) = (1 - (1 - X)**(1 / b))**(1 / a), with X in [0, 1]

LambertWTail transformation for heavy-tail Lambert W x F random variables.

Masked Autoregressive Density Estimator

Affine MaskedAutoregressiveFlow bijector

Autoregressively masked dense layer

Computes g(L) = inv(L), where L is a lower-triangular matrix

Matrix-vector multiply using LU decomposition

ComputesY = g(X) = NormalCDF(x)

Bijector which maps a tensor x_k that has increasing elements in the last dimension to an unconstrained tensor y_k

Pads a value to the event_shape of a Tensor.

Permutes the rightmost dimension of a Tensor

ComputesY = g(X) = (1 + X * c)**(1 / c), where X >= -1 / c

A piecewise rational quadratic spline, as developed in Conor et al.(2019).

RealNVP affine coupling layer for vector-valued events

Build a scale-and-shift function using a multi-layer neural network

A Bijector that computes b(x) = 1. / x

Reshapes the event_shape of a Tensor

Compute Y = g(X; scale) = scale * X.

Compute Y = g(X; scale) = scale @ X

Compute Y = g(X; scale) = scale @ X.

Matrix-vector multiply using LU decomposition.

Compute Y = g(X; scale) = scale @ X.

Transforms unconstrained vectors to TriL matrices with positive diagonal

Compute Y = g(X; shift) = X + shift.

Compute Y = g(X) = (1 - exp(-rate * X)) * exp(-c * exp(-rate * X))

ComputesY = g(X) = 1 / (1 + exp(-X))

Bijector that computes Y = sinh(X).

ComputesY = g(X) = Sinh( (Arcsinh(X) + skewness) * tailweight )

Computes Y = g(X) = exp([X 0]) / sum(exp([X 0]))

Computes Y = g(X) = Log[1 + exp(X)]

Computes Y = g(X) = X / (1 + |X|)

Split a Tensor event along an axis into a list of Tensors.

Computesg(X) = X^2; X is a positive real number.

Computes Y = tanh(X)

Applies a Bijector to the diagonal of a matrix

ComputesY = g(X) = transpose_rightmost_dims(X, rightmost_perm)

ComputesY = g(X) = 1 - exp((-X / scale) ** concentration) where X >= 0

Compute Y = g(X) = 1 - exp((-X / scale) ** concentration), X >= 0.

Autoregressive distribution

Batch-Reshaping distribution

Bates distribution.

Bernoulli distribution

Beta distribution

Beta-Binomial compound distribution

Binomial distribution

Blockwise distribution

Categorical distribution over integers

Cauchy distribution with location loc and scale scale