Package 'iwmm'

Title: Importance weighted moment matching
Description: iwmm provides functions for adaptive importance sampling.
Authors: Topi Paananen [cre, aut], Noa Kallioinen [ctb]
Maintainer: Topi Paananen <[email protected]>
License: GPL (>=3)
Version: 0.0.1
Built: 2024-11-07 05:43:14 UTC
Source: https://github.com/topipa/iwmm

Help Index

Constrain all draws from a fitted model

Description

Constrain all draws from a fitted model

Usage

constrain_draws(x, ...)

Arguments

x

model fit object
...

arguments passed to methods

Value

constrained draws

Example Stan model for importance weighted moment matching

Description

Provides example models (with data) that are ready for use with IWMM.

Usage

example_iwmm_model(model = "normal_model")

Arguments

model

Character specifying which model code to return. Currently "normal_model" is implemented.

Value

List containing model code and corresponding data.

Return log probability of posterior

Description

Return log probability of posterior

Usage

log_prob_draws(fit, ...)

Arguments

fit

model fit object
...

arguments passed to methods

Value

TODO

Generic importance weighted moment matching algorithm.

Description

Generic importance weighted moment matching algorithm.

Generic importance weighted moment matching algorithm for matrices.

Usage

moment_match(x, ...)

## S3 method for class 'matrix'
moment_match(
  x,
  log_prob_prop_fun,
  log_prob_target_fun = NULL,
  log_ratio_fun = NULL,
  expectation_fun = NULL,
  log_expectation_fun = FALSE,
  draws_transformation_fun = NULL,
  is_method = "psis",
  adaptation_method = "iwmm",
  k_threshold = 0.5,
  cov_transform = TRUE,
  split = FALSE,
  restart_transform = FALSE,
  ...
)

Arguments

x

A matrix of draws. Must be unconstrained.
...

Further arguments passed to log_prob_prop_fun, log_prob_target_fun and log_ratio_fun.
log_prob_prop_fun

Log density of the proposal. The function takes argument draws.
log_prob_target_fun

Log density of the target for importance sampling. The function takes argument draws.
log_ratio_fun

Log of the density ratio for importance sampling (target/proposal). The function takes argument draws.
expectation_fun

Optional argument, NULL by default. A function whose expectation is being computed. The function takes arguments draws.
log_expectation_fun

Logical indicating whether the expectation_fun returns its values as logarithms or not. Defaults to FALSE. If set to TRUE, the expectation function must be nonnegative (before taking the logarithm). Ignored if expectation_fun is NULL.
draws_transformation_fun

Optional argument, NULL by default. A function that transforms draws before computing expectation. The function takes arguments draws.
is_method

Which importance sampling method to use. Currently only psis is supported.
adaptation_method

Which adaptation method to use. Currently only iwmm is supported.
k_threshold

Threshold value for Pareto k values above which the moment matching algorithm is used. The default value is 0.5.
cov_transform

Logical; Indicates whether to match the covariance of the samples or not. If FALSE, only the mean and marginal variances are matched. Default is TRUE.
split

Logical; Indicate whether to do the split transformation or not at the end of moment matching. FALSE by default.
restart_transform

Logical; When split is TRUE, indicates whether to start the second transformation from the original model parameters or the transformed parameters. If split is FALSE, this is ignored.

Value

Returns a list with: transformed draws, updated importance weights, and the pareto k diagnostic value. If expectation_fun is given, also returns the expectation.

Generic importance weighted moment matching algorithm for brmsfit objects. See additional arguments from moment_match.matrix

Description

Generic importance weighted moment matching algorithm for brmsfit objects. See additional arguments from moment_match.matrix

Usage

## S3 method for class 'brmsfit'
moment_match(
  x,
  log_prob_target_fun = NULL,
  log_ratio_fun = NULL,
  target_observation_weights = NULL,
  expectation_fun = NULL,
  log_expectation_fun = FALSE,
  constrain = TRUE,
  ...
)

Arguments

x

A fitted brmsfit object.
log_prob_target_fun

Log density of the target. The function takes argument draws, which are the unconstrained draws. Can also take the argument fit which is the stan model fit.
log_ratio_fun

Log of the density ratio (target/proposal). The function takes argument draws, which are the unconstrained draws. Can also take the argument fit which is the stan model fit.
target_observation_weights

A vector of weights for observations for defining the target distribution. A value 0 means dropping the observation, a value 1 means including the observation similarly as in the current data, and a value 2 means including the observation twice.
expectation_fun

Optional argument, NULL by default. A function whose expectation is being computed. The function takes arguments draws.
log_expectation_fun

Logical indicating whether the expectation_fun returns its values as logarithms or not. Defaults to FALSE. If set to TRUE, the expectation function must be nonnegative (before taking the logarithm). Ignored if expectation_fun is NULL.
constrain

Logical specifying whether to return draws on the constrained space? Default is TRUE.
...

Further arguments passed to moment_match.matrix.

Value

Returns a list with 3 elements: transformed draws, updated importance weights, and the pareto k diagnostic value. If expectation_fun is given, also returns the expectation.

Generic importance weighted moment matching algorithm for CmdStanFit objects. See additional arguments from moment_match.matrix

Description

Generic importance weighted moment matching algorithm for CmdStanFit objects. See additional arguments from moment_match.matrix

Usage

## S3 method for class 'CmdStanFit'
moment_match(
  x,
  log_prob_target_fun = NULL,
  log_ratio_fun = NULL,
  constrain_draws = TRUE,
  ...
)

Arguments

x

A fitted CmdStanFit object.
log_prob_target_fun

Log density of the target. The function takes argument draws, which are the unconstrained draws.
log_ratio_fun

Log of the density ratio (target/proposal). The function takes argument draws, which are the unconstrained draws.
constrain_draws

Logical specifying whether to return draws on the constrained space. Draws are also constrained for computing expectations. Default is TRUE.
...

Further arguments passed to moment_match.matrix.

Value

Returns a list with 3 elements: transformed draws, updated importance weights, and the pareto k diagnostic value.

Generic importance weighted moment matching algorithm for stanfit objects. See additional arguments from moment_match.matrix

Description

Generic importance weighted moment matching algorithm for stanfit objects. See additional arguments from moment_match.matrix

Usage

## S3 method for class 'stanfit'
moment_match(
  x,
  log_prob_target_fun = NULL,
  log_ratio_fun = NULL,
  target_observation_weights = NULL,
  expectation_fun = NULL,
  log_expectation_fun = FALSE,
  constrain_draws = TRUE,
  ...
)

Arguments

x

A fitted stanfit object.
log_prob_target_fun

Log density of the target. The function takes argument draws, which are the unconstrained draws. Can also take the argument fit which is the stan model fit.
log_ratio_fun

Log of the density ratio (target/proposal). The function takes argument draws, which are the unconstrained draws. Can also take the argument fit which is the stan model fit.
target_observation_weights

A vector of weights for observations for defining the target distribution. A value 0 means dropping the observation, a value 1 means including the observation similarly as in the current data, and a value 2 means including the observation twice.
expectation_fun

Optional argument, NULL by default. A function whose expectation is being computed. The function takes arguments draws.
log_expectation_fun

Logical indicating whether the expectation_fun returns its values as logarithms or not. Defaults to FALSE. If set to TRUE, the expectation function must be nonnegative (before taking the logarithm). Ignored if expectation_fun is NULL.
constrain_draws

Logical specifying whether to return draws on the constrained space. Draws are also constrained for computing expectations. Default is TRUE.
...

Further arguments passed to moment_match.matrix.

Value

Returns a list with 3 elements: transformed draws, updated importance weights, and the pareto k diagnostic value. If expectation_fun is given, also returns the expectation.

Shift a matrix of draws to their weighted mean.

Description

Shift a matrix of draws to their weighted mean.

Usage

shift(draws, lw)

Arguments

draws

A matrix of draws.
lw

A vector representing the log-weight of each draw.

Value

List with the shift that was performed, and the new draws matrix.

Shift a matrix of draws to their weighted mean and scale the covariance to match the weighted covariance.

Description

Shift a matrix of draws to their weighted mean and scale the covariance to match the weighted covariance.

Usage

shift_and_cov(draws, lw)

Arguments

draws

A matrix of draws.
lw

A vector representing the log-weight of each draw.

Value

List with the shift and mapping that were performed, and the new draws matrix.

Shift a matrix of draws to their weighted mean and scale the marginal variances to match the weighted marginal variances.

Description

Shift a matrix of draws to their weighted mean and scale the marginal variances to match the weighted marginal variances.

Usage

shift_and_scale(draws, lw)

Arguments

draws

A matrix of draws.
lw

A vector representing the log-weight of each draw.

Value

List with the shift and scaling that were performed, and the new draws matrix.

Unconstrain all draws from a fitted model

Description

Unconstrain all draws from a fitted model

Usage

unconstrain_draws(x, ...)

Arguments

x

model fit object
...

arguments passed to methods

Value

unconstrained draws