mapestimate

Given data <code>Z</code>, a neural likelihood-to-evidence-ratio <code>estimator</code>, and a <code>prior</code>, computes the implied approximate maximum a posteriori (MAP) estimate
If a vector <code>theta0</code> of initial parameter estimates is given, the approximate posterior density is maximised by gradient descent. Otherwise, if a matrix of parameters <code>theta_grid</code> is given, the approximate posterior density is maximised by grid search.

An 'R' interface to the 'Julia' package 'NeuralEstimators.jl'. The package facilitates the user-friendly development of neural point estimators, which are neural networks that map data to a point summary of the posterior distribution. These estimators are likelihood-free and amortised, in the sense that, after an initial setup cost, inference from observed data can be made in a fraction of the time required by conventional approaches; see Sainsbury-Dale, Zammit-Mangion, and Huser (2024) <doi:10.1080/00031305.2023.2249522> for further details and an accessible introduction. The package also enables the construction of neural networks that approximate the likelihood-to-evidence ratio in an amortised manner, allowing one to perform inference based on the likelihood function or the entire posterior distribution; see Zammit-Mangion, Sainsbury-Dale, and Huser (2024, Sec. 5.2) <doi:10.48550/arXiv.2404.12484>, and the references therein. The package accommodates any model for which simulation is feasible by allowing the user to implicitly define their model through simulated data.

Matthew Sainsbury-Dale

NeuralEstimators

Likelihood-Free Parameter Estimation using Neural Networks

mapestimate function

<dl><dt>estimator</dt>
<dd>a neural likelihood-to-evidence-ratio estimator</dd>
<dt>Z</dt>
<dd>data; it's format should be amenable to the architecture of <code>estimator</code></dd>
<dt>prior</dt>
<dd>the prior (default uniform), specified as a Julia or R function</dd>
<dt>theta_grid</dt>
<dd>a (fine) gridding of the parameter space, given as a matrix with p rows, where p is the number of parameters in the model</dd>
<dt>theta0</dt>
<dd>a vector of initial parameter estimates</dd>
<dt>use_gpu</dt>
<dd>a boolean indicating whether to use the GPU if it is available (default true)</dd></dl>

Arguments

Maximum a posteriori estimation — mapestimate

<dl>

<dt>estimator</dt>
<dd>a neural likelihood-to-evidence-ratio estimator</dd>


<dt>Z</dt>
<dd>data; it's format should be amenable to the architecture of <code>estimator</code></dd>


<dt>prior</dt>
<dd>the prior (default uniform), specified as a Julia or R function</dd>


<dt>theta_grid</dt>
<dd>a (fine) gridding of the parameter space, given as a matrix with p rows, where p is the number of parameters in the model</dd>


<dt>theta0</dt>
<dd>a vector of initial parameter estimates</dd>


<dt>use_gpu</dt>
<dd>a boolean indicating whether to use the GPU if it is available (default true)</dd>

</dl>

mapestimate: Maximum a posteriori estimation

Description

Usage

Value

Arguments

See Also