importanceSamplingEvaluation: Evaluate a network object's predictions on newdata against...

In davharris/mistnet: Stochastic Neural Networks

Description

Estimate the model's log-likelihood on data held out of sample. Rather than storing a huge number of sample predictions in memory, we can do this in #' batches of a specified size and just work with the scalar likelihood.

Usage

importanceSamplingEvaluation(object, newdata, y, loss, batches, batch.size)

Arguments

object	a network object
newdata	a matrix of predictor variables.
y	a matrix of response variables
loss	a loss function (e.g. from a loss object), assumed to be a *negative* log-likelihood
batches	the number of importance sampling batches to perform
batch.size	the number of importance samples per batch

davharris/mistnet documentation built on May 14, 2019, 9:28 p.m.