gradNegL: Calculates the gradient of -l(x)

In dajmcdon/spasm: Implements the Laplace-Gaussian filter for sparse, additive, state-space models

Description

Calculates the gradient of -l(x)

Usage

gradNegL(x, prev.state, y, out.spam, HHinv)

Arguments

x	p-vector of current states (to minimize over)
prev.state	list containing mean and variance of \hat{p}(x_{t-1} | y_{t-1}) (as produced by posteriorPredictive
y	d-vector of observations at time t
out.spam	list of fitted splines for the observation equation, as produced by spam
HHinv	inverse of the observation error variance

Details

This function calculates

-l'(x) = d/dx -log p(y_{t} | x_{t}) p(x_{t} | y_{1:t-1}).

As the second distribution in the above product is approximated with a Gaussian, the result is also an approximation.

Value

The gradient of the (approximate) -l'(x), (p-vector)

dajmcdon/spasm documentation built on May 6, 2019, 1:31 a.m.