ssqOdeModel {simecol}

Sum of Squares Between odeModel and Data

Description

Compute the sum of squares between a given data and an odeModel object.

Usage

ssqOdeModel(p, simObj, obstime, yobs, 
  sd.yobs = as.numeric(lapply(yobs, sd)), 
  initialize = TRUE, lower. = -Inf, upper. = Inf, weights = NULL,
  debuglevel = 0, ..., pnames = NULL)

Arguments

p

vector of named parameter values of the model (can be a subset),

simObj

a valid object of class odeModel,

obstime

vector with time steps for which observational data are available,

yobs

data frame with observational data for all or a subset of state variables. Their names must correspond exacly with existing names of state variables in the odeModel.

sd.yobs

vector of given standard deviations for all observational variables given in yobs. If no standard deviations are given, these are estimated from yobs.

initialize

optional boolean value whether the simObj should be re-initialized after the assignment of new parameter values. This can be necessary in certain models to assign consistent values to initial state variables if they depend on parameters.

lower., upper.

named vectors with lower and upper bounds used in the optimisation,

weights

optional weights to be used in the fitting process. Should be NULL or a data frame with the same structure as yobs. If non-NULL, weighted least squares is used with weights (that is, minimizing sum(w*e^2)); otherwise ordinary least squares is used.

debuglevel

a positive number that specifies the amount of debugging information printed,

...

additional parameters passed to the solver method (e.g. lsoda),

pnames

names of the parameters, optionally passed from fitOdeModel. This argument is a workaround for R versions below 2.8.1. It may be removed in future versions of simecol.

Details

This is the default function called by function fitOdeModel. The source code of this function can be used as a starting point to develop user-defined optimization criteria (cost functions).

Value

The sum of squared differences between yobs and simulation, by default weighted by the inverse of the standard deviations of the respective variables.

See Also

fitOdeModel, optim, p.constrain

Examples

data(chemostat)
cs1 <- chemostat

## generate some noisy data
parms(cs1)[c("vm", "km")] <- c(2, 10)
times(cs1) <- c(from = 0, to = 20, by = 2)
yobs <- out(sim(cs1))
obstime <- yobs$time
yobs$time <- NULL
yobs$S <- yobs$S + rnorm(yobs$S, sd = 0.1 * sd(yobs$S))*2
yobs$X <- yobs$X + rnorm(yobs$X, sd = 0.1 * sd(yobs$X))

## SSQ between model and data
ssqOdeModel(NULL, cs1, obstime, yobs)

## SSQ between model and data, different parameter set
ssqOdeModel(p=c(vm=1, km=2), cs1, obstime, yobs)

## SSQ between model and data, downweight second observation
## (both variables)
weights <- data.frame(X=rep(1, nrow(yobs)), S = rep(1, nrow=(yobs)))
ssqOdeModel(p=c(vm=1, km=2), cs1, obstime, yobs, weights=weights)

## downweight 3rd data set (row)
weights[3,] <- 0.1
ssqOdeModel(p=c(vm=1, km=2), cs1, obstime, yobs, weights=weights)

## give one value double weight (e.g. 4th value of S)
weights$S[4] <- 2
ssqOdeModel(p=c(vm=1, km=2), cs1, obstime, yobs, weights=weights)