Y: Observation functions.
In dMod: Dynamic Modeling and Parameter Estimation in ODE Models

Description Usage Arguments Details Value Examples

Creates an object of type obsfn that evaluates an observation function and its derivatives based on the output of a model prediction function, see prdfn, as e.g. produced by Xs.

Y(
  g,
  f = NULL,
  states = NULL,
  parameters = NULL,
  condition = NULL,
  attach.input = TRUE,
  deriv = TRUE,
  compile = FALSE,
  modelname = NULL,
  verbose = FALSE
)

`g`	Named character vector or equation vector defining the observation function
`f`	Named character of equations or object that can be converted to eqnvec or object of class fn. If f is provided, states and parameters are guessed from f.
`states`	character vector, alternative definition of "states", usually the names of `f`. If both, f and states are provided, the states argument overwrites the states derived from f.
`parameters`	character vector, alternative definition of the "parameters", usually the symbols contained in "g" and "f" except for `states` and the code word `time`. If both, f and parameters are provided, the parameters argument overwrites the parameters derived from f and g.
`condition`	either NULL (generic prediction for any condition) or a character, denoting the condition for which the function makes a prediction.
`attach.input`	logical, indiating whether the original input should be returned with the output.
`deriv`	logical, generate function to evaluate derivatives of observables. Necessary for parameter estimation.
`compile`	Logical, compile the function (see funC0)
`modelname`	Character, used if `compile = TRUE`, sets a fixed filename for the C file.
`verbose`	Print compiler output to R command line.

For odemodels with forcings, it is best, to pass the prediction function x to the "f"-argument instead of the equations themselves. If an eqnvec is passed to "f" in this case, the forcings and states have to be specified manually via the "states"-argument.

Object of class obsfn, i.e. a function y(..., deriv = TRUE, conditions = NULL) representing the evaluation of the observation function. Arguments out (model prediction) and pars (parameter values) shoudl be passed by the ... argument. If out has the attribute "sensitivities", the result of y(out, pars), will have an attributed "deriv" which reflecs the sensitivities of the observation with respect to the parameters. If pars is the result of a parameter transformation p(pars) (see P), the Jacobian of the parameter transformation and the sensitivities of the observation function are multiplied according to the chain rule for differentiation.

# Define a time grid on which to make a prediction by peace-wise linear function.
# Then define a (generic) prediction function based on thid grid.
times <- 0:5
grid <- data.frame(name = "A", time = times, row.names = paste0("p", times))
x <- Xd(grid)

# Define an observable and an observation function
observables <- eqnvec(Aobs = "s*A")
g <- Y(g = observables, f = NULL, states = "A", parameters = "s")

# Collect parameters and define an overarching parameter transformation
# for two "experimental condtions".
dynpars <- attr(x, "parameters")
obspars <- attr(g, "parameters")
innerpars <- c(dynpars, obspars)

trafo <- structure(innerpars, names = innerpars)
trafo_C1 <- replaceSymbols(innerpars, paste(innerpars, "C1", sep = "_"), trafo)
trafo_C2 <- replaceSymbols(innerpars, paste(innerpars, "C2", sep = "_"), trafo)

p <- NULL
p <- p + P(trafo = trafo_C1, condition = "C1")
p <- p + P(trafo = trafo_C2, condition = "C2")

# Collect outer (overarching) parameters and 
# initialize with random values
outerpars <- attr(p, "parameters")
pars <- structure(runif(length(outerpars), 0, 1), names = outerpars)

# Predict internal/unobserved states
out1 <- (x*p)(times, pars)
plot(out1)

# Predict observed states in addition to unobserved
out2 <- (g*x*p)(times, pars)
plot(out2)