pmxNODE (version 0.1.0)

nn_converter_nlmixr: NN converter for nlmixr

Description

This function converts a nlmixr model function that includes pseudo-functions for NNs as described in Details into a model function that can be used in the nlmixr function.

Usage

nn_converter_nlmixr(
  f_ode,
  pop_only = FALSE,
  theta_scale = 0.1,
  eta_scale = 0.1,
  pre_fixef = NULL,
  seed = 1908
)

Value

A converted nlmixr model function

Arguments

f_ode

(nlmixr model function) Model function of nlmixr type with NN functions

pop_only

(boolean) If the generated nlmixr model function should be a fit without (TRUE) or with (FALSE) inter-individual variability on NN parameters

theta_scale

(numeric) Scale in which typical NN parameter values are initialized, default is 0.1, i.e., weights are initialized between -0.3 and 0.3

eta_scale

(numeric) Initial standard deviation of random effects on NN parameters, default is 0.1

pre_fixef

(named vector) Specific initial values for typical parameters, can be obtained from a run nlmixr model (e.g., run_1) through run_1$fixef

seed

(numeric) Seed for random parameter initialization.

Author

Dominic Bräm

Details

• state= defines the state to be used in the NN. For time, use t.

• min_init= defines the minimal activation point for the NN, i.e., minimal expected state

• max_init= defines the maximal activation point for the NN, i.e., maximal expected state

• n_hidden= (optional) defines the number of neurons in the hidden layer, default is 5

• act= (optional) defines activation function in the hidden layer, ReLU and Softplus implemented, default is ReLU

• time_nn= (optional) defines whether the NN should be assumed to be a time-dependent NN and consequently all weights from input to hidden layer should be strictly negative.

Examples

if (FALSE) {
f_ode_pop <- function(){
ini({
  lV <- 1
  prop.err <- 0.1
})
model({
  V <- lV
  d/dt(centr)  =  NN1(state=centr,min_init=0,max_init=300)
  cp = centr / V
  cp ~ prop(prop.err)
  })
}

f_new_pop <- nn_converter_nlmixr(f_ode_pop,pop_only = TRUE)

fit_pop <- nlmixr2(f_new_pop,data,est="bobyqa")

f_ode <- function(){
  ini({
    lV <- 1
    eta.V ~ 0.1
    prop.err <- 0.1
  })
  model({
    V <- lV * exp(eta.V)
    d/dt(centr)  =  NN1(state=centr,min_init=0,max_init=300)
    cp = centr / V
    cp ~ prop(prop.err)
  })
}

f_new <- nn_converter_nlmixr(f_ode,pop_only = FALSE, pre_fixef = fit_pop$fixef)
}