Easy tuning of the settings of a Monolix project

---

---

Overview

Description

Use a single accuracy level, between 1 and 9, to automatically tune all the settings of a Monolix project.

When the accuray level is equal to 1, the algorithms are very fast but the results may be not precise.

When the accuray level is equal to 9, the algorithms require more time but the results are much more accurate.

Default Monolix settings are obtained with level=5.

Usage

setSettings(project, new.project=NULL, level=5)

Arguments

project

a Monolix project.

new.project

the new created Monolix project (default is the original project)

level

an integer between 1 and 9 (default=5)

Example

Let us use the following PK project which was created with the default settings (level=5)

project <- "projects/warfarinPK1.mlxtran"

We first create a new project that will run extremly fast in Monolix.

library(Rsmlx)
setSettings(project, new.project="projects/warfarinPK1_level1.mlxtran", level=1)

We now modify the settings of the original project in order to get very accurate results:

setSettings(project, level=9)

Using setSettings with lixoftConnectors

setSettingscan be used as a Monolix connector.

library("lixoftConnectors")
initializeLixoftConnectors(software = "monolix")

project <- "projects/warfarinPK2.mlxtran"
loadProject(project)

All the tasks are performed with this project, without linearization:

getScenario()[1:2]

## $tasks
##   populationParameterEstimation conditionalDistributionSampling 
##                            TRUE                            TRUE 
##       conditionalModeEstimation         standardErrorEstimation 
##                            TRUE                            TRUE 
##         logLikelihoodEstimation                           plots 
##                            TRUE                           FALSE 
## 
## $linearization
## [1] FALSE

New settings are automatically defined with level=1:

setSettings(level=1)
unlist(getPopulationParameterEstimationSettings())

##          exploratoryautostop            smoothingautostop 
##                       "TRUE"                       "TRUE" 
##          nbburningiterations      nbexploratoryiterations 
##                          "5"                        "500" 
##        nbsmoothingiterations           simulatedannealing 
##                         "50"                       "TRUE" 
## simulatedannealingiterations             exploratoryalpha 
##                       "1000"                          "0" 
##               smoothingalpha          exploratoryinterval 
##                        "0.7"                         "50" 
##            smoothinginterval               smoothingratio 
##                         "20"                        "0.1" 
##                  variability                     tauomega 
##                       "none"                        "0.9" 
##                tauerrormodel     nboptimizationiterations 
##                        "0.9"                         "10" 
##        optimizationtolerance 
##                      "5e-04"

unlist(getStandardErrorEstimationSettings())

## miniterations maxiterations 
##            50           200

unlist(getConditionalDistributionSamplingSettings())

##       nbminiterations nbsimulatedparameters                 ratio 
##                  20.0                   5.0                   0.1

unlist(getLogLikelihoodEstimationSettings())

##       nbfixediterations        nbfreedomdegrees freedomdegreessampling1 
##                  "5000"                     "5"                     "1" 
## freedomdegreessampling2 freedomdegreessampling3 freedomdegreessampling4 
##                     "2"                     "5"                    "10" 
## freedomdegreessampling5          samplingmethod 
##                    "15"                 "fixed"

Running the complete sequence of tasks is extremely fast with these settings:

ptm <- proc.time()
runScenario()
print(proc.time() - ptm)

##    user  system elapsed 
##    6.77    0.27    2.10

Here are some results obtained with level=1:

getEstimatedPopulationParameters()

##     ka_pop      V_pop     Cl_pop   omega_ka    omega_V   omega_Cl 
## 0.60725714 7.70495328 0.13348864 0.76359681 0.20726923 0.28274915 
##         a1         b1 
## 0.58293777 0.06724816

getEstimatedStandardErrors()

## $stochasticApproximation
##     ka_pop      V_pop     Cl_pop   omega_ka    omega_V   omega_Cl 
## 0.14388645 0.32703877 0.00704306 0.24386013 0.02954997 0.03890614 
##         a1         b1 
## 0.09777818 0.01702185

getEstimatedLogLikelihood()

## $importanceSampling
##          -2LL standardError           AIC          BICc           BIC 
##    850.472426      0.184638    866.472426    888.416575    878.198313

Algorithms require much more time with the settings defined with level=9

setSettings(level=9)
unlist(getPopulationParameterEstimationSettings())

##          exploratoryautostop            smoothingautostop 
##                       "TRUE"                       "TRUE" 
##          nbburningiterations      nbexploratoryiterations 
##                          "5"                       "1000" 
##        nbsmoothingiterations           simulatedannealing 
##                       "1000"                       "TRUE" 
## simulatedannealingiterations             exploratoryalpha 
##                       "1000"                          "0" 
##               smoothingalpha          exploratoryinterval 
##                        "0.7"                        "300" 
##            smoothinginterval               smoothingratio 
##                        "200"                        "0.1" 
##                  variability                     tauomega 
##                       "none"                       "0.98" 
##                tauerrormodel     nboptimizationiterations 
##                       "0.98"                         "40" 
##        optimizationtolerance 
##                      "5e-05"

unlist(getStandardErrorEstimationSettings())

## miniterations maxiterations 
##           200           400

unlist(getConditionalDistributionSamplingSettings())

##       nbminiterations nbsimulatedparameters                 ratio 
##                200.00                 25.00                  0.01

unlist(getLogLikelihoodEstimationSettings())

##       nbfixediterations        nbfreedomdegrees freedomdegreessampling1 
##                 "50000"                     "5"                     "1" 
## freedomdegreessampling2 freedomdegreessampling3 freedomdegreessampling4 
##                     "2"                     "5"                    "10" 
## freedomdegreessampling5          samplingmethod 
##                    "15"                 "fixed"

ptm <- proc.time()
runScenario()
print(proc.time() - ptm)

##    user  system elapsed 
##   92.60    1.36   27.61

Of course, results obtained with level=9 are supposed to be ``better’’ than those obtained with level=1

getEstimatedPopulationParameters()

##     ka_pop      V_pop     Cl_pop   omega_ka    omega_V   omega_Cl 
## 0.58442519 7.76791533 0.13350195 0.76676366 0.22021213 0.28190062 
##         a1         b1 
## 0.55669386 0.07175258

getEstimatedStandardErrors()

## $stochasticApproximation
##      ka_pop       V_pop      Cl_pop    omega_ka     omega_V    omega_Cl 
## 0.151296033 0.350195987 0.007052124 0.210525231 0.036307467 0.039493863 
##          a1          b1 
## 0.094964592 0.017450940

getEstimatedLogLikelihood()

## $importanceSampling
##          -2LL standardError           AIC          BICc           BIC 
##  850.20843363    0.02635225  866.20843363  888.15258302  877.93432085