In lhjohn/EmcDementiaPredictionBase: A Skeleton for Prediction Development Studies that use the PatientLevelPrediction Framework

Analysis

The model that will be trained is a r attr(modelSettings$param,'settings')$name that uses the PatientLevelPrediction function r modelSettings$fitFunction to fit the model.

if(modelSettings$fitFunction == "fitCyclopsModel"){
 parameters <- data.frame(
   name = names(modelSettings$param),
   value = unlist(
     lapply(modelSettings$param, function(x) paste(names(x), x, collapse = '-', sep=':'))
   )
 )

 settings <- data.frame(
   name = names(attr(modelSettings$param,"settings")),
   value = unlist(
     lapply(
       attr(modelSettings$param,"settings"), 
       function(x) paste0(names(x), x, collapse = ':', sep=' ')
            )
     )
 )
 row.names(settings) <- NULL
} else{
   parameters <- 
       do.call('rbind', lapply(
       modelSettings$param, 
       function(x){
         unlist(lapply(x, function(x) paste0(x, sep=' ', collapse=':')))
       })
     )

 settings <- data.frame(
   name = names(attr(modelSettings$param,"settings")),
   value = unlist(
     lapply(
       attr(modelSettings$param,"settings"), 
       function(x) paste0(names(x), x, collapse = '-', sep='')
            )
     )
 )
 row.names(settings) <- NULL
}

Cross-validation settings

The cross validation settings are to use r splitSettings$nfold folds in the training data that are partitioned using the r attr(splitSettings,"fun") function and consist of r splitSettings$train*100 \% of the complete data. The seed used for splitting the data is r splitSettings$seed.

Hyper-parameter search

The hyper-parameters investigated while fitting the model are listed below. The combination of hyper-parameters that obtains the highest AUROC value in the training data via cross validation will be uses in the final model.

print(knitr::kable(x = parameters, caption = paste('Hyper-parameters combinations searched to fit the model')))

Other settings

The other settings used to fit the model, such as seeds used for reproducibility, are:

print(knitr::kable(x = settings, caption = paste('Other model fitting settings')))

Internal validation

The model will be assessed internally using a test set that consists of r splitSettings$test*100 \% of the complete data.

lhjohn/EmcDementiaPredictionBase documentation built on March 25, 2022, 12:22 a.m.