predict_one: Wrapper for executing the rNCV function.

In kforthman/caretStack: Stacks models from the caret package to create a more robust prediction

Description

Performs repeated nested cross-validation on the input dataset. Is intended for use on LIBR's T1000 data and expects data in this format. Accepts data that includes columns labeled 'id' and 'LC_Category' and will remove these columns before performing the rNCV.

Usage

predict_one(
  dset,
  var_to_predict,
  predictor_var_file_list,
  rdata_prefix,
  outDir = "",
  nFolds.outer = 5,
  methods = c("svmRadial", "ranger", "glmnet"),
  metric = "RMSE"
)

Arguments

dset	the dataset to use in matrix form including the predictors, the target, participant id's, and LifeChart (LC) categories.
var_to_predict	The column name of the target.
predictor_var_file_list	list of filenames. Each file is expected to contain a list of T1000 variable data names that will be included in the analysis as predictors.
rdata_prefix	label to put in output file names
nFolds.outer	Number of outer folds
methods	Similarly to the method argument in caret's train function, this argument is a list of strings specifying which classification or regression models to use. Possible values are found using names(getModelInfo()). See http://topepo.github.io/caret/train-models-by-tag.html. A list of functions can also be passed for a custom model function. See http://topepo.github.io/caret/using-your-own-model-in-train.html for details.
metric	A string that specifies what summary metric will be used to select the optimal model. By default, possible values are "RMSE" and "Rsquared" for regression and "Accuracy" and "Kappa" for classification. If custom performance metrics are used (via the summaryFunction argument in trainControl, the value of metric should match one of the arguments. If it does not, a warning is issued and the first metric given by the summaryFunction is used. (NOTE: If given, this argument must be named.)
ourDir	If you would like to save the output files into somewhere other than the working directory, specify that here. Make sure the folder name ends with '/'.

Details

Missing data: If dataset contains predictors with missing data, the missing entrys will be imputed using KNN imputation. If any subject is missing over 30% of their predictor variables, they are removed from the analysis. Any cases that have missing data for the target variable will be removed. Only 1 target allowed.

Saves a file containing 5 objects:

1. data.rncv is a list object that contains the response variables with the the imputed predictors. Cases with no entry for the response variable are removed. This is the dataset that is plugged into the function rNCV()

2. res.rncv is the object returned from the function rNCV()

3. output_label Label of output file name.

4. predictor_vars A list of the names of the predictors.

5. var_to_predict The name of the target variable.

Examples

prepped_data <- read.csv('Data/prepped_hc_data.csv', stringsAsFactors = F)
prepped_data[prepped_data$LC_Category == 'Dep', 'LC_Category'] <- 'Dep+Anx'
prepped_data[prepped_data$LC_Category == 'Anx', 'LC_Category'] <- 'Dep+Anx'
prepped_data <- prepped_data[which(prepped_data$LC_Category != 'Eating+'),]
prepped_data$LC_Category <- factor(prepped_data$LC_Category)

prepped_data[prepped_data$LC_Category == 'Dep', 'Dep.Anx'] <- 1
prepped_data[prepped_data$LC_Category == 'Anx', 'Dep.Anx'] <- 1

ft_data <- read.csv('Data/FT_summary.csv',  stringsAsFactors = F)

this_data <- merge(prepped_data, ft_data, by = c("id", "visit"), all.x = T)

predict_one(prepped_data, 'lme_slope_simple'  , c('Data/all_vars-clin_np.csv'),   'lme_slope_simple_vars-clin_np')
predict_one(prepped_data, 'lme_slope_simple'  , c('Data/all_vars-clinical.csv'),   'lme_slope_simple_vars-clinical')
predict_one(prepped_data, 'lme_slope_simple'  , c('Data/all_vars-np.csv'),   'lme_slope_simple_vars-np')

kforthman/caretStack documentation built on June 21, 2021, 8:38 a.m.