RCVLassoPar: Function to perform the random split analysis
In angy89/hyQSAR: Hybrid QSAR model

This function perform the random split analysis method to estimate the optimal lambda shrinkage parameter for the LASSO model

1
2
3

RCVLassoPar(y, x, lambda, n.splits = 25, train.prop = 0.9,
  measure = "MSE", intercept = TRUE, type.solution = "lci", nCores = 20,
  th = 1.96)

`y`	is the vector of response variables with same length of the number of samples
`x`	is the matrix of the input dataset with samples on the rows and features on the columns
`lambda`	is a numeric vector of lambda value to be used in the LASSO parameter selection step
`n.splits`	is the number of random split to be performed. Default value is 25
`train.prop`	is the percentage of samples in the dataset to be used as training set. Default value is 0.9
`measure`	is the measure used to perform the choice of the optimal lambda value. Possible values are MSE and R2. Default value is MSE
`intercept`	is a boolean valus indicating if we want to fit or not the intercept. Default valuw is TRUE
`type.solution`	is a string indicating the type of solution to compute. Possible values are: min, uci and lci; if standard min or max of the average CV function. if lci, take the the most parsimonous solution within the tot percentage of confidence bands around the standard solution. if uci a less partimosious solution within the tot percentage of confidence bands around the standard solution is selected
`nCores`	is the number of cores to be used
`th`	is the size of the confidence interval. Default value is 1.96

an object of class RCVLasso containing the following objects:

`cv`	list with matrices (of sizes n.slipts x n.lambda) containing statistics for each lambda and each splitting. mse is the matrix of predictive mse. R2 is the matrix of predictive R2 and active.size is the matrix with active beta in the trained model for each lambda and for each split.
`lambda`	numeric vector of the lambda values taken in input
`fitted`	predicted values
`residuals`	differences between predicted and real values
`intercept`	if the input parameter intercept is TRUE, it is the numeric value of the fitted intercep, otherwise is zero
`beta`	beta coefficients of the fitted model
`opt.lambda`	optimal lambda value
`idx.support`	index of the optimal lambda value
`pred.R2`	R2 of the optimal model on the test sets
`pred.mse`	mse of the optimal model on the test sets
`mse`	mse of the optimal model on the training sets
`R2`	R2 of the optimal model on the training sets
`measure`	measure used to compute the optimal solution
`fitted_model`	model fitted on the whole data