create_groups: Random Grouping of Predictors

In bips-hb/rgp: Identification of Risk Groups in Pharmacovigilance Using Penalized Regression and Machine Learning (RGP)

Description

A function to create a list of random variables clustered based on a predefined cluster size. Currently, cluster sizes are based on pathway size curated from KEGG database load(pathway_dr_ds_key_value), and stored as vector of integers (outlier excluded cluster_size <- cluster_size[-1]). In case of overlap, a variable can be present in more than one group.

Usage

create_groups(
  npred = 1000,
  nassoc = 10,
  nassoc_percent = 10,
  overlap = FALSE,
  overlap_size = 10
)

Arguments

npred	p, number of prediction variables.
nassoc	p_t, number of variables with true effect size = 0.
nassoc_percent	an integer from 0 to 100, indicating percentage of variables in a cluster with true effect size > 0.
overlap	boolean, default: FALSE. It determines if the clusters should be overlapping.
overlap_size	an integer, default: 10. It determines the percentage of variables to overlap across clusters.

Value

A list of

• csd: Cluster size data, as a data.table containing cluster, cluster size, group weights as in the sqrt(cluster size), and number of variables with true effect size.

• groups: A named list of integers of length length equal to the number npred containing the cluster and named with the variable name.

bips-hb/rgp documentation built on Feb. 3, 2021, 11:31 a.m.