condense_k_cv_output: Condense k cross validation output

In mdkessler/kTSCR: Implements k-Top Scoring Cluster Regression

Description

Condense and summarize output from k cross validation iterations. Specifically, this returns the union of elders across k-cv iterations, and the total per feature of how many sibling pairs it was in across all k-cv iterations

Usage

condense_k_cv_output(
  X,
  k_cv_res,
  k,
  app_minus_test_thresh = 0.1,
  weight_sum_by_app_minus_thresh = TRUE
)

Arguments

X	input feature matrix
k_cv_res	k cross validation result object from function run_cross_validation()
k	k number of cross validation iterations, passed from run_cross_validation function
app_minus_test_thresh	a threshold level for app_cor - test_cor...k-cv iterations that have app_cor - test_cor <= app_minus_test_thresh (i.e. low overfitting) have their siblings added to the consensus siblings ultimately used for prediction. A hyperparameter. Defaults to 0.10
weight_sum_by_app_minus_thresh	A logical. Indicates whether the sums across k-cv iterations used for feature importance should be weighted by app_cor - test_cor (i.e. higher weight to k-cv iterations that overfit less). Default is true.

Value

list of four vectors: app_cor, test_cor, elders, feature_importance

mdkessler/kTSCR documentation built on Feb. 25, 2021, 10:31 p.m.