trainingMisCov: trainingMisCov

In LTTGH/drkm4mc: Kernel machines with missing covariates

Description

This is train algorithm, we consider three different cases 1. CC; 2.WCC; 3.DR The algorithm uses the hinge loss function to do minimization procedure If CC case, return the estimated alpha based on complete data estimated parameters If WCC case, return the estimated alpha based on complete data and for propensity score which is also used in testing procedure; If DR case, return (1) the contrained quadratic program solution alpha (2) estimated parameters for propensity score (3) estimated parameters of conditional distribution of x2 given X1 and Y. (4) Covariates of the dataset after imputation- the large dataset

Usage

trainingMisCov(
  misCovDat,
  px,
  kerType,
  kerMethod,
  lambda,
  sigma,
  PSFunPath,
  IMPFunPath,
  B,
  testPurpose
)

Arguments

misCovDat	data set with missing covariates. (X,V,R,Y1,Y), X is the fully observed covariates; V is the popential missing covariates; R is the missing indicator; Y1=Y the binary response
px	dimsension of covariates X (totally observed)
kerType	type of kerenl function, "RBF"(if choose) or "linear" (else choose, automatically)
kerMethod	method to use "CC", "WCC", and "DR"
lambda	tuning parameter
sigma	parameter for RBF kernel, also act as tuning parameter
PSFunPath	Path of function used to estimated propensity score
IMPFunPath	Path of imputation function used to generte imputation data
B	The imputation time for DR (B>=1)
testPurpose	test purpose, for crossivalidation, the loss function is chosen as the phi loss; for test (default value) , the loss function is chosen as the classification loss.

Value

trainRes

LTTGH/drkm4mc documentation built on July 3, 2021, 4:15 p.m.