penAFT.coef {penAFT}

Extract regression coefficients from fitted model object

Description

A function to extract coefficients along the solution path for the regularized semiparametric acceleratred failure time model estimator.

Usage

penAFT.coef(fit, lambda = NULL)

Arguments

fit

A fitted model from penAFT or penAFT.cv.

lambda

The tuning parameter value at which to extract coefficients. If NULL and fit is a penAFT.cv object, will use the tuning parameter value with minimum cross-validation linear predictor score.

Details

The regression coefficients stored in the fitted model objects coming from penAFT or penAFT.cv will (i) be on the scale of standardized predictors if standardization was used (which is the default) and (ii) are stored as a specific sparse matrix so that coefficient extraction is cumbersome. This function returns the regression coefficient estimates on the original scale of the predictors for a particular tuning parmaeter value. It is important to note that this method does not return an estimate of the intercept: the intercept is absored into the error term as the Gehan loss function is invariant to location change.

Value

beta

The coefficient estimates

Examples

# --------------------------------------
# Generate data  
# --------------------------------------
set.seed(1)
genData <- genSurvData(n = 100, p = 50, s = 10, mag = 1, cens.quant = 0.6)
X <- genData$X
logY <- genData$logY
delta <- genData$status


# --------------------------------------
# Fit elastic net penalized estimator without CV
# --------------------------------------
fit <- penAFT(X = X, logY = logY, delta = delta,
                   nlambda = 50,
                   penalty = "EN",
                   alpha = 1)

coef.10 <- penAFT.coef(fit, lambda = fit$lambda[10])
coef.20 <- penAFT.coef(fit, lambda = fit$lambda[20])

# Cannot obtain fit at lambda not in fit$lambda
## Not run: coef.error <- penAFT.coef(fit, lambda = 10) # throws error


  # ------------------------------------------
  # Fit elastic net penalized estimator with CV
  # -------------------------------------------
  fit.cv <- penAFT.cv(X = X, logY = logY, delta = delta,
                   nlambda = 50,
                   penalty = "EN", 
                   alpha = 1, nfolds = 5)

  ## --- coefficients at lambda minimizing cross-validation error
  coef.cv <- penAFT.coef(fit.cv) 

  ## ---- coefficients at 10th considered lambda 
  coef.cv10 <- penAFT.coef(fit.cv, lambda = fit.cv$full.fit$lambda[10]) 

  # -------------------------------------------
  # Repeat with sparse group lasso without CV
  # -------------------------------------------
  groups <- rep(1:10, each = 5)
  fit.sg <- penAFT(X = X, logY = logY, delta = delta,
                   nlambda = 50, groups = groups,
                   penalty = "SG",
                   alpha = 0.5)

  coef.sg.10 <- penAFT.coef(fit.sg, lambda = fit.sg$lambda[10])
  coef.sg.20 <- penAFT.coef(fit.sg, lambda = fit.sg$lambda[20])


  # -------------------------------------------
  # Finally, fit sparse group lasso with CV
  # -------------------------------------------
  groups <- rep(1:10, each = 5)
  fit.sg.cv <- penAFT.cv(X = X, logY = logY, delta = delta,
                   nlambda = 50, groups = groups,
                   penalty = "SG",
                   alpha = 0.5, nfolds = 5)

  coef.sg.cv <- penAFT.coef(fit.sg.cv)
  coef.sg.cv10 <- penAFT.coef(fit.sg.cv, lambda = fit.sg$full.fit$lambda[20])