eval_classify: Extract True/False Positive/Negative Measures After...

In jmleach-bst/ssnet: Fit Spike-and-Slab Elastic Net GLM With Spatial Priors on Parameters

Extract True/False Positive/Negative Measures After Classification

Description

After fitting GLM's with, set a classification rule, and obtain measures for classification effectiveness.

Usage

eval_classify(
  num.class = 2,
  classify.rule,
  beta.hat,
  x,
  y,
  family,
  make.x0 = TRUE,
  gamma.link = "log",
  extra.measures = FALSE
)

Arguments

num.class	A scalar value defining the number of classes. Default value is two.
classify.rule	A vector of thresholds that determine assigned classes.
beta.hat	A vector of parameter estimates from some model fit.
x	A (numeric) design matrix whose number of columns is equal to length(beta.hat) - 1 and whose number of rows is equal to length(y). Strictly speaking, x may be a data frame, but categorical variables must have already been transformed into alternative coding schemes; e.g., dummy coding.
y	A numeric vector of outcome measurements. The length of this vector must equal nrow(x).
family	One of "gaussian", "gamma", "poisson", "binomial", which determines the distribution from which y is assumed to be drawn.
make.x0	Logical. When TRUE, creates a column of ones for the intercept. Default is TRUE.
gamma.link	Defines the link function in gamma GLM's. One of c("inverse", "log", "identity"). The default is "log".
extra.measures	Logical. When TRUE, also returns mean squared error (mse) and mean absolute error (mae), and when family = "binomial", returns AUC and misclassification. Default is FALSE.

Details

Note that y and x may be the same outcomes/design matrix used to generate beta.hat, but they may also, and perhaps more appropriately, be a held-out/independent data set on which to test the classification performance of whatever model was used to obtain beta.hat.

Value

A data frame with a single row containing columns for sensitivity, specificity, ppv, npv, and accuracy, where ppv is positive predictive value and npv is negative predictive value. When there are 3 or more classes, only accuracy, and possibly mse and mae, are returned.

Examples

set.seed(72874)

# sample size
n <- 10

# true betas
b <- c(1, 0.1, 0.5, 0, -0.1, -0.5)

# observed predictors
x <- matrix(rnorm(n * (length(b) - 1)), nrow = n, ncol = length(b) - 1)
x0 <- rep(1, n)

# observed outcomes
y <- cbind(x0, x) %*% b + rnorm(n, 0, 2.5)
y <- ifelse(y > 0, 1, 0)

# fitted model
yx.df <- data.frame(y, x)
beta.hat <- glm(y ~ ., data = yx.df , family = "binomial")$coefficients

# evaluate classification
eval_classify(classify.rule = 0.5, beta.hat = beta.hat,
              x = x, y = y, family = "binomial",
              extra.measures = TRUE)

jmleach-bst/ssnet documentation built on March 4, 2024, 5:04 p.m.