accuracy: Accuracy Measures Fort Predicted Responses

In kouroshz/creNet: Fit a GLM with a combination of lasso and overlapping group lasso regularization

Description

This function computes accuracy measures for predicted numerical or binary responses.

Usage

assessModel(pred, obs, type=c("linear","logit"), verbose=TRUE, prednames=NULL)

Arguments

pred	Vector or matrix of predicted responses. Each column corresponds to a different lambda value
obs	Vector or observed binary responses
type	Type of prediction model
verbose	logical, defaults to TRUE
prednames	Optional vector of names for the different sets of predictions

Details

For linear regression, the function simply returns the root mean square error (RMSE) of the predicted responses. For logistic regression, the function computes the negative log-likelihood using the original vector pred whose values should be in $(0,1)$ (soft classification). For the purpose of hard classification, the function also converts this vector to binary values and computes the following classification measures: sensitivity a.k.a. true positive rate (TP/(TP+FN)), specificity a.k.a. true negative rate (TN/(FP+TN)), accuracy (TP+TN)/(TP+TN+FP+FN), and precision a.k.a. positive predictive value (TP/(TP+FP)). Two cutoff values are considered for the binary conversion: 0.5 and the proportion of 1's in obs.

Value

If type="linear", the RMSE of the predictions.
If type="logit", a list with components

prior	proportion of 1's in obs
ll	negative log-likelihoods
classif	classification measures for the cutoff value 0.5
classif.adj	classification measures for the adjusted cutoff value prior

Author(s)

David Degras

See Also

creSGL, predictSGL, runPredictions

Examples

n = 50; p = 100; size.groups = 10
index <- ceiling(1:p / size.groups)
X = matrix(rnorm(n * p), ncol = p, nrow = n)
beta = (-2:2)
y = X[,1:5] %*% beta + 0.1*rnorm(n)
y = ifelse((exp(y) / (1 + exp(y))) > 0.5, 1, 0)
data = list(x = X, y = y)
weights = rep(1, size.groups)
fit = creSGL(data, index, weights, type = "linear", maxit = 1000, thresh = 0.001, 
	min.frac = 0.05, nlam = 100, gamma = 0.8, standardize = TRUE, verbose = FALSE, 
	step = 1, reset = 10, alpha = 0.05, lambdas = NULL)

pred = predict(Fit, X, 5)
prednames = substr(fit$lambdas,1,4)
accuracy(pred, y, type="logit", prednames=prednames)

kouroshz/creNet documentation built on May 20, 2019, 1:11 p.m.