Computes marginal conformal p-values using conformal prediction in binary classification tasks. Conformal prediction is a framework that augments machine learning algorithms with a measure of uncertainty, in the form of prediction regions that attain a user-specified level of confidence. This package specifically focuses on providing conformal p-values that can be used to assess the confidence of the classification predictions. For more details, see Tyagi and Guo (2023) < https://proceedings.mlr.press/v204/tyagi23a.html>.
To install the package, run:
devtools::install_github("ct364/conformalpvalue")
To use the function "naive_bayes_with_conformal_pvalues", use
# Train data with target variable "y"
# Calibration data with target variable "y"
# Test data without target variable "y"
result <- conformal_pvalues(train_data, calib_data, test_data, "y",method="naiveBayes")
This returns a matrix of conformal p-values.