November 2019
Intermediate to advanced
346 pages
9h 36m
English
We begin this recipe by loading in a previously featurized dataset and specifying a desired FPR constraint of 1% (step 1). The value to be used in practice depends highly on the situation and type of file being considered. There are a few considerations to follow: if the file is extremely common, but rarely malicious, such as a PDF, the desired FPR will have to be set very low, for example, 0.01%.
If the system is supported by additional systems that will double-check its verdict without human effort, then a high FPR might not be detrimental. Finally, a customer may have a preference, which will suggest a recommended value. We define a pair of convenience functions for FPR and TPR in step 2—these functions are very handy and ...