# How can I calculate FPR?

How can I compute the FPR for sentences with no labels?

Is there any relation between the FPR and the likelihood?

By definition, you must have labels to compute the false positive rate (FPR). If you don't have labels, you can't possibly compute it, because you don't know if the positive or negative predictions of your model are true or false.

$$FPR=\frac{\mathrm{FP}}{\mathrm{FP} + \mathrm{TN}}$$

(FP = false positives, TN = true negatives)

The likelihood represents the probability of a model, given some parameters, to match the data. When training a model, you usually find the parameters that maximize the likelihood of the data or, equivalently, minimize the negative log-likelihood; this method is called maximum likelihood estimation. To compute the likelihood, you also need labels. Its mathematical definition depends on the specific model.

Intuitively, the higher the likelihood, the lower the false positive rate.

• Well, my suggestion is to explain to him that their request cannot be applied to that specific situation and explain the reasons why.
– noe
Feb 26, 2023 at 19:31