CODEWITHSUNDEEP
thresholdpre-trained-modelminmax
Alberto Olivieri
Alberto Olivieri

Reputation: 75

Threshold optimization in unlabeled data

I am using a pretrained model that gives me an zero-shot output of three probabilities for three classes(day, night, and aerial photos), I need to find the threshold that maximizes the value of one class (day) and minimizes the other two, to label the data, but I am unsure how.

A colleague suggested me to use K fold cross validation, but I'm still unsure how to implement it, as far as I know the Kfold is used to test the accuracy of a model, and I do not have labelled data, and the model is a pretrained one.

I've looked around, but the solutions I found do not consider the use of unlabeled data, as they presume you're training the model, while I have a pretrained one that I am using as a tool to label new unknown data, but I still have to find that pesky threshold.

Any advice?

Upvotes: 1

Views: 60

Answers (0)

Related Questions