Function to assess precision/recall for predicted results
leandroradusky opened this issue · comments
Having as input Protein.go_terms()
, predictions and a probability_threshold
(what we consider positive and negative in our predictions) we should be able to assess our:
precision
orPPV
(positive predictive value)recall
orTPR
(true positive rate)
For further descriptions of what these terms mean the Wikipedia article for ROC curves is detailed enough. Indeed, all the methods and the competition papers show these kinds of curves so we should be able to plot this for our models.
This functionality should be available in manas-cafa5.metrics
.