tprf.py

def gen_tprf(predictions, positive_count=None, shuffle=True, sort=True):
    predictions = predictions if not shuffle else \
        fx.shuffle(predictions)
    predictions = predictions if not sort else \
        sorted(predictions, key=lambda p: -p.p)

    N = positive_count if positive_count is not None else \
        sum(1 for x in predictions if x.y == 1)

    tp = 0
    fp = 0
    fn = 0

    for p in predictions:
        if p.y == 1:
            tp += 1
        elif p.y == -1:
            fp += 1
        # else:
        #     assert False

        fn = N - tp

        T = p.p
        P = 0.0
        if tp + fp > 0:
            P = tp / (tp + fp)
        R = 0.0
        if tp + fn > 0:
            R = tp / (tp + fn)
        F = 0.0
        if P + R > 0:
            F = 2 * P * R / (P + R)

        yield TPRF(
            threshold=T,
            precision=P,
            recall=R,
            f1=F)