ap.py

import numpy as np
from sklearn.metrics import precision_recall_curve, average_precision_score


def naive_interpolated_precision(y_true, y_scores):
    precisions, recalls, _ = precision_recall_curve(y_true, y_scores)
    interp_precisions = []

    # the final point
    precisions = precisions[:-1]
    recalls = recalls[:-1]

    for recall_point in [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]:
        interp_precisions.append(precisions[recalls >= recall_point].max())

    return np.mean(interp_precisions)


rng = np.random.RandomState(4)

for _ in range(100):
    y_scores = rng.randn(100)
    y_true = rng.randn(100) > 0.4

    naive = naive_interpolated_precision(y_true, y_scores)
    fast = average_precision_score(y_true, y_scores, interpolation="eleven_point")

    print(np.abs(naive - fast))

stdout

0.0
0.000297088532383
0.0
0.0
0.0
0.0
0.0
0.000829967354617
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.000884127314333
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.00134815280426
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.00149653786017
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0