1
0
Fork 0
QuaPy/quapy/error.py

48 lines
951 B
Python

from sklearn.metrics import f1_score
from settings import SAMPLE_SIZE
def f1e(y_true, y_pred):
return 1. - f1_score(y_true, y_pred, average='macro')
def acce(y_true, y_pred):
acc = (y_true == y_pred).mean()
return 1. - acc
def mae(prevs, prevs_hat):
return ae(prevs, prevs_hat).mean()
def ae(p, p_hat):
assert p.shape == p_hat.shape, 'wrong shape'
return abs(p_hat-p).mean(axis=-1)
def mrae(p, p_hat, eps=1./(2. * SAMPLE_SIZE)):
return rae(p, p_hat, eps).mean()
def rae(p, p_hat, eps=1./(2. * SAMPLE_SIZE)):
p = smooth(p, eps)
p_hat = smooth(p_hat, eps)
return (abs(p-p_hat)/p).mean(axis=-1)
def smooth(p, eps):
n_classes = p.shape[-1]
return (p+eps)/(eps*n_classes + 1)
CLASSIFICATION_ERROR = {f1e, acce}
QUANTIFICATION_ERROR = {mae, mrae}
f1_error = f1e
acc_error = acce
mean_absolute_error = mae
absolute_error = ae
mean_relative_absolute_error = mrae
relative_absolute_error = rae