def precision(self, labels, pred_scores):
"""
Compute the precision
Parameters
----------
labels: value list. The labels of data set.
pred_scores: pred_scores: value list. The predict results of model. It should be corresponding to labels each data.
thresholds: value list. This parameter effective only for 'binary'. The predict scores will be 1 if it larger than thresholds, if not,
if will be 0. If not only one threshold in it, it will return several results according to the thresholds. default None
result_filter: value list. If result_filter is not None, it will filter the label results not in result_filter.
Returns
----------
dict
The key is threshold and the value is another dic, which key is label in parameter labels, and value is the label's precision.
"""
if self.eval_type == consts.BINARY:
precision_operator = classification_metric.BiClassPrecision()
metric_scores, score_threshold, cuts = precision_operator.compute(
labels, pred_scores)
return metric_scores, cuts, score_threshold
elif self.eval_type == consts.MULTY:
precision_operator = classification_metric.MultiClassPrecision()
return precision_operator.compute(labels, pred_scores)
else:
logging.warning("error:can not find classification type:{}".format(
self.eval_type))
def quantile_pr(self, labels, pred_scores):
if self.eval_type == consts.BINARY:
p = classification_metric.BiClassPrecision(cut_method='quantile', remove_duplicate=False)
r = classification_metric.BiClassRecall(cut_method='quantile', remove_duplicate=False)
p_scores, score_threshold, cuts = p.compute(labels, pred_scores)
r_scores, score_threshold, cuts = r.compute(labels, pred_scores)
p_scores = list(map(list, np.flip(p_scores, axis=0)))
r_scores = list(map(list, np.flip(r_scores, axis=0)))
score_threshold = list(np.flip(score_threshold))
return p_scores, r_scores, score_threshold
else:
logging.warning('error: pr quantile is for binary classification only')
from federatedml.evaluation.metrics.classification_metric \
import BiClassAccuracy, BiClassRecall, BiClassPrecision, KS, Lift, Gain, FScore
from federatedml.evaluation.metric_interface import MetricInterface
from federatedml.evaluation.metrics import classification_metric
import numpy as np
from federatedml.evaluation.backup.evaluation import BiClassPrecision as BiClassPrecision2
scores = np.random.random(100)
labels = (scores > 0.5) + 0
interface = MetricInterface(1, 'binary')
mat = interface.confusion_mat(labels, scores)
rs = classification_metric.ThresholdCutter.cut_by_quantile(scores, )
rs2 = classification_metric.BiClassRecall(cut_method='quantile').compute(
labels, scores)
rs3 = classification_metric.BiClassPrecision(cut_method='quantile').compute(
labels, scores)
rs4 = BiClassPrecision2().compute(labels, scores, rs2[1])
comp1 = [i[1] for i in rs3[0]]
comp2 = [i[1] for i in rs4[0]]
print(comp1 == comp2)
