CleanLabel清洗噪声数据

1、简介

cleanLabel：筛选带噪声数据；筛选难例样本

吴恩达大佬说人工智能是20%的算法+80%的数据决定。

2、安装

pip install cleanlab
#or
conda install -c cleanlab cleanlab

3、sample

from cleanlab.dataset import overall_label_health_score
health = overall_label_health_score(labels, pred_probs)

#查找数据集中哪些数据存在问题
from cleanlab.classification import CleanLearning
issues = CleanLearning(yourFavoriteModel).find_label_issues(data, labels)