BAAI 100000 yuan 670 Team764 participants
High-Energy Particle Classification Challenge
2019-12-09 - Launch
2020-03-20 - Team Merger Deadline
2020-03-22 - Close
ICON Home     Competitions    

MarkSHEN

评测方案中的ROC AUC是怎么做多标签平均的?

posted in   High-Energy Particle Classification Challenge

2019-12-12 06:54

23  comments

CH

2019-12-13 04:44

Reply

2
<p>同样疑惑,auc应该要提交概率值才能算,而样本的提交的类别</p>
  • 天王盖地虎 reply CH

    2019-12-14 13:27

    Reply

    0
    <p>我也是这个问题 如果用各个类的概率来算auc线下正常,把类别交上去分就很低,搞不懂。。。</p>

LogicJake

2019-12-14 16:09

Reply

0
<p>y_one_hot = label_binarize(df_train[ycol], np.arange(4))<br> <p>score = roc_auc_score(y_one_hot, oof_one_hot)<br>

LogicJake

2019-12-14 16:10

Reply

0
<pre><code class="lang-python">y_one_hot = label_binarize(df_train[ycol], np.arange(4)) oof_one_hot = label_binarize(oof.argmax(axis=1), np.arange(4)) score = roc_auc_score(y_one_hot, oof_one_hot) print('auc', score) </code></pre>

daydayup1111

2019-12-21 06:45

Reply

0
<p>from sklearn.preprocessing import label_binarize<br>import numpy as np<br> <p>y_pred = [1,4,5,21]<em>25<br>25<br>y_one_hot = label_binarize(y_true, classes=[1,4,5,21])<br>y_pred_one_hot = label_binarize(y_pred, classes=[1,4,5,21])<br>print(y_one_hot)<br>score = roc_auc_score(y_one_hot, y_pred_one_hot,average=’macro’)<br>

High-Energy Particle Classification Challenge

100000 yuan

764 participants

start

Final Submissions

2019-12-09

2020-03-22