知识卡片
指标键名不决定真实计算方法
将roc_auc_score传给名为acc的回调,输出acc仍可能是AUC而非准确率。需要检查回调收到连续分数还是硬预测,以及macro、micro等聚合设置;不显式指定时不能凭注释认定一定micro平均。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- · 查看原书依据
>>> result, model_outputs, wrong_predictions = model.eval_model(eval_df, ... acc=roc_auc_score) #1 >>> result {'LRAP': 0.9955934600588362, 'acc': 0.9812396881786198, 'eval_loss': 0.04415484298031397} #1 The Jigsaw Toxic Comment Classification Challenge used ROC AUC as the accuracy metric. The receiver operating characteristic (ROC) curve area under the curve (AUC) metric balances all the different ways a classifier can be wrong by computing the integral (area) under the precision versus recall plot (curve)
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第09章《可堆叠的Transformer》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-9.html)
- 结论依据:摘录中关于指标键名不决定真实计算方法的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:>>> result, model_outputs, wrong_predictions = model.eval_model(eval_df, ... acc=roc_auc_score) #1 >>> result {'LRAP': 0.9955934600588362, 'acc': 0.9812396881786198, 'eval_loss': 0.04415484298031397} #1 The Jigsaw Toxic Comment Classification Challenge used ROC AUC as the accuracy metric. The receiver operating characteristic (ROC) curve area under the curve (AUC) metric balances all the different ways a classifier can be wrong by computing the integral (area) under the precision versus recall plot (curve)