采用precision_recall_evaluator()进行性能评估,输出的eval是否有错
Created by: dreamfantacy
在gpu机器上运行sequence模型,采用的precision_recall_evaluator,其中输出的eval如下所示:
Test samples=346666 cost=0.121776 Eval: positive_label=-1 precision=1.2732e-313 recall=6.93541e-310 F1-score=1.19593e-316 macro-average-precision=0.900901 macro-average-recall=0.787111.....
precision和recall的值这么小,快逼近0了,但 macro-average-precision的值感觉也是合理的,这样的模型训练结果是否合理?