search

Murlocccc / hetero-secure-boost-origin

对着FATE V1.0版本的hetero secure boost模块写的
3 stars 0 forks source link

Predict accuracy #3

Open JiaHe664 opened 2 years ago

JiaHe664 commented 2 years ago

我发现，当选用的预测数据为整个数据集时，训练好的模型的预测效果很差，准确率只有45%左右，但是取整个数据集的1/5时预测效果却很好，有90%左右，这是为什么，是不是你在抽离源码时，忽略了一些部分？ I found that when the selected prediction data is the entire data set, the prediction effect of the trained model is very poor, with an accuracy rate of only about 45%, but when 1/5 of the entire data set is taken, the prediction effect is very good, with 90% Or so, why is this? Did you ignore some parts when you pulled away from the source code?

JiaHe664 commented 2 years ago

或者是在计算准确率时，存在部分硬编码吗？还是准确率计算有误？ Or is it partly hard-coded when calculating accuracy? Or is the accuracy calculated incorrectly?

Murlocccc commented 2 years ago

很奇怪的表现。应该没有硬编码和精度计算失误，请问这个现象是在所有数据集上都有吗？ Very strange behavior. There should be no hard coding and precision calculation errors. Does this phenomenon occur on all data sets?