duchao0726 / dual Goto Github PK

大佬你好，看了你的code的实现，有一些疑问

• 为什么p分布的协方差矩阵初始值要用cholesky分解K(Z, Z)的方式初始化

I have ran your batchrunner_dnn.py script and the results I have for DUAL (labelled dnn) appear to be consistent with those found in your paper. I have adapted your codes to run the DNN model without DUAL (labelled dnn_no_DUAL), however, the results for this are not consistent with those found in your paper. Box plots showing the range of various metrics for both models can be seen here:

It would be great if you could give me some insight on how you generated the results for "vanilla DNN". Were the hyper-parameters the same as those for the DNN-Dual model found in batchrunner_dnn.py?

Thank you

I have ran the Wide and Deep model on the Amazon Books dataset. The model only generates an AUC of around 0.57.
Has anyone else experienced a similar problem?

作者您好，请问能否提供 在Yahoo R6B 数据集上replay的实验代码