Due to the commercial sensitivity of promotions, releasing our datasets is not applicable.

The implementation of MOEF is based on our company-customized distributed Tensorflow framework for better industrial applications. Releasing code related to the framework is not allowed and detailed information about input features are also confidential. So with these information removed, part of source code of MOEF is extracted and simplified as supplementary materials. The released source code is not runnable but able to illustrate the implementation of model structure.