DeepFM binding model FM is the product DNN model structure and principle see: https://mp.weixin.qq.com/s/Hb6tKk1sw9pZ7qysO765nw , code logic: Hive access ➡️ generated onehot coding dictionary ➡️ convert data input to train ➡️ model
See model code GitHub: https://github.com/zwtgyh/algorithm/tree/master/deeplearning/DeepFM
Request line using a Java call interface, code reference: https://blog.csdn.net/luoyexuge/article/details/79941565?utm_source=blogxgwz8 , special attention is tensorflow model predictive interface to access data format issues, the corresponding good python and Java data formats , there would be a great deviation estimated results
After deployment uses docker listening port (specify your own), to achieve the online prediction model