![240](https://upload.jianshu.io/users/upload_avatars/1168461/f01b5214-acb5-4658-bc8c-9742cbab3ba6.jpg?imageMogr2/auto-orient/strip|imageView2/1/w/240/h/240)
IP屬地:北京
數(shù)據(jù)處理 preprocess.py tr_seqs[[1, 2], [1], [4], [6], [8, 9], [8], [10, 11,...
推薦系統(tǒng)知識(shí)追蹤每個(gè)session構(gòu)造一個(gè)有向圖炒辉,邊的權(quán)重歸一化,先把所有session的圖依次輸入GGNN纤泵,然后就能得到所有節(jié)點(diǎn)的embedd...
實(shí)現(xiàn)Seq2seq+Attention有以下幾個(gè)步驟: 定義RNNCell(封裝了attention機(jī)制) 定義traininghelper 定...
ABSTRACT 當(dāng)重現(xiàn)DKT模型時(shí),發(fā)現(xiàn)兩個(gè)問題: the model fails to reconstruct the observed i...