Awesome
HUAWEI-DIGIX-AgeGroup
2019 HUAWEI DIGIX Nurbs Solutions
Notes:
RAdam,AdamW,Lookahead,CycLearn在/snake/11. lstm-atten中
CTR模型的选择,只需要更换调用的模型,如FibiNet,FGCNN,xDeepFM,调用的是 deepctr 库
Meta Train在/snake/7和8中,参考的是Home Credit 17名和Elo Senkin大佬的Solutions
原生的adam与Lookahead配合较好,效果比较为adam<AdamW<RAdam<RAdam + Lookahead<AdamW + Lookahead<adam + lookahead,效率是纯AdamW最快
Graph特征构建在/snake/3. Graph Feature中,有一些特征较慢,可以注释掉,我们选择的建图为二分图建图方法
Sites: chizhu : https://github.com/chizhu nurbs: https://github.com/suncostanx deepctr : https://github.com/shenweichen/DeepCTR/