xiaomingtx

Hi LeoGrin, thanks for very meaningful work. I see in the paper you use deep models like MLP, ResNet etc, are you using some deep model dedicated to feature interaction...

hi @YannDubs, maybe np.random.seed needs to be defined before HashFamily? Now the a, b, and module are random, we can't get the same result if we reload the model.

Performance issue

2

Hi @porterehunley, I ran the example, 10000 node take 16s, 60000 node take 6min, not as fast as the report said May I ask where the problem may have occurred?...

Hi, 你好，看论文有几个问题想请教下： 1、在GIM模块，original feature的特征向量维度应该是小于interactive feature的，是怎么放在一起做attention的 2、同样是GIM模块，这些特征向量应该对应不同的空间吧，加权求和为什么会有提升 没想明白，麻烦作者大大解答下