shenweichen/DeepCTR-Torch

AutoInt implementation is tottaly different from the paper.

Closed this issue · 2 comments

diff7 commented
  1. No embeddings for continuous values
  2. No skip connections between attention blocks

1.This issue was explained there #62
2. There was no skip connections between attention blocks in the first version of the paper. It was revised and added by the authors after a long time since the first version.

diff7 commented

got it, thank you!