Dense value is not for DNN input in AutoInt

shenweichen / DeepCTR-Torch

【PyTorch】Easy-to-use,Modular and Extendible package of deep-learning based CTR models.

https://deepctr-torch.readthedocs.io/en/latest/index.html

Apache License 2.0

2.92k stars 687 forks source link

Dense value is not for DNN input in AutoInt #62

Closed SongRb closed 3 years ago

SongRb commented 4 years ago

Describe the bug(问题描述) In original paper of AutoInt, dense value is also converted to a fix length embedding. Using a DNN for dense value list may be wrong.

https://github.com/shenweichen/DeepCTR-Torch/blob/687a094135fa597697d926782a5634c79b627dac/deepctr_torch/models/autoint.py#L98

shenweichen commented 4 years ago

hi, in the previous version of deepctr(not deepctr-torch) we converted the dense value to embedding vectors.Now in order to be compatible with apis of other modes, we changed the implementation of dense values in autoint.

diff7 commented 3 years ago

You should rename it because it is very confusing when looking at the paper and your code. Or at least add comments in the code about it.

abedshantti commented 2 years ago

@shenweichen do you know where I can find the transformation of dense values into embedding vectors? I have been trying to locate it in the AutoInt code file but it is not there. Or am I supposed to define all the features as sparse before defining the model (as in the criteo classification example script)?