search

isjakewong / MIRACLE

Multi-view Graph Contrastive Representation Learning for Drug-drug Interaction Prediction
https://arxiv.org/abs/2010.11711
41 stars 8 forks source link

The implementation is seemingly inconsistent with the paper. Is this a bug? #3

Closed ZillaRU closed 2 years ago

ZillaRU commented 2 years ago

Hello! I have one more confusing issue now... As you wrote in Formula (11) and (12) (Section 2.5.2) of your paper published on WWW'21, the disagreement loss should measure the intra- and inter-view's disagreement on UNLABELED drug pairs. image

But in your implementation, the disagreement quantified by KL divergence is calculated on the LABELED drug pairs using the adj_mask rather than adj_mask_un. The adj_mask_un is never used in loss computation. Is this a bug? https://github.com/isjakewong/MIRACLE/blob/61ab653d41d095227cb9d0778b0780fe3afe4724/MIRACLE/main.py#L265 https://github.com/isjakewong/MIRACLE/blob/61ab653d41d095227cb9d0778b0780fe3afe4724/MIRACLE/main.py#L266 https://github.com/isjakewong/MIRACLE/blob/61ab653d41d095227cb9d0778b0780fe3afe4724/MIRACLE/main.py#L312 Thanks in advance! XD

isjakewong commented 2 years ago

Thanks for pointing this typo out! I have updated the related files.

ZillaRU commented 2 years ago

My pleasure~ I will conduct experiments on my datasets using your updated implementation.