search

THUDM / GATNE

Source code and dataset for KDD 2019 paper "Representation Learning for Attributed Multiplex Heterogeneous Network"
MIT License
527 stars 141 forks source link

Question about `trans_w` #96

Closed Wang-Yu-Qing closed 3 years ago

Wang-Yu-Qing commented 3 years ago

In main.py line 70, the trans_w is the weights for transforming aggregated neighbor nodes' embedding. To my understanding, trans_w refers to W^{k} in formula (2). For different edge types, their trans_w are different since the source lookup weights trans_weights (line 70) is of size [edge_type_count, embedding_u_size, embedding_size // att_head]. However, from formula (2), the W^{k} is the same for all edge types. So for this weights, the code are not the same as the paper. Am I right? Or I misunderstanding somethine?

Wang-Yu-Qing commented 3 years ago

It seems my previous understanding is wrong, trans_w refers to M_{r} in equation (6).