search

Hanjun-Dai / graph_comb_opt

Implementation of "Learning Combinatorial Optimization Algorithms over Graphs"
https://arxiv.org/abs/1704.01665
MIT License
491 stars 135 forks source link

Understanding s2v_mc #21

Open pandalgo opened 4 years ago

pandalgo commented 4 years ago

Hi, Thank you for sharing your code.

I am trying to understand how to write the equation for Q in s2v_mvc, given the additional auxiliary input. Following the notation in the paper, is it like this :

q_function

Where a^{T} is the auxiliary input and θ{8}_ is 3-dimensional. Am I right? If I am not, can you explain how to write the equation for s2v_mvc?

Thank you!

Hanjun-Dai commented 4 years ago

Yes a^T is the auxiliary input which can be node degree, coverage information, etc. There are some differences:

1) I guess in my implementation I simply concat a^T with the relu part (i.e., remove \theta_8) 2) the summation is over entire graph, instead of the neighborhood.