Regarding a weighted Graph

[afpmorgado]

Hey Kipf, firstly I want to express my gratitude for this project, it still proves really helpful in the big year of 2025.

I intend to use your GAE to enrich Glove embeddings on semantic and emotional categories associated to words, for a downstream regression task. At the moment I'm running into computational issues using a 5000x5000 weighted graph (floats) as the adjacency matrix, which doesn't happen with binary.
Do you reckon inputting a binary adjacency matrix and then multiplying the loss contribution of each edge by the respective edge weights would be able to separate my classes according to the given importance?
Or should I avoid this and try to fit in a weighted adjacency at all costs?

[zrq0123]

您的邮件已收到