Dear All DGL community members.
I really hope to not asking this lazy question. I’m really sorry in advance, i really new in programming.
I just have a vague goal of determining some attention value of evolving graph, which is based on the assumption that we can’t know the true node features without taking into account data of neighboring nodes features ones progressively, i hope this is in line with the basic concepts of GAT. Another condition i suspect in my case is that it will interact with nodes with features that will only be filled in the future (the outer border).
I have run the GAT code in tutorial (https://docs.dgl.ai/en/0.4.x/tutorials/models/1_gnn/9_gat.html), it goes smoothly with error around 1.9 and taking around 0.47second/epoch. next step i intended to working out is on how it works, i have tried to learn all crucial concept and expression (pytorch expression such as nn.linear, nn.init_xavier, except message passinng paradigm is still on proccess).
To do that i try to read the cora-dataset content and its structure, i have tried to make data that is stored in cora_v2.zip readable since a week ago, i try with several method to convert it into csv with panda but none works.
Meanwhile i have prepare some of my data and its features in edgelist.csv and nodelist.csv, and it plot nicely with panda and numpy (i haven’t tried with pytorch). my graph in its fullest phase will contain more than 1000 nodes with around 3-8 edges per nodes, it is fixed in space (got its own coordinates) with 20 parameters and 500-1000 entry each, i don’t know how to incorporate this spatial features thing, i have read the work on GCN and GAT approach in traffic forecasting but i don’t know whether it relevant in my case.
So in short, my questions are as follows:
- Is my assumption and condition compatible/correct with GAT?how to work on interaction and features at “post-existing nodes”?the spatial fixity is also a problem.
- How to read the content of cora-dataset data?how it works with the code and else, because i think it is very crucial for practical learner like me.
Enormous thanks in advance.
Best Regards