-
Notifications
You must be signed in to change notification settings - Fork 1
/
miscellaneous.txt
30 lines (26 loc) · 1.12 KB
/
miscellaneous.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Requirements
============
numpy
torch
torch_geometric
pandas
scikit-learn
tabulate
matplotlib
Flags
=====
--benchmark : benchmark dataset (default: cora)
choices are cora, citeseer, cora-f , pubmed, coauthor-p, coauthor-c, amazon-p, and amazon-c
--framework : model choices (default: NSGNN)
choices are NSGNN, GCN, GAT, GRAPHSAGE, and MLP
--layers : number of layers needed to construct your model (default:1)
--neurons : number of neurons to use for hidden layers of your model. ***not needed for NSGNN
--lr : learning rate (default:0.01)
--num_splits : number of different data-splits to use for training and testing mode (default:10)
--weight_decay : weight decay to use for Adam optmizer (default:0.0005)
--heads : number of attention-heads to use for GAT ***only needed for GAT (default:8)
--depth : propagation depth of NSGNN filter ***only needed for NODE-SELECT (default:1)
--random : whether to randomize the seeds used for training/testing the model (default:False)
Best Hyperparameter Configurations
==================================
see NODE-SELECT_Configurations.pdf