Skip to content

Latest commit

 

History

History
59 lines (33 loc) · 1.34 KB

README.md

File metadata and controls

59 lines (33 loc) · 1.34 KB

Usage

Build a configuration

Build a benchmark configuration in benchmark/<config_name>.json:

{
    "methods": ["vanilla", "ens"],
    "datasets": ["ogbn-arxiv"],
    "seeds": [100, 200]
}

So it will experiment in configuration of:

  1. vanilla ogbn-arxiv 100

  2. vanilla ogbn-arxiv 200

  3. ens ogbn-arxiv 100

  4. ens ogbn-arxiv 200` .

Methods are taken from ['vanilla', 'drgcn', 'smote', 'imgagn', 'ens', 'tam', 'lte4g', 'sann', 'sha', 'renode', 'pastel', 'hyperimba']

Datasets are taken from ['Cora_100', 'Cora_20', 'CiteSeer_100', 'CiteSeer_20', 'PubMed_100', 'PubMed_20', 'chameleon_100', 'chameleon_20', 'squirrel_100', 'squirrel_20', 'Actor_100', 'Actor_20', 'Wisconsin_100', 'Wisconsin_20', 'Amazon-Photo', 'Amazon-Computers', 'Coauthor-CS', 'ogbn-arxiv']

Seeds are taken from [100, 200, 300, 400, 500, 600, 700, 800, 900, 1000]

If set to an empty array, methods / datasets / seeds will be set to all of the options.

Run
python benchmark.py <config_name>

Options:

  • --gpu: use GPU, or use CPU if not set.
  • --debug: do not record the experiment for debug.

And the results will be recorded in records/<config_name>(_gpu).json.

Warning: Do not run the same config twice at the same time!

Print
python print.py

You can set what configurations to print in print.py.