PNC can be accessed from NIH, and ABCD can be accessed from NIMH Data Archive.
For those who can not access ABCD and PNC dataset, we also provide an open-source dataset, ABIDE. Please follow the instruction to download and process this dataset.
python main.py --config_filename setting/pnc_fbnetgen.yamlpython main.py --config_filename setting/abcd_fbnetgen.yamlpython main.py --config_filename setting/abide_fbnetgen.yamlAll hyper parameters can be tuned in setting files.
model:
# For the model type, there are 3 choices: "seq", "gnn" or "fbnetgen". type: fbnetgen# For the feature extractor, there are 2 choices: "gru" or "cnn".extractor_type: gru# For the feature extractor, there are 2 choices: "product" or "linear". # We suggest to use "product", since it is faster.graph_generation: product# Two hyperparameters are tuned in our paper.embedding_size: 8window_size: 8train:
# For training method, there are 2 choices: "normal" or "bilevel".# "bilevel" will be in effect only if the model.type is set as "fbnetgen"# We suggest to use "normal".method: normal# If the model.type is set as "gnn", this hyper parameter will be in effect.# There are 2 choices: "uniform" or "pearson".pure_gnn_graph: pearson