Dynamic-selftraining-GCN

Implementation of the paper

Zhou, Z., Zhang, S., and Huang, Z., "Dynamic Self-training Framework for Graph Convolutional Networks", arXiv e-prints, 2019.(https://arxiv.org/abs/1910.02684?context=stat.ML)

This code is based on the GCN code in kipf's version , implemented mainly on tensorflow.

How to run the experiments?

"cd" where "exps.py" is, and
"python exps.py", 
Wait for the results!

Or run it on server and nohup it:
"nohup python3 -u exps.py  > log.txt 2>&1 &"

Results can be seen in "results.txt".

How to modify the experiments you want?

See the "exps.py", 
"results=main(inputParameters=parameters)" is a standard form to convey parameters into 
experiment, "parameters" is a dictionary which restore value of parameters.

Default parameters can be seen in function "constructDefaultParameterDict", if a parameter 
is not in the keys of "parameters", we use its default value.

For example:
"for parameters in [{'dataset':'cora','train_size':20},
                    {'dataset':'pubmed','standard_split':True}]:

    for i in range(100):

        results=main(inputParameters=parameters)"