All files contained in this repo:
cora.cites/cora.content ----- original dataset
cora_trimmed.cites      ----- graph without loops and overcitation 
cutter.py               ----- split test and train nodes
graph_trim.py           ----- trim loops and overcitation
mrf.py                  ----- mainly used for development and debug

To regenerate the experiment results reported:
1. GD method with gibbs
    ->py mrf_gibbs.py square/softmax 1(asymmetric prediction)/0(symmetric prediction)
2. pseudo_likelihood with gibbs
    ->py mrf_gibbs_pseudo_likelihood
3. manual label and pure NB
    DO NOT SUPPORT COMMENDLINE INPUT!!!
    uncomment lines accordingly in the mrf_gibbs.py 
NOTE:
The result may vary each run, but expect to see the reported results within a few number of runs.