This project run on Pytory.
File'cora' records the operation that output adjacency matrix generated by cora dataset. File'output' records all of the output value in txt file. File'fig.m' visualizes an adjacency matrix constructed from a dataset
There are eight improvements in total. For each improvement, two training times were set, 200 and 400 times. 'GCN_x_epoch400.py'represents same improvement as 'GCN_x.py' but epoch time increase from 200 to 400 'GCN_1.py' keeps original improvement but adding the random number seed so that the results of each training run are the same. The principle can be simply understood as: every time the program is run, the variables of the model are set to the same initialization parameters, so the result of each run after a specific operation is the same. This can reproduce the previous experimental results. 'GCN_2.py' adds SD regularization. 'GCN_3.py' adds generalized cross entropy (gce) loss function and sample weighting. 'GCN_3_1.py' adds dynamic learning rate decay On the basis of 'GCN_3.py' 'GCN_4.py' changes the network structure from sigle to double. 'GCN_6.py' change the way of weighting. weighting ce loss and gce loss separately 'GCN_7.py' deletes the hidden layer 'GCN_8.py' change the way of weighting. The double network structure is weighted separately 'GCN_9.py' changes the network structure, express input layer in another way