trustagi-lab / arga Goto Github PK

In the paper, the reported results of clustering on cora dataset are :

Cora     Acc     NMI     F1    Precision    ARI
ARGE    0.640   0.449   0.619    0.646     0.352
ARVGE   0.638   0.450   0.627    0.624     0.374

In the experiment section, ARGA is not compared with gae. When I remove the adversarial module and use simple gae(not vae), the results on cora dataset look like this:

Epoch: 0030 ACC=0.6403 f1_macro=0.6222 f1_micro=0.6403 NMI=0.4732 ADJ_RAND_SCORE=0.4154
Epoch: 0032 ACC=0.6448 f1_macro=0.6264 f1_micro=0.6448 NMI=0.4796 ADJ_RAND_SCORE=0.4211
Epoch: 0034 ACC=0.6470 f1_macro=0.6276 f1_micro=0.6470 NMI=0.4782 ADJ_RAND_SCORE=0.4256
Epoch: 0036 ACC=0.6536 f1_macro=0.6362 f1_micro=0.6536 NMI=0.4809 ADJ_RAND_SCORE=0.4264
Epoch: 0038 ACC=0.6558 f1_macro=0.6397 f1_micro=0.6558 NMI=0.4786 ADJ_RAND_SCORE=0.4245
Epoch: 0040 ACC=0.6603 f1_macro=0.6435 f1_micro=0.6603 NMI=0.4840 ADJ_RAND_SCORE=0.4305
Epoch: 0042 ACC=0.6448 f1_macro=0.6356 f1_micro=0.6448 NMI=0.4771 ADJ_RAND_SCORE=0.4081
Epoch: 0044 ACC=0.6514 f1_macro=0.6430 f1_micro=0.6514 NMI=0.4745 ADJ_RAND_SCORE=0.4051
Epoch: 0046 ACC=0.6518 f1_macro=0.6433 f1_micro=0.6518 NMI=0.4786 ADJ_RAND_SCORE=0.4088
Epoch: 0048 ACC=0.6629 f1_macro=0.6464 f1_micro=0.6629 NMI=0.4655 ADJ_RAND_SCORE=0.4236
Epoch: 0050 ACC=0.6510 f1_macro=0.6404 f1_micro=0.6510 NMI=0.4722 ADJ_RAND_SCORE=0.4075

It seems that the adversarial module doesn't help much to improve the performance?

Can u publish the details of the node clustering experiments in your papers?
eg.
the repeat count of kmeans
is the experiment result best value or average value

Traceback (most recent call last):
File "ARGA/arga/run.py", line 18, in
runner.erun()
File "ARGA/arga/link_prediction.py", line 24, in erun
feas = format_data(self.data_name)
File "ARGA/arga/constructor.py", line 42, in format_data
adj, features, y_test, tx, ty, test_maks, true_labels = load_data(data_name)
File "ARGA/arga/input_data.py", line 24, in load_data
objects.append(pkl.load(open("data/ind.{}.{}".format(dataset, names[i]))))
File ".pyenv/versions/3.5.1/lib/python3.5/codecs.py", line 321, in decode
(result, consumed) = self._buffer_decode(data, self.errors, final)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x80 in position 0: invalid start byte

Hi @Ruiqi-Hu,
I see here https://github.com/Ruiqi-Hu/ARGA/blob/a970fa583d8c474b18f950da06bf91da03a647db/ARGA/arga/model.py#L67
you injected a Gaussian noise to the first layer.

• What's the purpose of doing so?

Based on the original implementation of GAE,
https://github.com/tkipf/gae/blob/a1aecb00de894859f509f2d392135086a86946a0/gae/model.py#L55-L69
there is no noise, and

• have you compared the results with and without the noise of ARGA with GAE?

• Actually adding noise will affect the Gaussian prior p(z), right?

1 . change all_edge_idx = range(edges.shape[0]) to all_edge_idx = list(range(edges.shape[0])) in preprocessing.py
2 . change objects.append(pkl.load(open("data/ind.{}.{}".format(dataset, names[i])))) to
objects.append(pkl.load(open("data/ind.{}.{}".format(dataset, names[i]),'rb'),encoding='latin1')) in input_data.py

Is there a way to embed multiple graphs with these models?

Or can I only use one graph at a time?
Thanks! This is amazing work!

Traceback (most recent call last):
File "/Users/linda/mywork/ARGA-master/ARGA/arga/run.py", line 18, in
runner.erun()
File "/Users/linda/mywork/ARGA-master/ARGA/arga/link_prediction.py", line 27, in erun
placeholders = get_placeholder(feas['adj'])
KeyError: 'adj'

Hi Ruiqi,
Thanks for your nice work! Does your code support running on GPU? I haven't modified it successfully.

Thank you very much!

preprocessing.py -> def mask_test_edges(adj): -> def ismember(a, b, tol=5):

The actual function is different from the meaning of the name. why???

Actual function： is same rows (https://www.coder.work/article/2401015)
Surface name function： is member

import numpy as np

def ismember(a, b, tol=5):
  rows_close = np.all(np.round(a - b[:, None], tol) == 0, axis=-1)
  return (np.all(np.any(rows_close, axis=-1), axis=-1) and np.all(np.any(rows_close, axis=0), axis=0))

#[is same rows]
a = np.array([[ 6,  7],
              [ 2,  3],
              [ 0,  1],
              [ 4,  5],
              [ 8,  9]])
b = np.array([[ 0,  1],
              [ 2,  3],
              [ 4,  5],
              [ 6,  7],
              [ 8,  9]])
print(ismember(a, b)) 
# >> True

#[is member]
a1 = np.array([[ 0,  1]]) # shape=(1, 2)
print(ismember(a1, b)) 
# >> False # error

a2 = np.array([ 0,  1]) # shape=(2,)
print(ismember(a2, b)) 
# >> False # error

Hi, it's a nice work and thanks for sharing the code.
I have a puzzle, in arga_vae model, when you define the OptimizerVAE in optimizer.py, the generator_loss doesn't plus the auto-encoder cost
self.generator_loss = tf.reduce_mean( tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(d_fake), logits=d_fake))
rather than
self.generator_loss = tf.reduce_mean( tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(d_fake), logits=d_fake))+self.cost.
Can you please tell me the reason to do this?
Thanks a lot.

How did you solve it in the end，When I use Python 37 + cuda10, tensorflow-gpu 1.14 appears / lib64/ libm.so .6: version `GLIBC_ 2.23' not found

There's an error when I run the model:
<UnicodeDecodeError: 'gbk' codec can't decode byte 0x80 in position 0: illegal multibyte sequence>
This error occurs during loading dataset files in line:
objects.append(pkl.load(open("data/ind.{}.{}".format(dataset, names[i])))) #input_data.py->load_data
I can NOT load these dataset files...
".x",".y"... these files are unreadable, so I can't get the content and format. Could you give some data examples and explain each file?

Sorry...but is there something wrong with my operation?

Traceback (most recent call last):
File "C:/Users/lily_bud/Desktop/ARGA-master/ARGA-master/ARGA/arga/run.py", line 3, in
from clustering import Clustering_Runner
File "C:\Users\lily_bud\Desktop\ARGA-master\ARGA-master\ARGA\arga\clustering.py", line 3, in
from sklearn.cluster import KMeans
File "C:\Python36\lib\site-packages\sklearn_init_.py", line 134, in
from .base import clone
File "C:\Python36\lib\site-packages\sklearn\base.py", line 12, in
from .utils.fixes import signature
File "C:\Python36\lib\site-packages\sklearn\utils_init_.py", line 7, in
from scipy.sparse import issparse
ImportError: cannot import name 'issparse'

Process finished with exit code 1

Create README me

Hi Ruiqi,
Thanks for your nice work! I tried to run ARGA on weighted adjacency matrix (A) and non-binary features (X), but it seems didn't work. Do you have any suggestions about this?