SDNE implementation error

[isikcakmak]

Dear authors,
I have issues using your implementation of sdne algorithm.
when I run "!pip install git+https://github.com/palash1992/GEM.git, I get this ERROR: textgenrnn 1.4.1 has requirement keras>=2.1.5, but you'll have keras 2.0.2 which is incompatible.

I am using keras version 2.3.0-tf,tensorflow version=2.2.0-rc3,networkx=2.4
I tried to upgrade to the previous version of tensorflow.
I tried "from tensorflow.python.keras import backend as k"

When I run embedding/sdne.py I get error:
Using TensorFlow backend.
Num nodes: 34, num edges: 77

AttributeError Traceback (most recent call last)
in ()
36 t1 = time()
37 # Learn embedding - accepts a networkx graph or file with edge list
---> 38 Y, t = embedding.learn_embedding(graph=G, edge_f=None, is_weighted=True, no_python=True)
39 print (embedding._method_name+':\n\tTraining time: %f' % (time() - t1))
40 # Evaluate on graph reconstruction

4 frames
/usr/local/lib/python3.6/dist-packages/gem/embedding/sdne.py in learn_embedding(self, graph, edge_f, is_weighted, no_python)
89 self._K, self._n_units,
90 self._nu1, self._nu2,
---> 91 self._actfn)
92 self._decoder = get_decoder(self._node_num, self._d,
93 self._K, self._n_units,

/usr/local/lib/python3.6/dist-packages/gem/embedding/sdne_utils.py in get_encoder(node_num, d, K, n_units, nu1, nu2, activation_fn)
63 def get_encoder(node_num, d, K, n_units, nu1, nu2, activation_fn):
64 # Input
---> 65 x = Input(shape=(node_num,))
66 # Encoder layers
67 y = [None] * (K + 1)

/usr/local/lib/python3.6/dist-packages/keras/engine/topology.py in Input(shape, batch_shape, name, dtype, sparse, tensor)
1386 name=name, dtype=dtype,
1387 sparse=sparse,
-> 1388 input_tensor=tensor)
1389 # Return tensor including _keras_shape and _keras_history.
1390 # Note that in this case train_output and test_output are the same pointer.

/usr/local/lib/python3.6/dist-packages/keras/engine/topology.py in init(self, input_shape, batch_size, batch_input_shape, dtype, input_tensor, sparse, name)
1251 if not name:
1252 prefix = 'input'
-> 1253 name = prefix + '_' + str(K.get_uid(prefix))
1254 super(InputLayer, self).init(dtype=dtype, name=name)
1255

/usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py in get_uid(prefix)
45 def get_uid(prefix=''):
46 global _GRAPH_UID_DICTS
---> 47 graph = tf.get_default_graph()
48 if graph not in _GRAPH_UID_DICTS:
49 _GRAPH_UID_DICTS[graph] = defaultdict(int)

AttributeError: module 'tensorflow' has no attribute 'get_default_graph'

Best regards,
ışık

[palash1992]

Please have a look here: keras-team/keras#12379

[isikcakmak]

Hi,Palash Goyal;
I looked keras-team/keras#12379 .I applied the solutions he said for the previous issue. but my problem was not solved. I took these error messages in SDNE implementation. I tried the things you suggested.

I didn’t solve my problems with 'from tensorflow.python.keras import backend'
I tried all of them.

I changed Tensorflow_version==1.13.1 and keras ==2.2.4 but I took this error .

Keras==2.2.4
Tensorflow==1.13.1

[juanpablo-munoz]

You need to downgrade your tensorflow install to 1.xx. I installed 1.14 and it worked. You still need to force install keras==2.0.2 though.