tkipf
what is the element of model.reconstructions in model.py?
Sorry, but I can't understand at the line 114 in file model.py, what's the element of self.reconstruction? probability value? But when I run the program, I print the variable reconstructions, and get the result: reconstruction [ 2.3758938 -0.10602921 0.103589 ... 0.8698976 3.3979175 3.6787224 ]
These are logits -- you can get probability values / normalized scores by passing these through a logistic sigmoid function.
On Thu, Jul 4, 2019 at 9:59 AM yjyGo [email protected] wrote:
Sorry, but I can't understand at the line 114 in file model.py, what's the element of self.reconstruction? probability value? But when I run the program, I print the variable reconstructions, and get the result: reconstruction [ 2.3758938 -0.10602921 0.103589 ... 0.8698976 3.3979175 3.6787224 ]
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/tkipf/gae/issues/41?email_source=notifications&email_token=ABYBYYHTBD5RT3NOAR2IYL3P5WUVZA5CNFSM4H5VFIEKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4G5J6AQA, or mute the thread https://github.com/notifications/unsubscribe-auth/ABYBYYCIMOUJ3IY55B4LOQDP5WUVZANCNFSM4H5VFIEA .
I'm currently trying to use GAE to do graph representation learning, thus I need to extract node features with dimension nxd, with n number of nodes and d number of features. Or so-called embeddings Z according to Kipf https://arxiv.org/pdf/1611.07308.pdf
As far, as I can see, the node features would be in model.reconstructions (@tkipf could you maybe confirm?).
However, I'm not able to get a vector of values like your example (@yjyGo). I can see shape and datatype (on the Cora example)
Tensor("gcnmodelae/innerproductdecoder_1/Reshape:0", shape=(?,), dtype=float32)
Would you mind sharing your code chunk?
@yjyGo got it: the embeddings can be extracted with
output = sess.run(model.z_mean, feed_dict=feed_dict)