different nodes' embeddings are quite similar in their absolute value in each dimension. Do you perform some column normalization for the downstream task, e.g. the node structure role classification?
In detail, for example, the output chi for
x1= [1.0 0.0 0.999737033462 0.000552053084944 0.999157958099]
x2 = [1.0 0.0 0.997427493114 0.00121764919074 0.9980411342 ]
For each dimension of x1,x2 , the value is quite similar. Do you do some normalization in order to use it?
different nodes' embeddings are quite similar in their absolute value in each dimension. Do you perform some column normalization for the downstream task, e.g. the node structure role classification?
In detail, for example, the output chi for x1= [1.0 0.0 0.999737033462 0.000552053084944 0.999157958099] x2 = [1.0 0.0 0.997427493114 0.00121764919074 0.9980411342 ] For each dimension of x1,x2 , the value is quite similar. Do you do some normalization in order to use it?