Node similarity metric learning module cannot perform gradient backpropagation

[syiswell]

The tensor in the Node similarity metric learning module cannot perform gradient backpropagation due to the greater operation, which leads to the parameters cannot being updated.

[jhljx]

I have the same confusion, too. This module also uses some ops, such as tf.sort. I don't know whether the weight param in cosine similarity calculation can get gradient back-propagation or not.

[ShowTBC]

\begin{equation}
A_{i j}= \begin{cases}1, & M_{i j}>=\operatorname{Rank}_{\varepsilon n^2}(M) \ 0, & \text { otherwise }\end{cases}
\end{equation}
I think this operation cannot perform gradient backpropagation