gcomdensemerge

Takes a 4d Tensor (-1,gs,ags,param) and transforms it using a simple Dense Layer into a 3d one (-1,gs,paramo). Respects Graph permutation Symmetry. So Basically Transforms each set of vectors into one vector of different size

Arguments

  • gs: The Number of Lists of Feature Vectors and the Number of Nodes in the Output
  • ags: The Number of Vectors in each List of Input vector, could be understood as the opposite of c
  • param: The Number of Features for each Input Node
  • paramo: The Number of Features each Outputvector should have
  • initializer=glorot_uniform: The Transformation initializer
  • learnabel=True: Shall the Transformation be learnable