gcomjpool¶
Takes a Feature Vector of size (-1,gs,param) and some indices representing order indices (-1,gs) and uses these indices to reorders the Feature Vectors and applies a simple Dense Transformation to transform it into (-1,gs/c,paramo)
Arguments
- gs: The initial Number of Nodes in each Graph.
- param: The initial Number of Features.
- c=2: How many Feature Vectors to combine into each Outputvector. gs has to divide this
- paramo: The Number of Features in each Outputvector
- metrik_init=”glorot_uniform”: The initializer of the Transformation
- trainable=True: weather the Transformation is trainable