search

akensert / molgraph

Graph neural networks for molecular machine learning. Implemented and compatible with TensorFlow and Keras.
https://molgraph.readthedocs.io/en/latest/
MIT License
47 stars 5 forks source link

Positional Encoding questions ? #13

Closed thegodone closed 1 year ago

thegodone commented 1 year ago

Which one of the following PE's can be added to the current code containing I guess LPE ?

https://github.com/cptq/SignNet-BasisNet/blob/main/GraphPrediction/data/molecules.py

This is from this paper: https://arxiv.org/pdf/2202.13013.pdf

akensert commented 1 year ago

They use the Laplacian positional encoding. Looks like the same as molgraph.layers.LaplacianPositionalEncoding. I did not immediately see the random flipping of the sign of the eigen vectors, but I guess they will perform it later on in the model along with a linear transformation.