liveBook · Manning

Property prediction layer

As explained in Graph Neural Networks in Action

A neural-network layer that maps a learned representation to a predicted property value.

A property prediction layer maps a graph’s learned latent representation to a predicted property value. In molecular graph generation, it adds property control alongside graph reconstruction. section 5.4.3

The implementation follows a short pipeline:

  • Encode the graph with a heterogeneous GCN, preserving distinctions between bond types. section 5.4.3

Molecular property prediction

A VGAE predicts QED by defining property_prediction_layer = nn.Linear(latent_dim, 1), then producing qed_pred from the sampled latent representation.

How Graph Neural Networks in Action covers it

How deep it goes
Defines a one-output linear head and places it beside reconstruction, without detailing its training loss.
Examples
Uses QED prediction within a molecular graph generation model.
What it assumes
Assumes an encoder, decoder, latent dimension, and variational reparameterization workflow.
What you can do after
Build it from scratch with PyTorch’s nn.Linear and connect it to a sampled latent vector.
How it connects
Connects variational graph autoencoders, heterogeneous graph convolution, latent representations, and graph decoding.
Cover of Graph Neural Networks in Action
liveBookmanning.com
Read Graph Neural Networks in Action on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Graph Neural Networks in Action. The section links point to the parts of the book it draws on.