TensorFlow - 使用 networkx 图作为输入

Question

I am trying to use TensorFlow to build an artificial neural network which can identify features in a CAD model.

Using the CAD model data I have computed the adjacency between certain faces and plotted it on a graph where each node is a face, each edge between faces represents adjacency and weights of 0 and 1 represent a convex and concave relationship respectively. Below is the code which draws the adjacency graph:

import networkx as nx
import matplotlib.pyplot as plt
%matplotlib inline

G=nx.Graph()

#adding nodes
G.add_nodes_from(range(1, 10))

#adding edges
G.add_weighted_edges_from([(1, 2, 0), (1, 3, 0), (1, 6, 1), (1, 8, 1),
(2, 5, 1), (2, 6, 1), (2, 8, 1), (3, 4, 1),(3, 6, 1), (3, 8, 1), 
(4, 6, 1), (4, 7, 1), (4, 8, 1), (5, 6, 1), (5, 8, 1), (5, 9, 1),
(6, 7, 1), (6, 9, 1), (6, 10, 1), (7, 8, 1), (7, 10, 1), (8, 9, 1), 
(8, 10, 1), (9, 10, 1)])

#draw AAG
nx.draw_circular(G, node_color = 'bisque', with_labels=True)

My question is: is it possible to feed the adjacency graphs into a neural network in TensorFlow, teaching it which faces make up the 'features' present in the model?

Any help on the subject would be appreciated

Answer 1

In principle you could, assuming you can present your graph as a list-like data structure. But the question is, how are you going to train your network? First, you need to have thousands (and thousands) of samples of such graphs, so you can feed them in batches into your TF model. Second, for each of the samples, you will need to make a clear distinction of what represents the features and what represents the label. You can't expect TF to figure out WHAT makes up the features, you will need to SPECIFY that.