DGL graph classification

Questions
#1

I would like to dgl graph so learnig for Grpah Convolutional Network model.
but I think I don’t know how to make make dgl graph.

I would like to question how to make dgl graph? like a exactly nodes featears and edges fearears.
I’m going to learing exactly nodes featears and edges fearears.

below example my sensor data
sensor0,sensor1,sensor2,sensor3,sensor4,label
59.6,24.7,274,2251,6333,2
44.8,21.6,212,2408,8585,2
43.6,21.4,98,12,416,0

and I drawed and learning graph like below
#graph [
Graph(num_nodes=5, num_edges=20,
ndata_schemes={‘x’: Scheme(shape=(), dtype=torch.float64)}
edata_schemes={}),
Graph(num_nodes=5, num_edges=20,
ndata_schemes={‘x’: Scheme(shape=(), dtype=torch.float64)}
edata_schemes={}),
Graph(num_nodes=5, num_edges=20,
ndata_schemes={‘x’: Scheme(shape=(), dtype=torch.float64)}
edata_schemes={}),
Graph(num_nodes=5, num_edges=20,
ndata_schemes={‘x’: Scheme(shape=(), dtype=torch.float64)}
edata_schemes={})]
#label [2, 0, 2, 0]

print(graph[0].ndata[‘x’] )
tensor([ 59.6000, 24.9000, 363.0000, 8849.0000, 22325.0000],
dtype=torch.float64)

image
image

My goal is graph classification

How to draw graph of my sensor data and make GCN model and learning?

Thank you for your insteresting.

#2

It seems that all your graphs have the same number of nodes and edges. Do they have the same graph structure? Have you read this on graph classification? Training a GNN for Graph Classification — DGL 1.1.1 documentation

1 Like
#3

I tried it like below(above link).

import matplotlib.pyplot as plt
import networkx as nx
…
fig, ax = plt.subplots()
nx.draw(graphs[0].to_networkx(), ax=ax, with_labels=True)
image
nx.draw(graphs[1]…
image
nx.draw(graphs[2]…
image

but my data is like below.
sensor0,sensor1,sensor2,sensor3,sensor4,label
59.6,24.7,274,2251,6333,2
44.8,21.6,212,2408,8585,2
43.6,21.4,98,12,416,0
…

What do you think about how to draw feature of nodes and feature of edges my data with dgl graph?
I think there are 5 nodes in my data at one row.
and one row is a graph.

And then how to learning use GCN model?
please help me!!

#4

Is sensor0 to sensor4 respectively the feature of each node? What are their semantic meanings?

1 Like
#5

@mufeili They are humidity, temperature, MQ139, TVOC, eCO2 sensors.
And they are fire occupied or not fire occupied dataset. In case fire happened are some causes of three fire situation.

#6

It sounds like these are just different features. I’m not sure if you really need a graph to model them. Maybe it suffices to treat them as tabular data.

1 Like
#7

Um … I have question.
How to check self node edge (identity matrix , unit matrix)?
in dgl graph

and How to check relation of node and edge ? in dgl graph

#8

I think GCN is suffices to know situation.

#9

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.