- Google Maps uses graph ML to predict ETA
- Netflix uses graph ML in recommendation
- Spotify uses graph ML in recommendation
- Pinterest uses graph ML in recommendation
Here are 6 must-know ways for graph feature engineering (with code):
Like images, text, and tabular datasets have features, so do graph datasets.
This means when building models on graph datasets, we can engineer these features to achieve better performance.
Let's discuss some feature engineering techniques below!
This means when building models on graph datasets, we can engineer these features to achieve better performance.
Let's discuss some feature engineering techniques below!
First, letās create a dummy social networking graph dataset with accounts and followers (which will also be accounts).
We create the two DataFrames shown below, an accounts DataFrame and a followers DataFrame.
Check this codeš
We create the two DataFrames shown below, an accounts DataFrame and a followers DataFrame.
Check this codeš
The above data is tabular, but we need to convert this into a graph format.
To do this, we use the networkx framework as follows:
ā¢ Initialize a graphĀ G.
ā¢ AddĀ nodes from theĀ accountsĀ DF.
ā¢ Added edges between the nodes using theĀ followers'Ā DF.
Here's the codeš
To do this, we use the networkx framework as follows:
ā¢ Initialize a graphĀ G.
ā¢ AddĀ nodes from theĀ accountsĀ DF.
ā¢ Added edges between the nodes using theĀ followers'Ā DF.
Here's the codeš
This produces the following graph.
Next, let's cover the 6 graph feature engineering techniques.
Next, let's cover the 6 graph feature engineering techniques.
1-3) Node degree
In a directed graph, there are two types of degrees:
ā¢ In-Degree:Ā The number of incoming edges (followers) a node has.
ā¢ Out-Degree:Ā The number of outgoing edges (followings) a node has.
In a directed graph, there are two types of degrees:
ā¢ In-Degree:Ā The number of incoming edges (followers) a node has.
ā¢ Out-Degree:Ā The number of outgoing edges (followings) a node has.
Hereās how we can compute them using NetworkX:
ā¢ in_degree(x)Ā counts edges directed toward the nodeĀ x.
ā¢ out_degree(x)Ā counts edges directed away from the nodeĀ x.
ā¢ degree(x)Ā is the sum of the in-degree and out-degree of nodeĀ x.
Check this codeš
ā¢ in_degree(x)Ā counts edges directed toward the nodeĀ x.
ā¢ out_degree(x)Ā counts edges directed away from the nodeĀ x.
ā¢ degree(x)Ā is the sum of the in-degree and out-degree of nodeĀ x.
Check this codeš
These features are now part of the accounts DataFrame.
Check thisš
Check thisš
4-6) Node centrality
Node degree features capture connectedness but fail to capture the influence of those connections.
For instance, a user can have many online friends just because they send friend requests to everyone.
Centrality features handle this.
Node degree features capture connectedness but fail to capture the influence of those connections.
For instance, a user can have many online friends just because they send friend requests to everyone.
Centrality features handle this.
4) Betweenness centrality
This measures how often a node appears on the shortest paths between other nodes.
If a node often acts as a ābridgeā between other nodes, it plays a key role in facilitating information flow.
Here's the codeš
This measures how often a node appears on the shortest paths between other nodes.
If a node often acts as a ābridgeā between other nodes, it plays a key role in facilitating information flow.
Here's the codeš
5) Closeness centrality
This indicates how close a node is to all other nodes in the network based on the shortest paths.
To compute closeness centrality for a node v, we sum the shortest path length from v to all other nodes and take its reciprocal.
Here's the codeš
This indicates how close a node is to all other nodes in the network based on the shortest paths.
To compute closeness centrality for a node v, we sum the shortest path length from v to all other nodes and take its reciprocal.
Here's the codeš
6) Eigenvector centrality
If a node is connected to other influential nodes, it amplifies its own influence.
It helps identify nodes that are influential not only due to their direct ties but also due to their connections with other influential nodes.
Here's the codeš
If a node is connected to other influential nodes, it amplifies its own influence.
It helps identify nodes that are influential not only due to their direct ties but also due to their connections with other influential nodes.
Here's the codeš
PyTorch Geometric is a PyTorch extension specifically developed for building graph-based neural networks.
It has an intuitive API that facilitates inspecting and analyzing graphs and building ML models on graph-based datasets.
Open-source with 22k+ stars!
It has an intuitive API that facilitates inspecting and analyzing graphs and building ML models on graph-based datasets.
Open-source with 22k+ stars!
Generated by Thread Navigator
Press ā + S to quick-export