Louvain Clustering

Detects communities in a network of nearest neighbours.

Inputs
Data
input dataset
Outputs
Data
dataset with a new cluster label
Network
network graph

The Louvain Clustering widget is a non-parametric clustering based on community detection in networks using the Louvain method. The method optimizes network modularity by looking at in-group versus out-of-group edge density.

../_images/LouvainClustering-stamped.png
  1. If ‘Apply PCA preprocessing’ is ticked, data will be transformed with PCA prior to clustering. Slider enables you to select the number of PCA components for clustering, maximum is 50.
  2. Graph parameters: - Use Euclidean or Manhattan distance metric. - Set k-neighbors for local clustering. - Resolution at which to observe the network. Default of 1.0 returns the macro level.
  3. Apply changes. If Apply automatically is ticked, changes will be communicated automatically. Alternatively, click Apply.
  4. Access help.

Example

We use iris data set for this simple example. Load iris with the File widget and send it to Louvain Clustering. The widget computes clustering and outputs the data with an additional column with cluster labels. Connect Scatter Plot to Louvain Clustering to observe the discovered clusters. In Scatter Plot, set the Color option to Cluster. Louvain recognized 4 clusters, Iris setosa, Iris virginica and Iris versicolor clusters, with the fourth cluster being at an intersection of virginica and versicolor.

If you have the Network add-on installed, you can connect Network Explorer to Louvain Clustering and observe the clusters in a network.

../_images/LouvainClustering-Example.png