Once the clusters are created, you can use Cluster Profiling to retain only those clusters that correspond to patterns that are meaningful to the business and look distinct.
Cluster profiling allows you to visually inspect the clusters created and retain only meaningful clusters or merge clusters showing similar behavior.
Use the ofs_auto_ml.feature_transform.ts_clustering Python module and the aif.save_transformation_results Python API to apply cluster profiling.
The following is the input:
· X: Behavioral data as a pandas data frame.
· feature: Name of the input feature for profiling.
· bmp_type: Type of Feature Extraction - clip/trend
· k: Max clusters to consider for profiling
· num_of_samples: Number of samples to be considered from each Cluster for Visualization (Default=10)
To apply cluster profiling, execute the commands as shown in the following paragraph:
|
%python
aif.set_plot_dimension( width = 14, height = 9 )
ts_obj.cluster_profiles( X = B_OSIT_PDF, feature = 'TOT_DEPST_AM', bmp_type = 'CLIP', k = 7, num_of_samples = 10 )
aif.show_plots() |