GitLab

The script examples/clustering.py now contains the example of chain for clustering #175 (closed) .

Added:

• tuning for clustering
• ensemble clustering
• more clustering models

Known issues:

• Silhouette metric is used for the composition, but it is not appropriate here (the value converges to -1 very quickly).
• DBScan model is added, but it should be more models to run the experiments
• The number of clusters for K-means is pre-defined now (should be identified from data)
• The secondary-only ensembling model shold be added for clustering to mix several predictions

Example of the composite clustering model:

Current metrics:

adjusted_rand_score for basic model 0.350061 with iris adjusted_rand_score for composite model 0.568116 with iris adjusted_rand_score for basic model 0.508453 with simple data adjusted_rand_score for tuned model 0.570815 with simple data