Allow choice of HDBSCAN's cluster_selection_method

netatmoqc/clustering.py

@@ -204,9 +204,9 @@ def run_clustering_on_df(
             metric="precomputed",
             core_dist_n_jobs=n_jobs,
             allow_single_cluster=True,
-            # Default cluster_selection_method: 'eom'. Sometimes it leads to
-            # clusters that are too big. Using 'leaf' seems better.
-            cluster_selection_method="leaf",
+            cluster_selection_method=config.get_clustering_opt(
+                "cluster_selection_method"
+            ),
         ).fit(distance_matrix)
     elif method == "optics":
         db = OPTICS(

netatmoqc/config_parser.py

@@ -417,6 +417,12 @@ with config_section("clustering_method.hdbscan") as section:
         choices=[None, "glosh", "lof", "iterative", "reclustering"],
     )
     config_metadata.copy_template("outlier_removal.iterative")
+    # 'eom' may sometimes lead to clusters that are too big and lower
+    # silhouette scores, but maybe the final station selection is better
+    # than that using "leaf"
+    config_metadata.register(
+        "cluster_selection_method", default="leaf", choices=["eom", "leaf"]
+    )
 
 # clustering_method.dbscan
 with config_section("clustering_method.dbscan") as section: