fix(cluster): fix edge case with connected_components() and no edges …

…but records

The .max() of the labels was getting called on a column with 0 entries, so the result was NULL

mismo/cluster/_connected_components.py

@@ -301,7 +301,8 @@ def _label_datasets(ds: Datasets, labels: ir.Table, *, label_as: str) -> Dataset
 def _get_additional_labels(labels: ir.Table, record_ids: ir.Column) -> ir.Table:
     nodes = record_ids.name("record_id").as_table()
     is_missing_label = nodes.record_id.notin(labels.record_id)
-    max_existing_label = labels.component.max()
+    # fill_null(0) in case labels is empty
+    max_existing_label = labels.component.max().fill_null(0)
     additional_labels = nodes.filter(is_missing_label).select(
         "record_id",
         component=(ibis.row_number() + max_existing_label + 1).cast("int64"),

mismo/cluster/test/test_connected_components.py

@@ -120,6 +120,17 @@ def test_cc_single_records(table_factory, label_as):
     assert clusters == {frozenset({0, 1, 2}), frozenset({3})}
 
 
+def test_cc_no_links_but_records(table_factory, label_as):
+    """If there are no links, each record should be its own cluster."""
+    link_schema = {"record_id_l": "int64", "record_id_r": "int64"}
+    link_df = pd.DataFrame({"record_id_l": [], "record_id_r": []})
+    links = table_factory(link_df, schema=link_schema)
+    nodes = table_factory({"record_id": [0, 1, 2]})
+    labeled = connected_components(links=links, records=nodes, label_as=label_as)
+    clusters = _labels_to_clusters(labeled, label_as)
+    assert clusters == {frozenset({0}), frozenset({1}), frozenset({2})}
+
+
 def test_cc_multi_records(table_factory, label_as):
     # multiple input record tables
     links = table_factory([(0, 1), (1, 2)], columns=["record_id_l", "record_id_r"])