[ENH] OWLouvain: Ensure deterministic clustering

Orange/clustering/louvain.py

@@ -8,6 +8,8 @@
 
 import numpy as np
 import networkx as nx
+# NOTE: The ``community`` package might be renamed in the near future, see
+# GH issue https://github.com/taynaud/python-louvain/issues/23
 from community import best_partition
 from sklearn.neighbors import NearestNeighbors
 
@@ -53,18 +55,27 @@ def table_to_knn_graph(data, k_neighbors, metric, progress_callback=None):
         if progress_callback:
             progress_callback(idx / num_nodes)
 
-        for neighbour in nearest_neighbors[node]:
-            graph.add_edge(node, neighbour, weight=jaccard(
-                nearest_neighbors[node], nearest_neighbors[neighbour]))
+        for neighbor in nearest_neighbors[node]:
+            graph.add_edge(
+                node,
+                neighbor,
+                weight=jaccard(nearest_neighbors[node], nearest_neighbors[neighbor]),
+            )
 
     return graph
 
 
 class Louvain:
-    preprocessors = [Orange.preprocess.Continuize(),
-                     Orange.preprocess.SklImpute()]
-
-    def __init__(self, k_neighbors=30, metric='l2', resolution=1., preprocessors=None):
+    preprocessors = [Orange.preprocess.Continuize(), Orange.preprocess.SklImpute()]
+
+    def __init__(
+            self,
+            k_neighbors=30,
+            metric="l2",
+            resolution=1.0,
+            random_state=None,
+            preprocessors=None,
+    ):
         """Louvain clustering for community detection in graphs.
 
         Louvain clustering is a community detection algorithm for detecting
@@ -77,12 +88,21 @@ def __init__(self, k_neighbors=30, metric='l2', resolution=1., preprocessors=Non
         k_neighbors : Optional[int]
             The number of nearest neighbors to use for the KNN graph if
             tabular data is passed.
+
         metric : Optional[str]
             The metric to use to compute the nearest neighbors.
+
         resolution : Optional[float]
             The resolution is a parameter of the Louvain method that affects
             the size of the recovered clusters.
 
+        random_state: Union[int, RandomState]
+            The random state parameter follows the convention used in scikit-learn.
+            If the value is an int, random_state is the seed used by the random
+            number generator. If the value is a RandomState instance, then it will
+            be used as the random number generator. If the value is None, the random
+            number generator is the RandomState instance used by `np.random`.
+
         """
         if preprocessors is None:
             preprocessors = type(self).preprocessors
@@ -91,6 +111,7 @@ def __init__(self, k_neighbors=30, metric='l2', resolution=1., preprocessors=Non
         self.k_neighbors = k_neighbors
         self.metric = metric
         self.resolution = resolution
+        self.random_state = random_state
 
         self.labels = None
 
@@ -106,14 +127,20 @@ def preprocess(self, data):
     def fit(self, X, y=None):
         # If we are given a table, we have to convert it to a graph first
         if isinstance(X, Table):
-            graph = table_to_knn_graph(X.X, metric=self.metric, k_neighbors=self.k_neighbors)
+            graph = table_to_knn_graph(
+                X.X, metric=self.metric, k_neighbors=self.k_neighbors
+            )
         # Same goes for a matrix
         elif isinstance(X, np.ndarray):
-            graph = table_to_knn_graph(X, metric=self.metric, k_neighbors=self.k_neighbors)
+            graph = table_to_knn_graph(
+                X, metric=self.metric, k_neighbors=self.k_neighbors
+            )
         elif isinstance(X, nx.Graph):
             graph = X
 
-        partition = best_partition(graph, resolution=self.resolution)
+        partition = best_partition(
+            graph, resolution=self.resolution, random_state=self.random_state
+        )
         partition = np.fromiter(list(zip(*sorted(partition.items())))[1], dtype=int)
 
         self.labels = partition

Orange/widgets/unsupervised/owlouvainclustering.py

@@ -211,7 +211,7 @@ def commit(self):
 
         # Preprocess the dataset
         if self.preprocessed_data is None:
-            louvain = Louvain()
+            louvain = Louvain(random_state=0)
             self.preprocessed_data = louvain.preprocess(self.data)
 
         state = TaskState(self)
@@ -590,7 +590,7 @@ def pcallback(val):
     if state.is_interuption_requested():
         return res
 
-    louvain = Louvain(resolution=resolution)
+    louvain = Louvain(resolution=resolution, random_state=0)
     res.partition = louvain.fit_predict(graph)
     state.set_partial_results(("partition", res.partition))
     return res
@@ -603,7 +603,7 @@ def run_on_graph(graph, resolution, state):
     """
     state = state  # type: TaskState
     res = Results(resolution=resolution)
-    louvain = Louvain(resolution=resolution)
+    louvain = Louvain(resolution=resolution, random_state=0)
     state.set_status("Detecting communities...")
     if state.is_interuption_requested():
         return res

Orange/widgets/unsupervised/tests/test_owlouvain.py

@@ -161,3 +161,20 @@ def test_invalidate(self):
         self.widget.commit()
         self.get_output(self.widget.Outputs.annotated_data)
         self.assertFalse(self.widget.Information.modified.is_shown())
+
+    def test_deterministic_clustering(self):
+        # Compute clustering on iris
+        self.send_signal(self.widget.Inputs.data, self.iris)
+        self.commit_and_wait()
+        result1 = self.get_output(self.widget.Outputs.annotated_data)
+
+        # Reset widget state
+        self.send_signal(self.widget.Inputs.data, None)
+
+        # Compute clustering on iris again
+        self.send_signal(self.widget.Inputs.data, self.iris)
+        self.commit_and_wait()
+        result2 = self.get_output(self.widget.Outputs.annotated_data)
+
+        # Ensure that clustering was the same in both instances
+        np.testing.assert_equal(result1.metas, result2.metas)