Merge pull request #61 from pmacg/sklearn-dependency

Fix sklearn dependency

check.sh

@@ -3,7 +3,7 @@ echo "|----------------|"
 echo "| Running pytest |"
 echo "|----------------|"
 cd tests || exit
-python3 -m pytest || exit
+python -m pytest || exit
 cd .. || exit
 echo ""
 echo ""

docs/source/changelog.rst

@@ -1,6 +1,13 @@
 Changelog
 =========
 
+Unreleased
+----------
+
+**Fixed**
+
+* Update sklearn dependency to scikit-learn
+
 0.4.6 - 2022-01-11
 ------------------
 

requirements.txt

@@ -1,9 +1,9 @@
 scipy
-sklearn
+scikit-learn
 pytest
 furo
 numpy
 pylint
-networkx
+networkx<3.0
 matplotlib
 pandas

setup.py

@@ -1,6 +1,6 @@
 from setuptools import setup, find_packages
 
-VERSION = '0.4.6'
+VERSION = '0.5.0'
 DESCRIPTION = 'SGTL - Spectral Graph Theory Library'
 LONG_DESCRIPTION =\
     "This library provides several methods and algorithms relating to spectral graph theory in python.\n\n" \
@@ -15,7 +15,7 @@
     description=DESCRIPTION,
     long_description=LONG_DESCRIPTION,
     packages=find_packages(),
-    install_requires=["numpy", "scipy", "sklearn"],
+    install_requires=["numpy", "scipy", "scikit-learn", "networkx<3.0", "matplotlib", "pandas"],
     long_description_content_type='text/markdown',
 
     keywords=['python', 'spectral', 'graph', 'algorithms', 'clustering', 'cheeger'],

sgtl/algorithms.py

@@ -34,7 +34,7 @@ def _sweep_set(graph: sgtl.Graph, vector: List[float]) -> Tuple[Set[int], Set[in
     vector = degree_matrix.power(-(1 / 2)).dot(vector)
 
     # First, sort the vertices based on their value in the given vector
-    sorted_vertices = [i for i, _ in sorted(enumerate(vector), key=(lambda y: y[1]))]
+    sorted_vertices = [i for i, _ in sorted(enumerate(vector), key=lambda y: y[1])]
 
     # Keep track of which edges to add/subtract from the cut each time
     edges_to_add = np.ones(num_vertices)

sgtl/clustering.py

@@ -40,7 +40,7 @@ def spectral_clustering(graph: sgtl.Graph, num_clusters: int, num_eigenvectors=N
     _, eigenvectors = scipy.sparse.linalg.eigsh(laplacian_matrix, num_eigenvectors, which='SM')
 
     # Perform k-means on the eigenvectors to find the clusters
-    labels = KMeans(n_clusters=num_clusters).fit_predict(eigenvectors)
+    labels = KMeans(n_clusters=num_clusters, n_init=10).fit_predict(eigenvectors)
 
     # Split the clusters.
     clusters = [[] for _ in range(num_clusters)]

sgtl/graph.py

@@ -185,7 +185,7 @@ def volume(self, vertex_set):
         :return: The volume of vertex_set
         """
         self._check_vert_num(vertex_set)
-        return sum([self.degrees[v] for v in vertex_set])
+        return sum(self.degrees[v] for v in vertex_set)
 
     def weight(self,
                vertex_set_l: List[int],
@@ -214,13 +214,13 @@ def weight(self,
         # self-loops
         if sets_are_equal:
             weight_to_remove = raw_weight / 2
-            weight_to_remove -= sum([self.adjacency_matrix()[i, i] for i in vertex_set_l]) / 2
+            weight_to_remove -= sum(self.adjacency_matrix()[i, i] for i in vertex_set_l) / 2
         elif not check_for_overlap:
             weight_to_remove = 0
         else:
             overlap = set.intersection(set(vertex_set_l), set(vertex_set_r))
             weight_to_remove = self.lil_adj_mat[list(overlap)][:, list(overlap)].sum() / 2
-            weight_to_remove -= sum([self.adjacency_matrix()[i, i] for i in overlap]) / 2
+            weight_to_remove -= sum(self.adjacency_matrix()[i, i] for i in overlap) / 2
 
         # Return the corrected weight
         return raw_weight - weight_to_remove
@@ -472,7 +472,7 @@ def knn_graph(data, k: int):
     :return: An ``sgtl.Graph`` object representing the ``k``-nearest neighbour graph of the input.
     """
     # Create the nearest neighbours for each vertex using sklearn
-    _, neighbours = NearestNeighbors(n_neighbors=(k+1)).fit(data).kneighbors(data)
+    _, neighbours = NearestNeighbors(n_neighbors=k+1).fit(data).kneighbors(data)
 
     # Now, let's construct the adjacency matrix of the graph iteratively
     adj_mat = scipy.sparse.lil_matrix((data.shape[0], data.shape[0]))