fix silhouette evaluation (Closes: #23)

Author: kno10

CHANGELOG.md

@@ -2,6 +2,11 @@
 
 For changes to the main Rust package, please see <https://github.com/kno10/rust-kmedoids/blob/main/CHANGELOG.md>
 
+## kmedoids 0.4.3 (2023-04-20)
+
+- fix silhouette evaluation for k > 2 (in Rust)
+- use np.unique in Python wrapper to ensure labels are 0..k
+
 ## kmedoids 0.4.2 (2023-03-07)
 
 - fix predict for non-precomputed distances

CITATION.cff

@@ -10,8 +10,8 @@ authors:
 title: "Fast k-medoids Clustering in Rust and Python"
 journal: "J. Open Source Softw."
 doi: 10.21105/joss.04183
-version: 0.4.2
-date-released: 2022-09-24
+version: 0.4.3
+date-released: 2023-04-20
 license: GPL-3.0
 preferred-citation:
   title: "Fast k-medoids Clustering in Rust and Python"

Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 edition = "2021"
 name = "kmedoids"
-version = "0.4.2"
+version = "0.4.3"
 authors = ["Erich Schubert <erich.schubert@tu-dortmund.de>", "Lars Lenssen <lars.lenssen@tu-dortmund.de>"]
 description = "k-Medoids clustering with the FasterPAM algorithm"
 homepage = "https://github.com/kno10/python-kmedoids"
@@ -14,7 +14,7 @@ name = "kmedoids"
 crate-type = ["cdylib"]
 
 [dependencies]
-rustkmedoids = { version = "0.4.2", package = "kmedoids", git = "https://github.com/kno10/rust-kmedoids" }
+rustkmedoids = { version = "0.4.3", package = "kmedoids", git = "https://github.com/kno10/rust-kmedoids" }
 numpy = "0.18"
 ndarray = "0.15"
 rand = "0.8"

kmedoids/__init__.py

@@ -617,8 +617,7 @@ def silhouette(diss, labels, samples=False, n_cpu=-1):
 
 	if not isinstance(diss, np.ndarray):
 		diss = np.array(diss)
-	if not isinstance(labels, np.ndarray):
-		labels = np.array(labels, dtype=np.uint64)
+	labels = np.unique(labels, return_inverse=True)[1].astype(np.uint64) # ensure labels are 0..k-1
 
 	if isinstance(diss, np.ndarray):
 		dtype = diss.dtype
@@ -628,20 +627,20 @@ def silhouette(diss, labels, samples=False, n_cpu=-1):
 		if n_cpu > 1:
 			assert not samples, "samples=true currently may only be used with n_cpu=1"
 			if dtype == np.float32:
-				return (_par_silhouette_f32(diss, labels.astype(np.uint64), n_cpu), [])
+				return (_par_silhouette_f32(diss, labels, n_cpu), [])
 			elif dtype == np.float64:
-				return (_par_silhouette_f64(diss, labels.astype(np.uint64), n_cpu), [])
+				return (_par_silhouette_f64(diss, labels, n_cpu), [])
 			elif dtype == np.int32:
-				return (_par_silhouette_i32(diss, labels.astype(np.uint64), n_cpu), [])
+				return (_par_silhouette_i32(diss, labels, n_cpu), [])
 			elif dtype == np.int64:
 				raise ValueError("Input of int64 is currently not supported, as it could overflow the float64 used internally when computing Silhouette. Use diss.astype(numpy.float64) if that is acceptable and you have the necessary memory for this copy.")
 		else:
 			if dtype == np.float32:
-				return _silhouette_f32(diss, labels.astype(np.uint64), samples)
+				return _silhouette_f32(diss, labels, samples)
 			elif dtype == np.float64:
-				return _silhouette_f64(diss, labels.astype(np.uint64), samples)
+				return _silhouette_f64(diss, labels, samples)
 			elif dtype == np.int32:
-				return _silhouette_i32(diss, labels.astype(np.uint64), samples)
+				return _silhouette_i32(diss, labels, samples)
 			elif dtype == np.int64:
 				raise ValueError("Input of int64 is currently not supported, as it could overflow the float64 used internally when computing Silhouette. Use diss.astype(numpy.float64) if that is acceptable and you have the necessary memory for this copy.")
 	raise ValueError("Input data not supported. Use a numpy array of floats.")

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "maturin"
 
 [project]
 name = "kmedoids"
-version = "0.4.2"
+version = "0.4.3"
 description = "k-Medoids Clustering in Python with FasterPAM"
 requires-dist = ["numpy"]
 classifier = [

tests/test_integration.py

@@ -47,7 +47,7 @@ def test_pamsil(self):
         dist = np.array([[0, 2, 3, 4, 5], [2, 0, 6, 7, 8], [3, 6, 0, 9, 10], [4, 7, 9, 0, 11], [5, 8, 10, 11, 0]], dtype=np.float32)
         pamsil = kmedoids.pamsil(dist, 2)
         pamsil_rust = kmedoids.kmedoids._pamsil_swap_f32(dist, pamsil.medoids, 100)
-        assert pamsil.loss == 0.5137878787878788
+        assert pamsil.loss == 0.3137878787878788
         assert pamsil.loss == pamsil_rust[0]
         assert np.array_equal(pamsil.medoids, pamsil_rust[2])