sparsekmeans 0.2__tar.gz → 0.2.2__tar.gz

sparsekmeans-0.2.2/PKG-INFO

@@ -0,0 +1,24 @@
+Metadata-Version: 2.4
+Name: sparsekmeans
+Version: 0.2.2
+Summary: A package for efficient implementation of the K-means clustering on high-dimensional sparse data sets
+Home-page: https://github.com/cjlin1/sparsekmeans
+Author: Chih-Jen Lin, He-Zhe Lin, Khoi Nguyen Pham Dang
+Author-email: cjlin@csie.ntu.edu.tw
+License: MIT
+Requires-Python: >=3.10
+License-File: LICENSE
+Requires-Dist: numpy
+Requires-Dist: python-graphblas
+Requires-Dist: scipy
+Dynamic: author
+Dynamic: author-email
+Dynamic: description
+Dynamic: home-page
+Dynamic: license
+Dynamic: license-file
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary
+
+See documentation here: https://github.com/cjlin1/sparsekmeans

{sparsekmeans-0.2 → sparsekmeans-0.2.2}/README.md

@@ -2,6 +2,8 @@
 
 The sparsekmeans package provides an efficient implementation of the K-means clustering algorithm optimized for sparse data sets. It is designed to handle high-dimensional and sparse data commonly found in text mining, recommender systems, and bioinformatics. By leveraging appropriate storage format and sparse matrix multiplication operations, our package ensures significant speedup in running time while maintaining consistency for clustering results compared with scikit-learn. Besides, the design of the package allows users to easily extend and customize, making it suitable for research or integrating into large-scale machine learning systems.
 
+Implementation details are in the paper ["SparseKmeans: Efficient K-means Clustering for Sparse Data"](https://www.csie.ntu.edu.tw/~cjlin/papers/sparse_kmeans/sparsekmeans-cikm.pdf), CIKM 2025.
+
 ## Installation
 
 Use the following command to install sparsekmeans with python >= 3.10.

{sparsekmeans-0.2 → sparsekmeans-0.2.2}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup
 
 PACKAGE_DIR = ["sparsekmeans"]
 PACKAGE_NAME = "sparsekmeans"
-VERSION = "0.2"
+VERSION = "0.2.2"
 
 
 # license parameters
@@ -17,8 +17,8 @@ setup(
     version=VERSION,
     python_requires=">=3.10",
     install_requires=["numpy","python-graphblas","scipy"],
-    description="",
-    long_description_content_type="",
+    description="A package for efficient implementation of the K-means clustering on high-dimensional sparse data sets",
+    long_description="See documentation here: https://github.com/cjlin1/sparsekmeans",
     author="Chih-Jen Lin, He-Zhe Lin, Khoi Nguyen Pham Dang",
     author_email="cjlin@csie.ntu.edu.tw",
     url="https://github.com/cjlin1/sparsekmeans",

{sparsekmeans-0.2 → sparsekmeans-0.2.2}/sparsekmeans/sparse_kmeans.py

@@ -33,7 +33,7 @@ def squared_row_norms(X: gb.Matrix, n_threads: int=default_threads):
 
 
 def check_centroids_density(centroids: gb.Matrix):
-    gamma = 0.1
+    gamma = 0.05
 
     if centroids.ss.format == "fullr":
         is_centroid_dense = True
@@ -52,6 +52,10 @@ def check_centroids_density(centroids: gb.Matrix):
 
 
 def predict_labels(X: gb.Matrix, centroids: gb.Matrix, is_centroid_dense: bool, n_threads: int=default_threads):
+    if X.dtype != dtypes.FP64:
+        X = X.dup(dtype="FP64")
+    if centroids.dtype != dtypes.FP64:
+        centroids = centroids.dup(dtype="FP64")
     n_samples = X.shape[0]
     n_clusters = centroids.shape[0]
 
@@ -65,16 +69,13 @@ def predict_labels(X: gb.Matrix, centroids: gb.Matrix, is_centroid_dense: bool,
         centroids = centroids.ss.export("fullc")
         centroids = gb.Matrix.ss.import_fullc(**centroids)
 
-        XCt << 0
-        XCt(accum=gb.binary.plus, nthreads=n_threads) << X.mxm(centroids.T)
-        XCt = 2 * XCt.to_dense()
-
     else:
         centroids = centroids.ss.export("csc")
         centroids = gb.Matrix.ss.import_csc(**centroids)
 
-        XCt(nthreads=n_threads) << X.mxm(centroids.T)
-        XCt = 2 * XCt.to_dense(fill_value=0)
+    XCt << 0
+    XCt(accum=gb.binary.plus, nthreads=n_threads) << X.mxm(centroids.T)
+    XCt = 2 * XCt.to_dense()
 
     # x_squared_norms is not needed
     c_squared_norms = squared_row_norms(centroids, n_threads)
@@ -298,6 +299,8 @@ class SparseKmeans:
 
         if not isinstance(X, gb.Matrix):
             X = gb.io.from_scipy_sparse(X)
+        if X.dtype != dtypes.FP64:
+            X = X.dup(dtype="FP64")
 
         start_init_centroids = time.time()
         self.centroids = self._initialize_centroids(X)
@@ -355,6 +358,7 @@ class SparseKmeans:
         # If a selected point belongs to a cluster with only one point, we skip it and move on to the next furthest point.
         # Each empty cluster gets one point
         # We also need to update cluster_sizes
+        # See Algorithm E.1 in supplementary materials of Dang et al., CIKM 2025.
         if n_empty > 0:
             far_samples_idx = np.argsort(self.sample_centroids_closest_distance)[:: -1]
             i, j = 0, 0

sparsekmeans-0.2.2/sparsekmeans.egg-info/PKG-INFO

@@ -0,0 +1,24 @@
+Metadata-Version: 2.4
+Name: sparsekmeans
+Version: 0.2.2
+Summary: A package for efficient implementation of the K-means clustering on high-dimensional sparse data sets
+Home-page: https://github.com/cjlin1/sparsekmeans
+Author: Chih-Jen Lin, He-Zhe Lin, Khoi Nguyen Pham Dang
+Author-email: cjlin@csie.ntu.edu.tw
+License: MIT
+Requires-Python: >=3.10
+License-File: LICENSE
+Requires-Dist: numpy
+Requires-Dist: python-graphblas
+Requires-Dist: scipy
+Dynamic: author
+Dynamic: author-email
+Dynamic: description
+Dynamic: home-page
+Dynamic: license
+Dynamic: license-file
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary
+
+See documentation here: https://github.com/cjlin1/sparsekmeans

sparsekmeans-0.2/PKG-INFO

@@ -1,12 +0,0 @@
-Metadata-Version: 2.1
-Name: sparsekmeans
-Version: 0.2
-Home-page: https://github.com/cjlin1/sparsekmeans
-Author: Chih-Jen Lin, He-Zhe Lin, Khoi Nguyen Pham Dang
-Author-email: cjlin@csie.ntu.edu.tw
-License: MIT
-Requires-Python: >=3.10
-License-File: LICENSE
-Requires-Dist: numpy
-Requires-Dist: python-graphblas
-Requires-Dist: scipy

sparsekmeans-0.2/sparsekmeans.egg-info/PKG-INFO

@@ -1,12 +0,0 @@
-Metadata-Version: 2.1
-Name: sparsekmeans
-Version: 0.2
-Home-page: https://github.com/cjlin1/sparsekmeans
-Author: Chih-Jen Lin, He-Zhe Lin, Khoi Nguyen Pham Dang
-Author-email: cjlin@csie.ntu.edu.tw
-License: MIT
-Requires-Python: >=3.10
-License-File: LICENSE
-Requires-Dist: numpy
-Requires-Dist: python-graphblas
-Requires-Dist: scipy