prosemble 2.3.0__tar.gz → 2.4.0__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (328) hide show
  1. {prosemble-2.3.0 → prosemble-2.4.0}/PKG-INFO +1 -1
  2. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/distance.py +183 -0
  3. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/manifolds.py +166 -0
  4. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/__init__.py +14 -2
  5. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/celvq_ng_mixin.py +11 -3
  6. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_glvq_ng.py +17 -2
  7. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_matrix_lvq_ng.py +17 -2
  8. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_relevance_lvq_ng.py +17 -2
  9. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/mrslvq_ng.py +18 -6
  10. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_dk_glvq.py +8 -6
  11. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_dk_matrix_lvq.py +5 -4
  12. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_dk_relevance_lvq.py +5 -4
  13. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_glvq.py +56 -9
  14. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_glvq_ng_mixin.py +10 -2
  15. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_gmlvq.py +20 -2
  16. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_grlvq.py +15 -4
  17. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_gtlvq.py +15 -2
  18. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_lgmlvq.py +22 -2
  19. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_mrslvq.py +9 -5
  20. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_rslvq.py +1 -1
  21. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_rslvq_ng_mixin.py +10 -4
  22. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_glvq.py +6 -2
  23. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_m_slng.py +12 -1
  24. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_m_smng.py +12 -1
  25. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_m_stng.py +12 -1
  26. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_matrix_lvq.py +10 -3
  27. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_r_slng.py +12 -1
  28. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_r_smng.py +12 -1
  29. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_r_stng.py +12 -1
  30. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_relevance_lvq.py +10 -3
  31. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_slng.py +12 -1
  32. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_smng.py +12 -1
  33. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_dk_stng.py +12 -1
  34. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_neural_gas.py +6 -1
  35. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_slng.py +10 -2
  36. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_smng.py +15 -3
  37. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_srng.py +100 -5
  38. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/riemannian_stng.py +11 -2
  39. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/rslvq_ng.py +9 -3
  40. prosemble-2.4.0/prosemble/models/scmng.py +289 -0
  41. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/slng.py +17 -2
  42. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/smng.py +17 -2
  43. prosemble-2.4.0/prosemble/models/sng.py +259 -0
  44. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/srng.py +18 -2
  45. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/stng.py +17 -2
  46. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/svq_occ.py +2 -2
  47. prosemble-2.4.0/prosemble/models/wasserstein_glvq.py +231 -0
  48. prosemble-2.4.0/prosemble/models/wasserstein_gmlvq.py +270 -0
  49. prosemble-2.4.0/prosemble/models/wasserstein_grlvq.py +259 -0
  50. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble.egg-info/PKG-INFO +1 -1
  51. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble.egg-info/SOURCES.txt +11 -1
  52. {prosemble-2.3.0 → prosemble-2.4.0}/pyproject.toml +1 -1
  53. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/api/models.rst +8 -0
  54. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/conf.py +1 -1
  55. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/guides/onnx.rst +4 -1
  56. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/guides/riemannian.rst +91 -2
  57. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/guides/supervised.rst +147 -2
  58. prosemble-2.4.0/tests/test_hyperbolic.py +292 -0
  59. prosemble-2.4.0/tests/test_ng_lr_ratio.py +153 -0
  60. prosemble-2.4.0/tests/test_scmng.py +258 -0
  61. prosemble-2.4.0/tests/test_sng.py +225 -0
  62. prosemble-2.4.0/tests/test_wasserstein.py +302 -0
  63. {prosemble-2.3.0 → prosemble-2.4.0}/uv.lock +202 -208
  64. {prosemble-2.3.0 → prosemble-2.4.0}/.bumpversion.cfg +0 -0
  65. {prosemble-2.3.0 → prosemble-2.4.0}/.editorconfig +0 -0
  66. {prosemble-2.3.0 → prosemble-2.4.0}/.envrc +0 -0
  67. {prosemble-2.3.0 → prosemble-2.4.0}/.github/ISSUE_TEMPLATE/bug_report.md +0 -0
  68. {prosemble-2.3.0 → prosemble-2.4.0}/.github/ISSUE_TEMPLATE/config.yml +0 -0
  69. {prosemble-2.3.0 → prosemble-2.4.0}/.github/ISSUE_TEMPLATE/feature_request.md +0 -0
  70. {prosemble-2.3.0 → prosemble-2.4.0}/.github/workflows/ci.yml +0 -0
  71. {prosemble-2.3.0 → prosemble-2.4.0}/.github/workflows/release.yml +0 -0
  72. {prosemble-2.3.0 → prosemble-2.4.0}/.gitignore +0 -0
  73. {prosemble-2.3.0 → prosemble-2.4.0}/.readthedocs.yaml +0 -0
  74. {prosemble-2.3.0 → prosemble-2.4.0}/CONSTITUTION.md +0 -0
  75. {prosemble-2.3.0 → prosemble-2.4.0}/LICENSE +0 -0
  76. {prosemble-2.3.0 → prosemble-2.4.0}/MANIFEST.in +0 -0
  77. {prosemble-2.3.0 → prosemble-2.4.0}/ONNX_COVERAGE.md +0 -0
  78. {prosemble-2.3.0 → prosemble-2.4.0}/README.md +0 -0
  79. {prosemble-2.3.0 → prosemble-2.4.0}/TASK.md +0 -0
  80. {prosemble-2.3.0 → prosemble-2.4.0}/benchmarks/benchmark_distance_jax.py +0 -0
  81. {prosemble-2.3.0 → prosemble-2.4.0}/benchmarks/benchmark_fcm_jax.py +0 -0
  82. {prosemble-2.3.0 → prosemble-2.4.0}/benchmarks/benchmark_pcm_jax.py +0 -0
  83. {prosemble-2.3.0 → prosemble-2.4.0}/devenv.lock +0 -0
  84. {prosemble-2.3.0 → prosemble-2.4.0}/devenv.nix +0 -0
  85. {prosemble-2.3.0 → prosemble-2.4.0}/devenv.yaml +0 -0
  86. {prosemble-2.3.0 → prosemble-2.4.0}/examples/afcm_iris.py +0 -0
  87. {prosemble-2.3.0 → prosemble-2.4.0}/examples/bgpc_iris.py +0 -0
  88. {prosemble-2.3.0 → prosemble-2.4.0}/examples/cbc_iris.py +0 -0
  89. {prosemble-2.3.0 → prosemble-2.4.0}/examples/celvq_ng_iris.py +0 -0
  90. {prosemble-2.3.0 → prosemble-2.4.0}/examples/fcm_iris.py +0 -0
  91. {prosemble-2.3.0 → prosemble-2.4.0}/examples/fpcm_iris.py +0 -0
  92. {prosemble-2.3.0 → prosemble-2.4.0}/examples/glvq_advanced_iris.py +0 -0
  93. {prosemble-2.3.0 → prosemble-2.4.0}/examples/glvq_iris.py +0 -0
  94. {prosemble-2.3.0 → prosemble-2.4.0}/examples/glvq_unequal_prototypes_iris.py +0 -0
  95. {prosemble-2.3.0 → prosemble-2.4.0}/examples/gmlvq_iris.py +0 -0
  96. {prosemble-2.3.0 → prosemble-2.4.0}/examples/grlvq_iris.py +0 -0
  97. {prosemble-2.3.0 → prosemble-2.4.0}/examples/hcm_iris.py +0 -0
  98. {prosemble-2.3.0 → prosemble-2.4.0}/examples/heskes_som_iris.py +0 -0
  99. {prosemble-2.3.0 → prosemble-2.4.0}/examples/ipcm2_iris.py +0 -0
  100. {prosemble-2.3.0 → prosemble-2.4.0}/examples/ipcm_iris.py +0 -0
  101. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kafcm_iris.py +0 -0
  102. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kfcm_iris.py +0 -0
  103. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kfpcm_iris.py +0 -0
  104. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kipcm2_iris.py +0 -0
  105. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kipcm_iris.py +0 -0
  106. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kmeans_iris.py +0 -0
  107. {prosemble-2.3.0 → prosemble-2.4.0}/examples/knn_iris.py +0 -0
  108. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kohonen_som_iris.py +0 -0
  109. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kpcm_iris.py +0 -0
  110. {prosemble-2.3.0 → prosemble-2.4.0}/examples/kpfcm_iris.py +0 -0
  111. {prosemble-2.3.0 → prosemble-2.4.0}/examples/large_dataset_training.py +0 -0
  112. {prosemble-2.3.0 → prosemble-2.4.0}/examples/lcelvq_ng_iris.py +0 -0
  113. {prosemble-2.3.0 → prosemble-2.4.0}/examples/lmrslvq_iris.py +0 -0
  114. {prosemble-2.3.0 → prosemble-2.4.0}/examples/lmrslvq_ng_iris.py +0 -0
  115. {prosemble-2.3.0 → prosemble-2.4.0}/examples/lvq1_iris.py +0 -0
  116. {prosemble-2.3.0 → prosemble-2.4.0}/examples/lvqmln_iris.py +0 -0
  117. {prosemble-2.3.0 → prosemble-2.4.0}/examples/mcelvq_ng_iris.py +0 -0
  118. {prosemble-2.3.0 → prosemble-2.4.0}/examples/mrslvq_iris.py +0 -0
  119. {prosemble-2.3.0 → prosemble-2.4.0}/examples/mrslvq_ng_iris.py +0 -0
  120. {prosemble-2.3.0 → prosemble-2.4.0}/examples/neural_gas_iris.py +0 -0
  121. {prosemble-2.3.0 → prosemble-2.4.0}/examples/npc_iris.py +0 -0
  122. {prosemble-2.3.0 → prosemble-2.4.0}/examples/oc_lmrslvq_iris.py +0 -0
  123. {prosemble-2.3.0 → prosemble-2.4.0}/examples/oc_lmrslvq_ng_iris.py +0 -0
  124. {prosemble-2.3.0 → prosemble-2.4.0}/examples/oc_mrslvq_iris.py +0 -0
  125. {prosemble-2.3.0 → prosemble-2.4.0}/examples/oc_mrslvq_ng_iris.py +0 -0
  126. {prosemble-2.3.0 → prosemble-2.4.0}/examples/oc_rslvq_iris.py +0 -0
  127. {prosemble-2.3.0 → prosemble-2.4.0}/examples/oc_rslvq_ng_iris.py +0 -0
  128. {prosemble-2.3.0 → prosemble-2.4.0}/examples/pcm_iris.py +0 -0
  129. {prosemble-2.3.0 → prosemble-2.4.0}/examples/pfcm_iris.py +0 -0
  130. {prosemble-2.3.0 → prosemble-2.4.0}/examples/plvq_iris.py +0 -0
  131. {prosemble-2.3.0 → prosemble-2.4.0}/examples/rslvq_ng_iris.py +0 -0
  132. {prosemble-2.3.0 → prosemble-2.4.0}/examples/slng_iris.py +0 -0
  133. {prosemble-2.3.0 → prosemble-2.4.0}/examples/smng_iris.py +0 -0
  134. {prosemble-2.3.0 → prosemble-2.4.0}/examples/som_iris.py +0 -0
  135. {prosemble-2.3.0 → prosemble-2.4.0}/examples/srng_iris.py +0 -0
  136. {prosemble-2.3.0 → prosemble-2.4.0}/examples/stng_iris.py +0 -0
  137. {prosemble-2.3.0 → prosemble-2.4.0}/examples/tcelvq_ng_iris.py +0 -0
  138. {prosemble-2.3.0 → prosemble-2.4.0}/examples/vis_fcm_iris.py +0 -0
  139. {prosemble-2.3.0 → prosemble-2.4.0}/examples/vis_glvq_iris.py +0 -0
  140. {prosemble-2.3.0 → prosemble-2.4.0}/examples/vis_neural_gas_iris.py +0 -0
  141. {prosemble-2.3.0 → prosemble-2.4.0}/examples/vis_som_iris.py +0 -0
  142. {prosemble-2.3.0 → prosemble-2.4.0}/flake.nix +0 -0
  143. {prosemble-2.3.0 → prosemble-2.4.0}/mkdocs.yml +0 -0
  144. {prosemble-2.3.0 → prosemble-2.4.0}/nix/package.nix +0 -0
  145. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/__init__.py +0 -0
  146. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/__init__.py +0 -0
  147. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/activations.py +0 -0
  148. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/callbacks.py +0 -0
  149. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/competitions.py +0 -0
  150. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/curriculum.py +0 -0
  151. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/data.py +0 -0
  152. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/distributed.py +0 -0
  153. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/geodesic.py +0 -0
  154. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/initializers.py +0 -0
  155. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/kernel.py +0 -0
  156. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/losses.py +0 -0
  157. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/model_selection.py +0 -0
  158. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/onnx_export.py +0 -0
  159. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/optimizers.py +0 -0
  160. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/pipeline.py +0 -0
  161. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/pooling.py +0 -0
  162. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/protocols.py +0 -0
  163. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/quantization.py +0 -0
  164. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/regularization.py +0 -0
  165. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/reject.py +0 -0
  166. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/serialization.py +0 -0
  167. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/similarities.py +0 -0
  168. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/utils.py +0 -0
  169. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/vis.py +0 -0
  170. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/core/visualizer.py +0 -0
  171. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/datasets/__init__.py +0 -0
  172. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/datasets/dataset.py +0 -0
  173. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/afcm.py +0 -0
  174. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/base.py +0 -0
  175. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/bgpc.py +0 -0
  176. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/cbc.py +0 -0
  177. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/celvq_ng.py +0 -0
  178. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/crossentropy_lvq.py +0 -0
  179. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_glvq.py +0 -0
  180. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_heskes_som.py +0 -0
  181. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_kohonen_som.py +0 -0
  182. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_matrix_lvq.py +0 -0
  183. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_neural_gas.py +0 -0
  184. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/dk_relevance_lvq.py +0 -0
  185. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/fcm.py +0 -0
  186. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/fpcm.py +0 -0
  187. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/glvq.py +0 -0
  188. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/growing_neural_gas.py +0 -0
  189. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/hcm.py +0 -0
  190. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/heskes_som.py +0 -0
  191. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/image_cbc.py +0 -0
  192. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/image_lvq.py +0 -0
  193. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/ipcm.py +0 -0
  194. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/ipcm2.py +0 -0
  195. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kafcm.py +0 -0
  196. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kfcm.py +0 -0
  197. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kfpcm.py +0 -0
  198. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kipcm.py +0 -0
  199. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kipcm2.py +0 -0
  200. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kmeans.py +0 -0
  201. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/knn.py +0 -0
  202. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kohonen_som.py +0 -0
  203. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kpcm.py +0 -0
  204. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/kpfcm.py +0 -0
  205. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/lcelvq_ng.py +0 -0
  206. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/local_matrix_lvq.py +0 -0
  207. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/lvq1.py +0 -0
  208. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/lvq21.py +0 -0
  209. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/lvqmln.py +0 -0
  210. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/matrix_lvq.py +0 -0
  211. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/matrix_rslvq.py +0 -0
  212. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/mcelvq_ng.py +0 -0
  213. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/median_lvq.py +0 -0
  214. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/neural_gas.py +0 -0
  215. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/npc.py +0 -0
  216. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_dk_glvq_ng.py +0 -0
  217. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_dk_matrix_lvq_ng.py +0 -0
  218. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_dk_relevance_lvq_ng.py +0 -0
  219. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_glvq_ng.py +0 -0
  220. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_gmlvq_ng.py +0 -0
  221. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_grlvq_ng.py +0 -0
  222. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_gtlvq_ng.py +0 -0
  223. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_lgmlvq_ng.py +0 -0
  224. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_mrslvq_ng.py +0 -0
  225. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/oc_rslvq_ng.py +0 -0
  226. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/pcm.py +0 -0
  227. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/pfcm.py +0 -0
  228. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/plvq.py +0 -0
  229. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/probabilistic_lvq.py +0 -0
  230. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/prototype_base.py +0 -0
  231. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/relevance_lvq.py +0 -0
  232. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/siamese_lvq.py +0 -0
  233. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/som.py +0 -0
  234. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/svq_occ_lm.py +0 -0
  235. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/svq_occ_m.py +0 -0
  236. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/svq_occ_r.py +0 -0
  237. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/svq_occ_t.py +0 -0
  238. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/tangent_lvq.py +0 -0
  239. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/models/tcelvq_ng.py +0 -0
  240. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble/py.typed +0 -0
  241. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble.egg-info/dependency_links.txt +0 -0
  242. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble.egg-info/requires.txt +0 -0
  243. {prosemble-2.3.0 → prosemble-2.4.0}/prosemble.egg-info/top_level.txt +0 -0
  244. {prosemble-2.3.0 → prosemble-2.4.0}/requirements_docs.txt +0 -0
  245. {prosemble-2.3.0 → prosemble-2.4.0}/setup.cfg +0 -0
  246. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/Makefile +0 -0
  247. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/api/core.rst +0 -0
  248. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/guides/advanced.rst +0 -0
  249. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/guides/clustering.rst +0 -0
  250. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/guides/differentiating_kernels.rst +0 -0
  251. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/guides/one_class.rst +0 -0
  252. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/guides/topology.rst +0 -0
  253. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/index.rst +0 -0
  254. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/installation.rst +0 -0
  255. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/make.bat +0 -0
  256. {prosemble-2.3.0 → prosemble-2.4.0}/sphinx-docs/quickstart.rst +0 -0
  257. {prosemble-2.3.0 → prosemble-2.4.0}/tests/__init__.py +0 -0
  258. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_activations.py +0 -0
  259. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_cbc.py +0 -0
  260. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_celvq_ng.py +0 -0
  261. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_checkpoint.py +0 -0
  262. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_clustering_patience.py +0 -0
  263. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_competitions.py +0 -0
  264. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_core.py +0 -0
  265. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_core_jax.py +0 -0
  266. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_data.py +0 -0
  267. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_dataset.py +0 -0
  268. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_dataset_jax.py +0 -0
  269. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_distance_jax.py +0 -0
  270. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_distributed.py +0 -0
  271. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_dk_models.py +0 -0
  272. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_dk_ng_models.py +0 -0
  273. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_examples.sh +0 -0
  274. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_fcm_jax.py +0 -0
  275. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_glvq.py +0 -0
  276. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_gmlvq.py +0 -0
  277. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_heskes_som.py +0 -0
  278. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_image_lvq.py +0 -0
  279. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_improvements.py +0 -0
  280. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_initializers.py +0 -0
  281. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_jax_models.py +0 -0
  282. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_jit_inference.py +0 -0
  283. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_kohonen_som.py +0 -0
  284. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_lcelvq_ng.py +0 -0
  285. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_losses.py +0 -0
  286. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_lvq_nongradient.py +0 -0
  287. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_lvqmln_plvq.py +0 -0
  288. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_matrix_rslvq.py +0 -0
  289. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_mcelvq_ng.py +0 -0
  290. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_minibatch.py +0 -0
  291. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_mixed_precision.py +0 -0
  292. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_model_selection.py +0 -0
  293. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_mrslvq_ng.py +0 -0
  294. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_neural_gas.py +0 -0
  295. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_oc_dk_models.py +0 -0
  296. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_oc_dk_ng_models.py +0 -0
  297. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_oc_glvq.py +0 -0
  298. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_oc_glvq_ng.py +0 -0
  299. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_oc_glvq_variants.py +0 -0
  300. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_oc_mrslvq.py +0 -0
  301. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_oc_rslvq.py +0 -0
  302. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_oc_rslvq_ng.py +0 -0
  303. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_onnx_export.py +0 -0
  304. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_pcm_jax.py +0 -0
  305. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_pfcm_jax.py +0 -0
  306. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_pipeline.py +0 -0
  307. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_pooling.py +0 -0
  308. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_probabilistic.py +0 -0
  309. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_protocols.py +0 -0
  310. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_prototype_base.py +0 -0
  311. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_quantization.py +0 -0
  312. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_riemannian_dk_models.py +0 -0
  313. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_riemannian_neural_gas.py +0 -0
  314. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_riemannian_supervised_ng.py +0 -0
  315. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_rslvq_ng.py +0 -0
  316. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_scan_training.py +0 -0
  317. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_serialization.py +0 -0
  318. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_siamese_lvq.py +0 -0
  319. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_similarities.py +0 -0
  320. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_slng.py +0 -0
  321. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_smng.py +0 -0
  322. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_srng.py +0 -0
  323. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_stng.py +0 -0
  324. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_svq_occ.py +0 -0
  325. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_svq_occ_r.py +0 -0
  326. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_svq_occ_variants.py +0 -0
  327. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_tcelvq_ng.py +0 -0
  328. {prosemble-2.3.0 → prosemble-2.4.0}/tests/test_vis.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: prosemble
3
- Version: 2.3.0
3
+ Version: 2.4.0
4
4
  Summary: A python project for prototype-based machine learning models
5
5
  Author-email: Nana Abeka Otoo <nanaabekaotoo@gmail.com>
6
6
  License: MIT
@@ -431,6 +431,185 @@ def tangent_distance_matrix(
431
431
  return jnp.sum(tangent_diff ** 2, axis=2)
432
432
 
433
433
 
434
+ # ============================================================================
435
+ # Wasserstein Distance Functions
436
+ # ============================================================================
437
+
438
+
439
+ @jit
440
+ def wasserstein2_distance_matrix(
441
+ X: chex.Array,
442
+ means: chex.Array,
443
+ log_variances: chex.Array
444
+ ) -> chex.Array:
445
+ """
446
+ Compute pairwise squared 2-Wasserstein distances from points to Gaussian prototypes.
447
+
448
+ Each prototype is a diagonal Gaussian :math:`\\mathcal{N}(\\mu_k, \\text{diag}(\\sigma_k^2))`.
449
+ Each input point :math:`x` is treated as a Dirac delta distribution :math:`\\delta_x`.
450
+
451
+ The squared 2-Wasserstein distance from a point to a diagonal Gaussian is:
452
+
453
+ .. math::
454
+
455
+ W_2^2(\\delta_x, \\mathcal{N}(\\mu_k, \\text{diag}(\\sigma_k^2)))
456
+ = \\sum_j (x_j - \\mu_{kj})^2 + \\sum_j \\sigma_{kj}^2
457
+
458
+ This decomposes into the squared Euclidean distance from the point to the
459
+ mean, plus the total variance (trace of covariance). Prototypes with
460
+ smaller variance are effectively "more certain" and attract nearby points
461
+ more strongly.
462
+
463
+ Parameters
464
+ ----------
465
+ X : array of shape (n, d)
466
+ Data points.
467
+ means : array of shape (p, d)
468
+ Prototype mean vectors.
469
+ log_variances : array of shape (p, d)
470
+ Log of prototype variances (ensures positivity via ``exp``).
471
+
472
+ Returns
473
+ -------
474
+ D : array of shape (n, p)
475
+ Squared 2-Wasserstein distances.
476
+
477
+ References
478
+ ----------
479
+ .. [1] Villani, C. (2009). Optimal Transport: Old and New.
480
+ Springer. Chapter 2.
481
+ .. [2] Givens, C. R. & Shortt, R. M. (1984). A class of Wasserstein
482
+ metrics for probability distributions. Michigan Math. J., 31(2).
483
+ """
484
+ chex.assert_rank(X, 2)
485
+ chex.assert_rank(means, 2)
486
+ chex.assert_rank(log_variances, 2)
487
+ chex.assert_equal(X.shape[1], means.shape[1])
488
+ chex.assert_equal(means.shape, log_variances.shape)
489
+
490
+ # Squared Euclidean from points to means
491
+ eucl = squared_euclidean_distance_matrix(X, means) # (n, p)
492
+
493
+ # Variance spread penalty per prototype
494
+ variances = jnp.exp(log_variances) # (p, d)
495
+ spread = jnp.sum(variances, axis=1) # (p,)
496
+
497
+ return eucl + spread[None, :]
498
+
499
+
500
+ @jit
501
+ def wasserstein2_omega_distance_matrix(
502
+ X: chex.Array,
503
+ means: chex.Array,
504
+ log_variances: chex.Array,
505
+ omega: chex.Array
506
+ ) -> chex.Array:
507
+ """
508
+ Squared 2-Wasserstein distance with global metric adaptation.
509
+
510
+ Projects data and means through :math:`\\Omega` before computing
511
+ the Euclidean component, while variances contribute directly:
512
+
513
+ .. math::
514
+
515
+ W_2^2(x, k) = \\|\\Omega(x - \\mu_k)\\|^2 + \\sum_j \\sigma_{kj}^2
516
+
517
+ Parameters
518
+ ----------
519
+ X : array of shape (n, d)
520
+ Data points.
521
+ means : array of shape (p, d)
522
+ Prototype mean vectors.
523
+ log_variances : array of shape (p, d)
524
+ Log of prototype variances.
525
+ omega : array of shape (d, l)
526
+ Global projection matrix.
527
+
528
+ Returns
529
+ -------
530
+ D : array of shape (n, p)
531
+ Squared 2-Wasserstein distances in projected space.
532
+ """
533
+ chex.assert_rank(X, 2)
534
+ chex.assert_rank(means, 2)
535
+ chex.assert_rank(log_variances, 2)
536
+ chex.assert_rank(omega, 2)
537
+ chex.assert_equal(X.shape[1], means.shape[1])
538
+ chex.assert_equal(means.shape, log_variances.shape)
539
+ chex.assert_equal(X.shape[1], omega.shape[0])
540
+
541
+ # Project and compute squared Euclidean in projected space
542
+ X_proj = X @ omega # (n, l)
543
+ M_proj = means @ omega # (p, l)
544
+ eucl = squared_euclidean_distance_matrix(X_proj, M_proj) # (n, p)
545
+
546
+ # Variance spread penalty
547
+ variances = jnp.exp(log_variances) # (p, d)
548
+ spread = jnp.sum(variances, axis=1) # (p,)
549
+
550
+ return eucl + spread[None, :]
551
+
552
+
553
+ @jit
554
+ def wasserstein2_relevance_distance_matrix(
555
+ X: chex.Array,
556
+ means: chex.Array,
557
+ log_variances: chex.Array,
558
+ relevances: chex.Array
559
+ ) -> chex.Array:
560
+ """
561
+ Squared 2-Wasserstein distance with feature relevance weighting.
562
+
563
+ Applies per-feature relevance weights :math:`\\lambda_j` to the
564
+ Euclidean component:
565
+
566
+ .. math::
567
+
568
+ W_2^2(x, k) = \\sum_j \\lambda_j (x_j - \\mu_{kj})^2 + \\sum_j \\sigma_{kj}^2
569
+
570
+ where :math:`\\lambda_j = \\text{softmax}(r)_j` ensures non-negative
571
+ weights that sum to 1.
572
+
573
+ Parameters
574
+ ----------
575
+ X : array of shape (n, d)
576
+ Data points.
577
+ means : array of shape (p, d)
578
+ Prototype mean vectors.
579
+ log_variances : array of shape (p, d)
580
+ Log of prototype variances.
581
+ relevances : array of shape (d,)
582
+ Raw relevance logits (softmax applied internally).
583
+
584
+ Returns
585
+ -------
586
+ D : array of shape (n, p)
587
+ Relevance-weighted squared 2-Wasserstein distances.
588
+ """
589
+ chex.assert_rank(X, 2)
590
+ chex.assert_rank(means, 2)
591
+ chex.assert_rank(log_variances, 2)
592
+ chex.assert_rank(relevances, 1)
593
+ chex.assert_equal(X.shape[1], means.shape[1])
594
+ chex.assert_equal(means.shape, log_variances.shape)
595
+ chex.assert_equal(X.shape[1], relevances.shape[0])
596
+
597
+ # Softmax relevance weights
598
+ lambdas = jax.nn.softmax(relevances) # (d,)
599
+
600
+ # Weighted squared differences
601
+ # (n, 1, d) - (1, p, d) -> (n, p, d)
602
+ diff_sq = (X[:, None, :] - means[None, :, :]) ** 2
603
+ weighted = diff_sq * lambdas[None, None, :] # (n, p, d)
604
+ eucl = jnp.sum(weighted, axis=2) # (n, p)
605
+
606
+ # Variance spread penalty
607
+ variances = jnp.exp(log_variances) # (p, d)
608
+ spread = jnp.sum(variances, axis=1) # (p,)
609
+
610
+ return eucl + spread[None, :]
611
+
612
+
434
613
  # ============================================================================
435
614
  # Kernel Functions
436
615
  # ============================================================================
@@ -764,6 +943,10 @@ __all__ = [
764
943
  'omega_distance_matrix',
765
944
  'lomega_distance_matrix',
766
945
  'tangent_distance_matrix',
946
+ # Wasserstein distance functions
947
+ 'wasserstein2_distance_matrix',
948
+ 'wasserstein2_omega_distance_matrix',
949
+ 'wasserstein2_relevance_distance_matrix',
767
950
  # Kernel functions
768
951
  'gaussian_kernel_matrix',
769
952
  'polynomial_kernel_matrix',
@@ -7,6 +7,7 @@ for manifolds commonly arising in machine learning:
7
7
  - SO(n): Special orthogonal group (rotation matrices)
8
8
  - SPD(n): Symmetric positive definite matrices
9
9
  - Grassmannian Gr(n, k): k-dimensional subspaces of R^n
10
+ - HyperbolicPoincare(d): Poincare ball model of hyperbolic space
10
11
 
11
12
  All operations are JIT-compilable and vmap-compatible.
12
13
 
@@ -15,6 +16,9 @@ References
15
16
  Schwarz, Psenickova, Villmann, Röhrbein (2026).
16
17
  Topology-Preserving Prototype Learning on Riemannian Manifolds.
17
18
  ESANN 2026.
19
+
20
+ Ganea, O., Becigneul, G., & Hofmann, T. (2018). Hyperbolic Neural
21
+ Networks. NeurIPS 2018.
18
22
  """
19
23
 
20
24
  from __future__ import annotations
@@ -323,3 +327,165 @@ class Grassmannian:
323
327
  def injectivity_radius(self, Q):
324
328
  """Injectivity radius of Gr(n,k) is :math:`\\pi/2`."""
325
329
  return jnp.pi / 2.0
330
+
331
+
332
+ # ---------------------------------------------------------------------------
333
+ # HyperbolicPoincare(d) — Poincaré Ball Model of Hyperbolic Space
334
+ # ---------------------------------------------------------------------------
335
+
336
+ def _mobius_add(x, y, eps=1e-7):
337
+ """Möbius addition in the Poincaré ball.
338
+
339
+ .. math::
340
+
341
+ x \\oplus y = \\frac{(1 + 2\\langle x, y\\rangle + \\|y\\|^2) x
342
+ + (1 - \\|x\\|^2) y}
343
+ {1 + 2\\langle x, y\\rangle + \\|x\\|^2 \\|y\\|^2}
344
+
345
+ Parameters
346
+ ----------
347
+ x, y : arrays of shape (d,)
348
+
349
+ Returns
350
+ -------
351
+ result : array of shape (d,)
352
+ """
353
+ x_sq = jnp.sum(x ** 2)
354
+ y_sq = jnp.sum(y ** 2)
355
+ xy = jnp.sum(x * y)
356
+ num = (1.0 + 2.0 * xy + y_sq) * x + (1.0 - x_sq) * y
357
+ denom = 1.0 + 2.0 * xy + x_sq * y_sq
358
+ return num / (denom + eps)
359
+
360
+
361
+ def _conformal_factor(x, eps=1e-7):
362
+ """Conformal factor :math:`\\lambda_x = 2 / (1 - \\|x\\|^2)`.
363
+
364
+ Parameters
365
+ ----------
366
+ x : array of shape (d,)
367
+
368
+ Returns
369
+ -------
370
+ scalar
371
+ """
372
+ return 2.0 / (1.0 - jnp.sum(x ** 2) + eps)
373
+
374
+
375
+ class HyperbolicPoincare:
376
+ """Poincaré ball model of hyperbolic space :math:`\\mathbb{H}^d`.
377
+
378
+ Points are vectors in :math:`\\mathbb{B}^d = \\{x \\in \\mathbb{R}^d : \\|x\\| < 1\\}`.
379
+ The Riemannian metric is :math:`g_x = \\lambda_x^2 g_E` where
380
+ :math:`\\lambda_x = 2/(1-\\|x\\|^2)` is the conformal factor.
381
+
382
+ This manifold is the natural geometry for hierarchical and
383
+ tree-structured data — it can embed trees with arbitrarily low
384
+ distortion (Sarkar 2011).
385
+
386
+ Parameters
387
+ ----------
388
+ d : int
389
+ Dimension of the hyperbolic space.
390
+ eps : float
391
+ Numerical safety margin for boundary clamping.
392
+ """
393
+
394
+ def __init__(self, d: int, eps: float = 1e-5):
395
+ self.d = d
396
+ self.eps = eps
397
+ self.point_shape = (d,)
398
+
399
+ def distance(self, x, y):
400
+ """Geodesic distance in the Poincaré ball.
401
+
402
+ .. math::
403
+
404
+ d(x, y) = \\operatorname{arcosh}\\!\\left(1 + \\frac{2\\|x - y\\|^2}
405
+ {(1 - \\|x\\|^2)(1 - \\|y\\|^2)}\\right)
406
+ """
407
+ diff_sq = jnp.sum((x - y) ** 2)
408
+ x_sq = jnp.sum(x ** 2)
409
+ y_sq = jnp.sum(y ** 2)
410
+ denom = (1.0 - x_sq) * (1.0 - y_sq)
411
+ arg = 1.0 + 2.0 * diff_sq / (denom + self.eps)
412
+ # Stable arcosh with safe gradient: arcosh(z) = log(z + sqrt(z^2-1))
413
+ # Clamp arg >= 1 + eps to prevent gradient blow-up near z=1
414
+ arg = jnp.maximum(arg, 1.0 + 1e-6)
415
+ return jnp.arccosh(arg)
416
+
417
+ def distance_squared(self, x, y):
418
+ """Squared geodesic distance."""
419
+ d = self.distance(x, y)
420
+ return d ** 2
421
+
422
+ def log_map(self, x, y):
423
+ """Logarithmic map: tangent vector at x pointing toward y.
424
+
425
+ .. math::
426
+
427
+ \\text{Log}_x(y) = \\frac{2}{\\lambda_x}
428
+ \\operatorname{arctanh}(\\|-x \\oplus y\\|)
429
+ \\frac{-x \\oplus y}{\\|-x \\oplus y\\|}
430
+ """
431
+ neg_x = -x
432
+ add_result = _mobius_add(neg_x, y, eps=self.eps)
433
+ norm = jnp.linalg.norm(add_result)
434
+ norm_safe = jnp.maximum(norm, 1e-10)
435
+ # arctanh is only defined for |z| < 1, clamp
436
+ norm_clamped = jnp.minimum(norm, 1.0 - 1e-7)
437
+ lam_x = _conformal_factor(x, eps=self.eps)
438
+ scale = (2.0 / lam_x) * jnp.arctanh(norm_clamped)
439
+ return scale * add_result / norm_safe
440
+
441
+ def exp_map(self, x, v):
442
+ """Exponential map: move from x along tangent vector v.
443
+
444
+ .. math::
445
+
446
+ \\text{Exp}_x(v) = x \\oplus \\left(
447
+ \\tanh\\!\\left(\\frac{\\lambda_x \\|v\\|}{2}\\right)
448
+ \\frac{v}{\\|v\\|}\\right)
449
+ """
450
+ v_norm = jnp.linalg.norm(v)
451
+ v_norm_safe = jnp.maximum(v_norm, 1e-10)
452
+ lam_x = _conformal_factor(x, eps=self.eps)
453
+ direction = v / v_norm_safe
454
+ t = jnp.tanh(lam_x * v_norm / 2.0) * direction
455
+ return _mobius_add(x, t, eps=self.eps)
456
+
457
+ def random_point(self, key):
458
+ """Generate a random point uniformly in the Poincaré ball.
459
+
460
+ Uses the radial transform: sample direction uniformly on S^{d-1},
461
+ then radius r ~ Beta(1, d) to get uniform distribution in B^d,
462
+ scaled to stay safely inside the ball.
463
+ """
464
+ key1, key2 = jax.random.split(key)
465
+ # Random direction on unit sphere
466
+ direction = jax.random.normal(key1, (self.d,))
467
+ direction = direction / (jnp.linalg.norm(direction) + 1e-10)
468
+ # Uniform radius in ball: r^d ~ U[0,1] => r = U^{1/d}
469
+ u = jax.random.uniform(key2, (), minval=0.01, maxval=0.95)
470
+ r = u ** (1.0 / self.d)
471
+ # Scale to stay inside ball with margin
472
+ r = r * (1.0 - self.eps)
473
+ return r * direction
474
+
475
+ def belongs(self, x):
476
+ """Check if x is in the Poincaré ball: :math:`\\|x\\| < 1`."""
477
+ return jnp.sum(x ** 2) < 1.0
478
+
479
+ def project(self, x):
480
+ """Project to the interior of the Poincaré ball.
481
+
482
+ Clamps the norm to be strictly less than 1.
483
+ """
484
+ max_norm = 1.0 - self.eps
485
+ norm = jnp.linalg.norm(x)
486
+ scale = jnp.minimum(max_norm / (norm + 1e-10), 1.0)
487
+ return x * scale
488
+
489
+ def injectivity_radius(self, x):
490
+ """Hyperbolic space has infinite injectivity radius."""
491
+ return jnp.inf
@@ -32,8 +32,10 @@ try:
32
32
  # Supervised prototype models (LVQ family)
33
33
  from .glvq import GLVQ, GLVQ1, GLVQ21
34
34
  from .relevance_lvq import GRLVQ
35
+ from .sng import SNG
35
36
  from .srng import SRNG
36
37
  from .smng import SMNG
38
+ from .scmng import SCMNG
37
39
  from .slng import SLNG
38
40
  from .stng import STNG
39
41
  from .matrix_lvq import GMLVQ
@@ -117,6 +119,12 @@ try:
117
119
  from .dk_kohonen_som import DKKohonenSOM
118
120
  from .dk_heskes_som import DKHeskesSOM
119
121
 
122
+ # Wasserstein LVQ models
123
+ from .wasserstein_glvq import WGLVQ
124
+ from .wasserstein_gmlvq import WGMLVQ
125
+ from .wasserstein_grlvq import WGRLVQ
126
+
127
+
120
128
  # One-Class Differentiating Kernel models
121
129
  from .oc_dk_glvq import OCDKGLVQ
122
130
  from .oc_dk_relevance_lvq import OCDKGRLVQ
@@ -133,7 +141,7 @@ try:
133
141
  'KPFCM': KPFCM, 'KIPCM': KIPCM, 'KIPCM2': KIPCM2,
134
142
  # Supervised LVQ
135
143
  'GLVQ': GLVQ, 'GLVQ1': GLVQ1, 'GLVQ21': GLVQ21,
136
- 'GRLVQ': GRLVQ, 'SRNG': SRNG, 'SMNG': SMNG, 'SLNG': SLNG, 'STNG': STNG,
144
+ 'GRLVQ': GRLVQ, 'SNG': SNG, 'SRNG': SRNG, 'SMNG': SMNG, 'SCMNG': SCMNG, 'SLNG': SLNG, 'STNG': STNG,
137
145
  'GMLVQ': GMLVQ, 'LGMLVQ': LGMLVQ,
138
146
  'GTLVQ': GTLVQ, 'CELVQ': CELVQ, 'CELVQ_NG': CELVQ_NG,
139
147
  'MCELVQ_NG': MCELVQ_NG, 'LCELVQ_NG': LCELVQ_NG, 'TCELVQ_NG': TCELVQ_NG,
@@ -191,6 +199,8 @@ try:
191
199
  'OCDKGMLVQ': OCDKGMLVQ,
192
200
  'OCDKGLVQ_NG': OCDKGLVQ_NG, 'OCDKGRLVQ_NG': OCDKGRLVQ_NG,
193
201
  'OCDKGMLVQ_NG': OCDKGMLVQ_NG,
202
+ # Wasserstein LVQ
203
+ 'WGLVQ': WGLVQ, 'WGMLVQ': WGMLVQ, 'WGRLVQ': WGRLVQ,
194
204
  }
195
205
 
196
206
  JAX_AVAILABLE = True
@@ -214,7 +224,7 @@ __all__ = [
214
224
  'PCM', 'PFCM', 'SOM',
215
225
  # Supervised LVQ family
216
226
  'GLVQ', 'GLVQ1', 'GLVQ21',
217
- 'GRLVQ', 'SRNG', 'SMNG', 'SLNG', 'STNG',
227
+ 'GRLVQ', 'SNG', 'SRNG', 'SMNG', 'SCMNG', 'SLNG', 'STNG',
218
228
  'GMLVQ', 'LGMLVQ', 'GTLVQ',
219
229
  'CELVQ', 'CELVQ_NG', 'MCELVQ_NG', 'LCELVQ_NG', 'TCELVQ_NG',
220
230
  'LVQ1', 'LVQ21', 'MedianLVQ',
@@ -250,6 +260,8 @@ __all__ = [
250
260
  # One-Class Differentiating Kernel
251
261
  'OCDKGLVQ', 'OCDKGRLVQ', 'OCDKGMLVQ',
252
262
  'OCDKGLVQ_NG', 'OCDKGRLVQ_NG', 'OCDKGMLVQ_NG',
263
+ # Wasserstein LVQ
264
+ 'WGLVQ', 'WGMLVQ', 'WGRLVQ',
253
265
  # Status
254
266
  'JAX_AVAILABLE',
255
267
  ]
@@ -35,6 +35,9 @@ class CELVQNGMixin:
35
35
  gamma_decay : float, optional
36
36
  Per-step multiplicative decay for gamma.
37
37
  Default: computed from max_iter so gamma reaches gamma_final.
38
+ lr_ratio : float
39
+ Ratio of wrong-class to correct-class learning rate (ε⁻/ε⁺).
40
+ Default: 0.5.
38
41
  n_prototypes_per_class : int
39
42
  Number of prototypes per class.
40
43
  max_iter : int
@@ -157,7 +160,7 @@ class CELVQNGMixin:
157
160
  params['gamma'] = jnp.array(gamma, dtype=jnp.float32)
158
161
  return params
159
162
 
160
- def _compute_gamma_init(self):
163
+ def _compute_gamma_init(self, X):
161
164
  """Compute gamma_init from prototype count if not set."""
162
165
  if isinstance(self.n_prototypes_per_class, int):
163
166
  max_per_class = self.n_prototypes_per_class
@@ -173,7 +176,12 @@ class CELVQNGMixin:
173
176
  if self.gamma_decay is not None:
174
177
  self._gamma_decay = self.gamma_decay
175
178
  else:
176
- self._gamma_decay = (self.gamma_final / gamma_init) ** (1.0 / self.max_iter)
179
+ if self.batch_size is not None:
180
+ steps_per_epoch = (X.shape[0] + self.batch_size - 1) // self.batch_size
181
+ else:
182
+ steps_per_epoch = 1
183
+ total_steps = self.max_iter * steps_per_epoch
184
+ self._gamma_decay = (self.gamma_final / gamma_init) ** (1.0 / total_steps)
177
185
 
178
186
  return gamma_init
179
187
 
@@ -183,7 +191,7 @@ class CELVQNGMixin:
183
191
  X, y, self.n_prototypes_per_class, key1
184
192
  )
185
193
 
186
- gamma_init = self._compute_gamma_init()
194
+ gamma_init = self._compute_gamma_init(X)
187
195
 
188
196
  params = {
189
197
  'prototypes': prototypes,
@@ -72,6 +72,9 @@ class DKGLVQ_NG(SupervisedPrototypeModel):
72
72
  gamma_decay : float, optional
73
73
  Per-step multiplicative decay for gamma.
74
74
  Default: computed from max_iter so gamma reaches gamma_final.
75
+ lr_ratio : float
76
+ Ratio of wrong-class to correct-class learning rate (ε⁻/ε⁺).
77
+ Default: 0.5.
75
78
  n_prototypes_per_class : int
76
79
  Number of prototypes per class.
77
80
  max_iter : int
@@ -140,7 +143,7 @@ class DKGLVQ_NG(SupervisedPrototypeModel):
140
143
 
141
144
  def __init__(self, sigma_init='median', sigma_min=1e-3, beta=10.0,
142
145
  gamma_init=None, gamma_final=0.01, gamma_decay=None,
143
- n_prototypes_per_class=1, max_iter=100,
146
+ lr_ratio=0.5, n_prototypes_per_class=1, max_iter=100,
144
147
  lr=0.01, epsilon=1e-6, random_seed=42, distance_fn=None,
145
148
  optimizer='adam', transfer_fn=None, margin=0.0,
146
149
  callbacks=None, use_scan=True, batch_size=None,
@@ -171,6 +174,7 @@ class DKGLVQ_NG(SupervisedPrototypeModel):
171
174
  self.gamma_init = gamma_init
172
175
  self.gamma_final = gamma_final
173
176
  self.gamma_decay = gamma_decay
177
+ self.lr_ratio = lr_ratio
174
178
  self.sigmas_ = None
175
179
  self.gamma_ = None
176
180
 
@@ -227,7 +231,12 @@ class DKGLVQ_NG(SupervisedPrototypeModel):
227
231
  if self.gamma_decay is not None:
228
232
  self._gamma_decay = self.gamma_decay
229
233
  else:
230
- self._gamma_decay = (self.gamma_final / gamma_init) ** (1.0 / self.max_iter)
234
+ if self.batch_size is not None:
235
+ steps_per_epoch = (X.shape[0] + self.batch_size - 1) // self.batch_size
236
+ else:
237
+ steps_per_epoch = 1
238
+ total_steps = self.max_iter * steps_per_epoch
239
+ self._gamma_decay = (self.gamma_final / gamma_init) ** (1.0 / total_steps)
231
240
 
232
241
  params = {
233
242
  'prototypes': prototypes,
@@ -274,6 +283,11 @@ class DKGLVQ_NG(SupervisedPrototypeModel):
274
283
  d_diff = jnp.where(~same_class, distances, INF)
275
284
  dm = jnp.min(d_diff, axis=1)
276
285
 
286
+ # Separate learning rates (Hammer et al. 2003: ε⁻ = lr_ratio × ε⁺)
287
+ # Scale gradient through dm by lr_ratio; forward pass unchanged.
288
+ dm = jax.lax.stop_gradient(dm) + self.lr_ratio * (
289
+ dm - jax.lax.stop_gradient(dm))
290
+
277
291
  # GLVQ mu for each (sample, prototype) pair
278
292
  mu = (distances - dm[:, None]) / (distances + dm[:, None] + 1e-10)
279
293
 
@@ -353,4 +367,5 @@ class DKGLVQ_NG(SupervisedPrototypeModel):
353
367
  hp['gamma_init'] = self.gamma_init
354
368
  hp['gamma_final'] = self.gamma_final
355
369
  hp['gamma_decay'] = self.gamma_decay
370
+ hp['lr_ratio'] = self.lr_ratio
356
371
  return hp
@@ -68,6 +68,9 @@ class DKGMLVQ_NG(SupervisedPrototypeModel):
68
68
  Final neighborhood range. Default: 0.01.
69
69
  gamma_decay : float, optional
70
70
  Per-step multiplicative decay for gamma.
71
+ lr_ratio : float
72
+ Ratio of wrong-class to correct-class learning rate (ε⁻/ε⁺).
73
+ Default: 0.5.
71
74
  n_prototypes_per_class : int
72
75
  Number of prototypes per class.
73
76
  max_iter : int
@@ -136,7 +139,7 @@ class DKGMLVQ_NG(SupervisedPrototypeModel):
136
139
 
137
140
  def __init__(self, latent_dim=None, omega_hat_scale=0.1, beta=10.0,
138
141
  gamma_init=None, gamma_final=0.01, gamma_decay=None,
139
- n_prototypes_per_class=1, max_iter=100,
142
+ lr_ratio=0.5, n_prototypes_per_class=1, max_iter=100,
140
143
  lr=0.01, epsilon=1e-6, random_seed=42, distance_fn=None,
141
144
  optimizer='adam', transfer_fn=None, margin=0.0,
142
145
  callbacks=None, use_scan=True, batch_size=None,
@@ -167,6 +170,7 @@ class DKGMLVQ_NG(SupervisedPrototypeModel):
167
170
  self.gamma_init = gamma_init
168
171
  self.gamma_final = gamma_final
169
172
  self.gamma_decay = gamma_decay
173
+ self.lr_ratio = lr_ratio
170
174
  self.omega_hat_ = None
171
175
  self.gamma_ = None
172
176
 
@@ -210,7 +214,12 @@ class DKGMLVQ_NG(SupervisedPrototypeModel):
210
214
  if self.gamma_decay is not None:
211
215
  self._gamma_decay = self.gamma_decay
212
216
  else:
213
- self._gamma_decay = (self.gamma_final / gamma_init) ** (1.0 / self.max_iter)
217
+ if self.batch_size is not None:
218
+ steps_per_epoch = (X.shape[0] + self.batch_size - 1) // self.batch_size
219
+ else:
220
+ steps_per_epoch = 1
221
+ total_steps = self.max_iter * steps_per_epoch
222
+ self._gamma_decay = (self.gamma_final / gamma_init) ** (1.0 / total_steps)
214
223
 
215
224
  params = {
216
225
  'prototypes': prototypes,
@@ -255,6 +264,11 @@ class DKGMLVQ_NG(SupervisedPrototypeModel):
255
264
  d_diff = jnp.where(~same_class, distances, INF)
256
265
  dm = jnp.min(d_diff, axis=1)
257
266
 
267
+ # Separate learning rates (Hammer et al. 2003: ε⁻ = lr_ratio × ε⁺)
268
+ # Scale gradient through dm by lr_ratio; forward pass unchanged.
269
+ dm = jax.lax.stop_gradient(dm) + self.lr_ratio * (
270
+ dm - jax.lax.stop_gradient(dm))
271
+
258
272
  mu = (distances - dm[:, None]) / (distances + dm[:, None] + 1e-10)
259
273
 
260
274
  from prosemble.core.activations import sigmoid_beta
@@ -331,6 +345,7 @@ class DKGMLVQ_NG(SupervisedPrototypeModel):
331
345
  hp['gamma_init'] = self.gamma_init
332
346
  hp['gamma_final'] = self.gamma_final
333
347
  hp['gamma_decay'] = self.gamma_decay
348
+ hp['lr_ratio'] = self.lr_ratio
334
349
  if self.latent_dim is not None:
335
350
  hp['latent_dim'] = self.latent_dim
336
351
  return hp
@@ -74,6 +74,9 @@ class DKGRLVQ_NG(SupervisedPrototypeModel):
74
74
  Final neighborhood range. Default: 0.01.
75
75
  gamma_decay : float, optional
76
76
  Per-step multiplicative decay for gamma.
77
+ lr_ratio : float
78
+ Ratio of wrong-class to correct-class learning rate (ε⁻/ε⁺).
79
+ Default: 0.5.
77
80
  n_prototypes_per_class : int
78
81
  Number of prototypes per class.
79
82
  max_iter : int
@@ -144,7 +147,7 @@ class DKGRLVQ_NG(SupervisedPrototypeModel):
144
147
 
145
148
  def __init__(self, sigma_init='median', sigma_min=1e-3, beta=10.0,
146
149
  gamma_init=None, gamma_final=0.01, gamma_decay=None,
147
- n_prototypes_per_class=1, max_iter=100,
150
+ lr_ratio=0.5, n_prototypes_per_class=1, max_iter=100,
148
151
  lr=0.01, epsilon=1e-6, random_seed=42, distance_fn=None,
149
152
  optimizer='adam', transfer_fn=None, margin=0.0,
150
153
  callbacks=None, use_scan=True, batch_size=None,
@@ -175,6 +178,7 @@ class DKGRLVQ_NG(SupervisedPrototypeModel):
175
178
  self.gamma_init = gamma_init
176
179
  self.gamma_final = gamma_final
177
180
  self.gamma_decay = gamma_decay
181
+ self.lr_ratio = lr_ratio
178
182
  self.sigmas_ = None
179
183
  self.relevances_ = None
180
184
  self.gamma_ = None
@@ -234,7 +238,12 @@ class DKGRLVQ_NG(SupervisedPrototypeModel):
234
238
  if self.gamma_decay is not None:
235
239
  self._gamma_decay = self.gamma_decay
236
240
  else:
237
- self._gamma_decay = (self.gamma_final / gamma_init) ** (1.0 / self.max_iter)
241
+ if self.batch_size is not None:
242
+ steps_per_epoch = (X.shape[0] + self.batch_size - 1) // self.batch_size
243
+ else:
244
+ steps_per_epoch = 1
245
+ total_steps = self.max_iter * steps_per_epoch
246
+ self._gamma_decay = (self.gamma_final / gamma_init) ** (1.0 / total_steps)
238
247
 
239
248
  relevances = jnp.ones(n_features) / n_features
240
249
  params = {
@@ -282,6 +291,11 @@ class DKGRLVQ_NG(SupervisedPrototypeModel):
282
291
  d_diff = jnp.where(~same_class, distances, INF)
283
292
  dm = jnp.min(d_diff, axis=1)
284
293
 
294
+ # Separate learning rates (Hammer et al. 2003: ε⁻ = lr_ratio × ε⁺)
295
+ # Scale gradient through dm by lr_ratio; forward pass unchanged.
296
+ dm = jax.lax.stop_gradient(dm) + self.lr_ratio * (
297
+ dm - jax.lax.stop_gradient(dm))
298
+
285
299
  mu = (distances - dm[:, None]) / (distances + dm[:, None] + 1e-10)
286
300
 
287
301
  from prosemble.core.activations import sigmoid_beta
@@ -373,4 +387,5 @@ class DKGRLVQ_NG(SupervisedPrototypeModel):
373
387
  hp['gamma_init'] = self.gamma_init
374
388
  hp['gamma_final'] = self.gamma_final
375
389
  hp['gamma_decay'] = self.gamma_decay
390
+ hp['lr_ratio'] = self.lr_ratio
376
391
  return hp