prosemble 2.2.0__tar.gz → 2.3.0__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 (318) hide show
  1. {prosemble-2.2.0 → prosemble-2.3.0}/PKG-INFO +1 -1
  2. {prosemble-2.2.0 → prosemble-2.3.0}/nix/package.nix +9 -8
  3. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/__init__.py +25 -0
  4. prosemble-2.3.0/prosemble/core/curriculum.py +171 -0
  5. prosemble-2.3.0/prosemble/core/geodesic.py +217 -0
  6. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/onnx_export.py +146 -3
  7. prosemble-2.3.0/prosemble/core/optimizers.py +292 -0
  8. prosemble-2.3.0/prosemble/core/regularization.py +210 -0
  9. prosemble-2.3.0/prosemble/core/reject.py +246 -0
  10. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/__init__.py +33 -0
  11. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_glvq.py +5 -0
  12. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_glvq_ng.py +5 -0
  13. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_matrix_lvq.py +4 -0
  14. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_matrix_lvq_ng.py +4 -0
  15. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_relevance_lvq.py +6 -0
  16. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_relevance_lvq_ng.py +6 -0
  17. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/local_matrix_lvq.py +6 -0
  18. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/matrix_lvq.py +6 -0
  19. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/prototype_base.py +2 -1
  20. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/relevance_lvq.py +5 -0
  21. prosemble-2.3.0/prosemble/models/riemannian_dk_glvq.py +311 -0
  22. prosemble-2.3.0/prosemble/models/riemannian_dk_m_slng.py +303 -0
  23. prosemble-2.3.0/prosemble/models/riemannian_dk_m_smng.py +303 -0
  24. prosemble-2.3.0/prosemble/models/riemannian_dk_m_stng.py +307 -0
  25. prosemble-2.3.0/prosemble/models/riemannian_dk_matrix_lvq.py +343 -0
  26. prosemble-2.3.0/prosemble/models/riemannian_dk_r_slng.py +352 -0
  27. prosemble-2.3.0/prosemble/models/riemannian_dk_r_smng.py +353 -0
  28. prosemble-2.3.0/prosemble/models/riemannian_dk_r_stng.py +359 -0
  29. prosemble-2.3.0/prosemble/models/riemannian_dk_relevance_lvq.py +380 -0
  30. prosemble-2.3.0/prosemble/models/riemannian_dk_slng.py +319 -0
  31. prosemble-2.3.0/prosemble/models/riemannian_dk_smng.py +319 -0
  32. prosemble-2.3.0/prosemble/models/riemannian_dk_stng.py +328 -0
  33. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/tangent_lvq.py +8 -0
  34. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble.egg-info/PKG-INFO +1 -1
  35. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble.egg-info/SOURCES.txt +19 -0
  36. {prosemble-2.2.0 → prosemble-2.3.0}/pyproject.toml +1 -1
  37. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/api/core.rst +35 -0
  38. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/api/models.rst +51 -0
  39. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/conf.py +1 -1
  40. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/guides/advanced.rst +262 -0
  41. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/guides/differentiating_kernels.rst +74 -0
  42. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/guides/onnx.rst +52 -1
  43. prosemble-2.3.0/tests/test_improvements.py +703 -0
  44. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_onnx_export.py +90 -0
  45. prosemble-2.3.0/tests/test_riemannian_dk_models.py +1043 -0
  46. {prosemble-2.2.0 → prosemble-2.3.0}/uv.lock +1 -1
  47. {prosemble-2.2.0 → prosemble-2.3.0}/.bumpversion.cfg +0 -0
  48. {prosemble-2.2.0 → prosemble-2.3.0}/.editorconfig +0 -0
  49. {prosemble-2.2.0 → prosemble-2.3.0}/.envrc +0 -0
  50. {prosemble-2.2.0 → prosemble-2.3.0}/.github/ISSUE_TEMPLATE/bug_report.md +0 -0
  51. {prosemble-2.2.0 → prosemble-2.3.0}/.github/ISSUE_TEMPLATE/config.yml +0 -0
  52. {prosemble-2.2.0 → prosemble-2.3.0}/.github/ISSUE_TEMPLATE/feature_request.md +0 -0
  53. {prosemble-2.2.0 → prosemble-2.3.0}/.github/workflows/ci.yml +0 -0
  54. {prosemble-2.2.0 → prosemble-2.3.0}/.github/workflows/release.yml +0 -0
  55. {prosemble-2.2.0 → prosemble-2.3.0}/.gitignore +0 -0
  56. {prosemble-2.2.0 → prosemble-2.3.0}/.readthedocs.yaml +0 -0
  57. {prosemble-2.2.0 → prosemble-2.3.0}/CONSTITUTION.md +0 -0
  58. {prosemble-2.2.0 → prosemble-2.3.0}/LICENSE +0 -0
  59. {prosemble-2.2.0 → prosemble-2.3.0}/MANIFEST.in +0 -0
  60. {prosemble-2.2.0 → prosemble-2.3.0}/ONNX_COVERAGE.md +0 -0
  61. {prosemble-2.2.0 → prosemble-2.3.0}/README.md +0 -0
  62. {prosemble-2.2.0 → prosemble-2.3.0}/TASK.md +0 -0
  63. {prosemble-2.2.0 → prosemble-2.3.0}/benchmarks/benchmark_distance_jax.py +0 -0
  64. {prosemble-2.2.0 → prosemble-2.3.0}/benchmarks/benchmark_fcm_jax.py +0 -0
  65. {prosemble-2.2.0 → prosemble-2.3.0}/benchmarks/benchmark_pcm_jax.py +0 -0
  66. {prosemble-2.2.0 → prosemble-2.3.0}/devenv.lock +0 -0
  67. {prosemble-2.2.0 → prosemble-2.3.0}/devenv.nix +0 -0
  68. {prosemble-2.2.0 → prosemble-2.3.0}/devenv.yaml +0 -0
  69. {prosemble-2.2.0 → prosemble-2.3.0}/examples/afcm_iris.py +0 -0
  70. {prosemble-2.2.0 → prosemble-2.3.0}/examples/bgpc_iris.py +0 -0
  71. {prosemble-2.2.0 → prosemble-2.3.0}/examples/cbc_iris.py +0 -0
  72. {prosemble-2.2.0 → prosemble-2.3.0}/examples/celvq_ng_iris.py +0 -0
  73. {prosemble-2.2.0 → prosemble-2.3.0}/examples/fcm_iris.py +0 -0
  74. {prosemble-2.2.0 → prosemble-2.3.0}/examples/fpcm_iris.py +0 -0
  75. {prosemble-2.2.0 → prosemble-2.3.0}/examples/glvq_advanced_iris.py +0 -0
  76. {prosemble-2.2.0 → prosemble-2.3.0}/examples/glvq_iris.py +0 -0
  77. {prosemble-2.2.0 → prosemble-2.3.0}/examples/glvq_unequal_prototypes_iris.py +0 -0
  78. {prosemble-2.2.0 → prosemble-2.3.0}/examples/gmlvq_iris.py +0 -0
  79. {prosemble-2.2.0 → prosemble-2.3.0}/examples/grlvq_iris.py +0 -0
  80. {prosemble-2.2.0 → prosemble-2.3.0}/examples/hcm_iris.py +0 -0
  81. {prosemble-2.2.0 → prosemble-2.3.0}/examples/heskes_som_iris.py +0 -0
  82. {prosemble-2.2.0 → prosemble-2.3.0}/examples/ipcm2_iris.py +0 -0
  83. {prosemble-2.2.0 → prosemble-2.3.0}/examples/ipcm_iris.py +0 -0
  84. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kafcm_iris.py +0 -0
  85. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kfcm_iris.py +0 -0
  86. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kfpcm_iris.py +0 -0
  87. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kipcm2_iris.py +0 -0
  88. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kipcm_iris.py +0 -0
  89. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kmeans_iris.py +0 -0
  90. {prosemble-2.2.0 → prosemble-2.3.0}/examples/knn_iris.py +0 -0
  91. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kohonen_som_iris.py +0 -0
  92. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kpcm_iris.py +0 -0
  93. {prosemble-2.2.0 → prosemble-2.3.0}/examples/kpfcm_iris.py +0 -0
  94. {prosemble-2.2.0 → prosemble-2.3.0}/examples/large_dataset_training.py +0 -0
  95. {prosemble-2.2.0 → prosemble-2.3.0}/examples/lcelvq_ng_iris.py +0 -0
  96. {prosemble-2.2.0 → prosemble-2.3.0}/examples/lmrslvq_iris.py +0 -0
  97. {prosemble-2.2.0 → prosemble-2.3.0}/examples/lmrslvq_ng_iris.py +0 -0
  98. {prosemble-2.2.0 → prosemble-2.3.0}/examples/lvq1_iris.py +0 -0
  99. {prosemble-2.2.0 → prosemble-2.3.0}/examples/lvqmln_iris.py +0 -0
  100. {prosemble-2.2.0 → prosemble-2.3.0}/examples/mcelvq_ng_iris.py +0 -0
  101. {prosemble-2.2.0 → prosemble-2.3.0}/examples/mrslvq_iris.py +0 -0
  102. {prosemble-2.2.0 → prosemble-2.3.0}/examples/mrslvq_ng_iris.py +0 -0
  103. {prosemble-2.2.0 → prosemble-2.3.0}/examples/neural_gas_iris.py +0 -0
  104. {prosemble-2.2.0 → prosemble-2.3.0}/examples/npc_iris.py +0 -0
  105. {prosemble-2.2.0 → prosemble-2.3.0}/examples/oc_lmrslvq_iris.py +0 -0
  106. {prosemble-2.2.0 → prosemble-2.3.0}/examples/oc_lmrslvq_ng_iris.py +0 -0
  107. {prosemble-2.2.0 → prosemble-2.3.0}/examples/oc_mrslvq_iris.py +0 -0
  108. {prosemble-2.2.0 → prosemble-2.3.0}/examples/oc_mrslvq_ng_iris.py +0 -0
  109. {prosemble-2.2.0 → prosemble-2.3.0}/examples/oc_rslvq_iris.py +0 -0
  110. {prosemble-2.2.0 → prosemble-2.3.0}/examples/oc_rslvq_ng_iris.py +0 -0
  111. {prosemble-2.2.0 → prosemble-2.3.0}/examples/pcm_iris.py +0 -0
  112. {prosemble-2.2.0 → prosemble-2.3.0}/examples/pfcm_iris.py +0 -0
  113. {prosemble-2.2.0 → prosemble-2.3.0}/examples/plvq_iris.py +0 -0
  114. {prosemble-2.2.0 → prosemble-2.3.0}/examples/rslvq_ng_iris.py +0 -0
  115. {prosemble-2.2.0 → prosemble-2.3.0}/examples/slng_iris.py +0 -0
  116. {prosemble-2.2.0 → prosemble-2.3.0}/examples/smng_iris.py +0 -0
  117. {prosemble-2.2.0 → prosemble-2.3.0}/examples/som_iris.py +0 -0
  118. {prosemble-2.2.0 → prosemble-2.3.0}/examples/srng_iris.py +0 -0
  119. {prosemble-2.2.0 → prosemble-2.3.0}/examples/stng_iris.py +0 -0
  120. {prosemble-2.2.0 → prosemble-2.3.0}/examples/tcelvq_ng_iris.py +0 -0
  121. {prosemble-2.2.0 → prosemble-2.3.0}/examples/vis_fcm_iris.py +0 -0
  122. {prosemble-2.2.0 → prosemble-2.3.0}/examples/vis_glvq_iris.py +0 -0
  123. {prosemble-2.2.0 → prosemble-2.3.0}/examples/vis_neural_gas_iris.py +0 -0
  124. {prosemble-2.2.0 → prosemble-2.3.0}/examples/vis_som_iris.py +0 -0
  125. {prosemble-2.2.0 → prosemble-2.3.0}/flake.nix +0 -0
  126. {prosemble-2.2.0 → prosemble-2.3.0}/mkdocs.yml +0 -0
  127. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/__init__.py +0 -0
  128. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/activations.py +0 -0
  129. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/callbacks.py +0 -0
  130. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/competitions.py +0 -0
  131. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/data.py +0 -0
  132. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/distance.py +0 -0
  133. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/distributed.py +0 -0
  134. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/initializers.py +0 -0
  135. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/kernel.py +0 -0
  136. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/losses.py +0 -0
  137. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/manifolds.py +0 -0
  138. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/model_selection.py +0 -0
  139. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/pipeline.py +0 -0
  140. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/pooling.py +0 -0
  141. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/protocols.py +0 -0
  142. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/quantization.py +0 -0
  143. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/serialization.py +0 -0
  144. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/similarities.py +0 -0
  145. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/utils.py +0 -0
  146. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/vis.py +0 -0
  147. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/core/visualizer.py +0 -0
  148. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/datasets/__init__.py +0 -0
  149. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/datasets/dataset.py +0 -0
  150. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/afcm.py +0 -0
  151. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/base.py +0 -0
  152. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/bgpc.py +0 -0
  153. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/cbc.py +0 -0
  154. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/celvq_ng.py +0 -0
  155. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/celvq_ng_mixin.py +0 -0
  156. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/crossentropy_lvq.py +0 -0
  157. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_heskes_som.py +0 -0
  158. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_kohonen_som.py +0 -0
  159. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/dk_neural_gas.py +0 -0
  160. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/fcm.py +0 -0
  161. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/fpcm.py +0 -0
  162. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/glvq.py +0 -0
  163. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/growing_neural_gas.py +0 -0
  164. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/hcm.py +0 -0
  165. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/heskes_som.py +0 -0
  166. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/image_cbc.py +0 -0
  167. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/image_lvq.py +0 -0
  168. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/ipcm.py +0 -0
  169. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/ipcm2.py +0 -0
  170. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kafcm.py +0 -0
  171. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kfcm.py +0 -0
  172. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kfpcm.py +0 -0
  173. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kipcm.py +0 -0
  174. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kipcm2.py +0 -0
  175. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kmeans.py +0 -0
  176. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/knn.py +0 -0
  177. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kohonen_som.py +0 -0
  178. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kpcm.py +0 -0
  179. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/kpfcm.py +0 -0
  180. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/lcelvq_ng.py +0 -0
  181. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/lvq1.py +0 -0
  182. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/lvq21.py +0 -0
  183. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/lvqmln.py +0 -0
  184. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/matrix_rslvq.py +0 -0
  185. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/mcelvq_ng.py +0 -0
  186. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/median_lvq.py +0 -0
  187. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/mrslvq_ng.py +0 -0
  188. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/neural_gas.py +0 -0
  189. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/npc.py +0 -0
  190. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_dk_glvq.py +0 -0
  191. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_dk_glvq_ng.py +0 -0
  192. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_dk_matrix_lvq.py +0 -0
  193. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_dk_matrix_lvq_ng.py +0 -0
  194. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_dk_relevance_lvq.py +0 -0
  195. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_dk_relevance_lvq_ng.py +0 -0
  196. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_glvq.py +0 -0
  197. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_glvq_ng.py +0 -0
  198. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_glvq_ng_mixin.py +0 -0
  199. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_gmlvq.py +0 -0
  200. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_gmlvq_ng.py +0 -0
  201. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_grlvq.py +0 -0
  202. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_grlvq_ng.py +0 -0
  203. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_gtlvq.py +0 -0
  204. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_gtlvq_ng.py +0 -0
  205. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_lgmlvq.py +0 -0
  206. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_lgmlvq_ng.py +0 -0
  207. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_mrslvq.py +0 -0
  208. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_mrslvq_ng.py +0 -0
  209. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_rslvq.py +0 -0
  210. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_rslvq_ng.py +0 -0
  211. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/oc_rslvq_ng_mixin.py +0 -0
  212. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/pcm.py +0 -0
  213. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/pfcm.py +0 -0
  214. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/plvq.py +0 -0
  215. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/probabilistic_lvq.py +0 -0
  216. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/riemannian_neural_gas.py +0 -0
  217. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/riemannian_slng.py +0 -0
  218. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/riemannian_smng.py +0 -0
  219. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/riemannian_srng.py +0 -0
  220. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/riemannian_stng.py +0 -0
  221. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/rslvq_ng.py +0 -0
  222. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/siamese_lvq.py +0 -0
  223. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/slng.py +0 -0
  224. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/smng.py +0 -0
  225. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/som.py +0 -0
  226. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/srng.py +0 -0
  227. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/stng.py +0 -0
  228. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/svq_occ.py +0 -0
  229. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/svq_occ_lm.py +0 -0
  230. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/svq_occ_m.py +0 -0
  231. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/svq_occ_r.py +0 -0
  232. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/svq_occ_t.py +0 -0
  233. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/models/tcelvq_ng.py +0 -0
  234. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble/py.typed +0 -0
  235. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble.egg-info/dependency_links.txt +0 -0
  236. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble.egg-info/requires.txt +0 -0
  237. {prosemble-2.2.0 → prosemble-2.3.0}/prosemble.egg-info/top_level.txt +0 -0
  238. {prosemble-2.2.0 → prosemble-2.3.0}/requirements_docs.txt +0 -0
  239. {prosemble-2.2.0 → prosemble-2.3.0}/setup.cfg +0 -0
  240. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/Makefile +0 -0
  241. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/guides/clustering.rst +0 -0
  242. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/guides/one_class.rst +0 -0
  243. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/guides/riemannian.rst +0 -0
  244. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/guides/supervised.rst +0 -0
  245. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/guides/topology.rst +0 -0
  246. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/index.rst +0 -0
  247. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/installation.rst +0 -0
  248. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/make.bat +0 -0
  249. {prosemble-2.2.0 → prosemble-2.3.0}/sphinx-docs/quickstart.rst +0 -0
  250. {prosemble-2.2.0 → prosemble-2.3.0}/tests/__init__.py +0 -0
  251. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_activations.py +0 -0
  252. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_cbc.py +0 -0
  253. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_celvq_ng.py +0 -0
  254. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_checkpoint.py +0 -0
  255. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_clustering_patience.py +0 -0
  256. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_competitions.py +0 -0
  257. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_core.py +0 -0
  258. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_core_jax.py +0 -0
  259. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_data.py +0 -0
  260. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_dataset.py +0 -0
  261. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_dataset_jax.py +0 -0
  262. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_distance_jax.py +0 -0
  263. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_distributed.py +0 -0
  264. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_dk_models.py +0 -0
  265. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_dk_ng_models.py +0 -0
  266. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_examples.sh +0 -0
  267. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_fcm_jax.py +0 -0
  268. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_glvq.py +0 -0
  269. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_gmlvq.py +0 -0
  270. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_heskes_som.py +0 -0
  271. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_image_lvq.py +0 -0
  272. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_initializers.py +0 -0
  273. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_jax_models.py +0 -0
  274. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_jit_inference.py +0 -0
  275. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_kohonen_som.py +0 -0
  276. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_lcelvq_ng.py +0 -0
  277. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_losses.py +0 -0
  278. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_lvq_nongradient.py +0 -0
  279. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_lvqmln_plvq.py +0 -0
  280. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_matrix_rslvq.py +0 -0
  281. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_mcelvq_ng.py +0 -0
  282. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_minibatch.py +0 -0
  283. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_mixed_precision.py +0 -0
  284. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_model_selection.py +0 -0
  285. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_mrslvq_ng.py +0 -0
  286. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_neural_gas.py +0 -0
  287. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_oc_dk_models.py +0 -0
  288. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_oc_dk_ng_models.py +0 -0
  289. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_oc_glvq.py +0 -0
  290. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_oc_glvq_ng.py +0 -0
  291. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_oc_glvq_variants.py +0 -0
  292. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_oc_mrslvq.py +0 -0
  293. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_oc_rslvq.py +0 -0
  294. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_oc_rslvq_ng.py +0 -0
  295. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_pcm_jax.py +0 -0
  296. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_pfcm_jax.py +0 -0
  297. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_pipeline.py +0 -0
  298. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_pooling.py +0 -0
  299. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_probabilistic.py +0 -0
  300. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_protocols.py +0 -0
  301. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_prototype_base.py +0 -0
  302. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_quantization.py +0 -0
  303. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_riemannian_neural_gas.py +0 -0
  304. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_riemannian_supervised_ng.py +0 -0
  305. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_rslvq_ng.py +0 -0
  306. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_scan_training.py +0 -0
  307. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_serialization.py +0 -0
  308. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_siamese_lvq.py +0 -0
  309. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_similarities.py +0 -0
  310. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_slng.py +0 -0
  311. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_smng.py +0 -0
  312. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_srng.py +0 -0
  313. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_stng.py +0 -0
  314. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_svq_occ.py +0 -0
  315. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_svq_occ_r.py +0 -0
  316. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_svq_occ_variants.py +0 -0
  317. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_tcelvq_ng.py +0 -0
  318. {prosemble-2.2.0 → prosemble-2.3.0}/tests/test_vis.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: prosemble
3
- Version: 2.2.0
3
+ Version: 2.3.0
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
@@ -5,8 +5,8 @@
5
5
  # 2. Add to pkgs/top-level/python-packages.nix:
6
6
  # prosemble = callPackage ../development/python-modules/prosemble { };
7
7
  # 3. Update the hash after publishing to PyPI:
8
- # nix-prefetch-url --unpack https://github.com/naotoo1/prosemble/archive/refs/tags/v2.0.0.tar.gz
9
- # 4. PR title: python3Packages.prosemble: init at 2.0.0
8
+ # nix-prefetch-url --unpack https://github.com/naotoo1/prosemble/archive/refs/tags/v2.3.0.tar.gz
9
+ # 4. PR title: python3Packages.prosemble: init at 2.3.0
10
10
  {
11
11
  lib,
12
12
  buildPythonPackage,
@@ -29,14 +29,14 @@
29
29
 
30
30
  buildPythonPackage (finalAttrs: {
31
31
  pname = "prosemble";
32
- version = "2.0.0";
32
+ version = "2.3.0";
33
33
  pyproject = true;
34
34
 
35
35
  src = fetchFromGitHub {
36
36
  owner = "naotoo1";
37
37
  repo = "prosemble";
38
- rev = "refs/tags/v${finalAttrs.version}";
39
- hash = "sha256-6Oa7V7tFle74oYBNEaJ1ZCrrku8cav92YkzpnCxsH5c=";
38
+ tag = "v${finalAttrs.version}";
39
+ hash = "sha256-296Z1lHBcOVQbkfO7KlNnWe7f7xjbKFHJRfGUPaHKWY=";
40
40
  };
41
41
 
42
42
  build-system = [
@@ -68,6 +68,7 @@ buildPythonPackage (finalAttrs: {
68
68
  ];
69
69
 
70
70
  disabledTestPaths = [
71
+ # Requires geomstats which is not packaged
71
72
  "tests/test_riemannian_neural_gas.py"
72
73
  ];
73
74
 
@@ -77,11 +78,11 @@ buildPythonPackage (finalAttrs: {
77
78
  "prosemble.core"
78
79
  ];
79
80
 
80
- meta = with lib; {
81
+ meta = {
81
82
  description = "JAX-based prototype machine learning library";
82
83
  homepage = "https://github.com/naotoo1/prosemble";
83
84
  changelog = "https://github.com/naotoo1/prosemble/releases/tag/v${finalAttrs.version}";
84
- license = licenses.mit;
85
- maintainers = with maintainers; [ ];
85
+ license = lib.licenses.mit;
86
+ maintainers = with lib.maintainers; [ naotoo1 ];
86
87
  };
87
88
  })
@@ -41,6 +41,19 @@ from .protocols import (
41
41
  from .serialization import SerializationMixin
42
42
  from .data import shuffle_arrays, padded_batches, batched_iterator
43
43
  from .distributed import create_mesh, shard_data, replicate_params
44
+ from .optimizers import per_group_clip, hypergradient_descent, riemannian_nesterov
45
+ from .regularization import (
46
+ prototype_diversity_loss, prototype_diversity_loss_vectorized,
47
+ sparse_relevance_proximal, elastic_net_proximal,
48
+ )
49
+ from .curriculum import (
50
+ curriculum_weights, curriculum_threshold, apply_curriculum_to_loss,
51
+ )
52
+ from .reject import RejectOptionMixin
53
+ from .geodesic import (
54
+ geodesic_interpolation, geodesic_midpoint, decision_boundary_point,
55
+ prototype_geodesic_distances, inter_class_geodesics,
56
+ )
44
57
 
45
58
  try:
46
59
  import jax
@@ -93,6 +106,18 @@ __all__ = _distance_all + [
93
106
  'shuffle_arrays', 'padded_batches', 'batched_iterator',
94
107
  # Distributed training
95
108
  'create_mesh', 'shard_data', 'replicate_params',
109
+ # Custom optimizers
110
+ 'per_group_clip', 'hypergradient_descent', 'riemannian_nesterov',
111
+ # Regularization
112
+ 'prototype_diversity_loss', 'prototype_diversity_loss_vectorized',
113
+ 'sparse_relevance_proximal', 'elastic_net_proximal',
114
+ # Curriculum learning
115
+ 'curriculum_weights', 'curriculum_threshold', 'apply_curriculum_to_loss',
116
+ # Reject option
117
+ 'RejectOptionMixin',
118
+ # Geodesic interpolation
119
+ 'geodesic_interpolation', 'geodesic_midpoint', 'decision_boundary_point',
120
+ 'prototype_geodesic_distances', 'inter_class_geodesics',
96
121
  # ONNX export (optional)
97
122
  'export_onnx',
98
123
  ]
@@ -0,0 +1,171 @@
1
+ """Curriculum learning (self-paced learning) for prototype-based models.
2
+
3
+ Implements the self-paced learning framework (Kumar et al. 2010) adapted
4
+ for LVQ models. Samples are presented in order of difficulty — easy
5
+ samples first, hard samples later.
6
+
7
+ For LVQ, sample difficulty is measured by the absolute value of the
8
+ mu-ratio: |mu(x)| close to 0 means the sample is on the decision
9
+ boundary (hard), while |mu(x)| >> 0 means it's far from the boundary
10
+ (easy).
11
+
12
+ Usage
13
+ -----
14
+ Curriculum learning is applied via sample masking in the loss function.
15
+ At each training step, a difficulty threshold lambda_t determines which
16
+ samples participate. The threshold increases over training to gradually
17
+ include harder samples.
18
+
19
+ References
20
+ ----------
21
+ .. [1] Kumar, M. P., Packer, B., & Koller, D. (2010). Self-paced
22
+ learning for latent variable models. NeurIPS.
23
+ .. [2] Bengio, Y., Louradour, J., Collobert, R., & Weston, J. (2009).
24
+ Curriculum learning. ICML.
25
+ """
26
+
27
+ import jax.numpy as jnp
28
+
29
+
30
+ def curriculum_weights(per_sample_loss, threshold, mode='hard'):
31
+ """Compute per-sample curriculum weights based on loss magnitude.
32
+
33
+ Parameters
34
+ ----------
35
+ per_sample_loss : array of shape (n_samples,)
36
+ Per-sample loss values (e.g., from GLVQ mu-ratio).
37
+ threshold : float
38
+ Difficulty threshold lambda_t. Samples with loss below this
39
+ threshold are included. Increases over training.
40
+ mode : str, {'hard', 'soft', 'linear'}
41
+ Weight assignment mode:
42
+ - 'hard': binary weights (0 or 1). Sample is either in or out.
43
+ - 'soft': smooth exponential weighting.
44
+ w_i = exp(-loss_i / threshold) if loss_i > threshold, else 1.
45
+ - 'linear': linear ramp-down.
46
+ w_i = max(0, 1 - (loss_i - threshold) / threshold).
47
+ Default: 'hard'.
48
+
49
+ Returns
50
+ -------
51
+ weights : array of shape (n_samples,)
52
+ Per-sample weights in [0, 1]. Weights sum is used to normalize
53
+ the loss.
54
+ """
55
+ if mode == 'hard':
56
+ return jnp.where(per_sample_loss <= threshold, 1.0, 0.0)
57
+ elif mode == 'soft':
58
+ # Smooth transition: full weight below threshold, exponential decay above
59
+ excess = jnp.maximum(per_sample_loss - threshold, 0.0)
60
+ return jnp.exp(-excess / (threshold + 1e-10))
61
+ elif mode == 'linear':
62
+ # Linear ramp-down above threshold
63
+ return jnp.clip(1.0 - (per_sample_loss - threshold) / (threshold + 1e-10), 0.0, 1.0)
64
+ else:
65
+ raise ValueError(f"Unknown curriculum mode '{mode}'. Use 'hard', 'soft', or 'linear'.")
66
+
67
+
68
+ def curriculum_threshold(iteration, max_iter, init_threshold=0.3,
69
+ final_threshold=None, schedule='linear'):
70
+ """Compute the difficulty threshold at a given training iteration.
71
+
72
+ The threshold increases over training: initially only easy samples
73
+ are included, and harder samples are gradually added.
74
+
75
+ Parameters
76
+ ----------
77
+ iteration : int or array
78
+ Current training iteration.
79
+ max_iter : int
80
+ Maximum number of iterations.
81
+ init_threshold : float
82
+ Initial threshold (fraction of max loss to include). At the
83
+ start, only samples with loss < init_threshold * max_loss are
84
+ included. Default: 0.3.
85
+ final_threshold : float, optional
86
+ Final threshold. Default: None (uses a large value to include
87
+ all samples by end of training).
88
+ schedule : str, {'linear', 'exponential', 'cosine'}
89
+ How the threshold grows over training:
90
+ - 'linear': lambda_t = init + (final - init) * t / T
91
+ - 'exponential': lambda_t = init * (final / init) ^ (t / T)
92
+ - 'cosine': cosine annealing from init to final.
93
+ Default: 'linear'.
94
+
95
+ Returns
96
+ -------
97
+ threshold : float
98
+ Difficulty threshold at the current iteration.
99
+ """
100
+ if final_threshold is None:
101
+ final_threshold = 100.0 # Effectively include all samples
102
+
103
+ progress = jnp.clip(iteration / jnp.maximum(max_iter - 1, 1), 0.0, 1.0)
104
+
105
+ if schedule == 'linear':
106
+ return init_threshold + (final_threshold - init_threshold) * progress
107
+ elif schedule == 'exponential':
108
+ # Avoid log(0): ensure init > 0
109
+ log_ratio = jnp.log(final_threshold / (init_threshold + 1e-10))
110
+ return init_threshold * jnp.exp(log_ratio * progress)
111
+ elif schedule == 'cosine':
112
+ # Cosine annealing: starts slow, accelerates, then slows
113
+ cosine_factor = 0.5 * (1.0 - jnp.cos(jnp.pi * progress))
114
+ return init_threshold + (final_threshold - init_threshold) * cosine_factor
115
+ else:
116
+ raise ValueError(
117
+ f"Unknown schedule '{schedule}'. Use 'linear', 'exponential', or 'cosine'."
118
+ )
119
+
120
+
121
+ def apply_curriculum_to_loss(per_sample_losses, iteration, max_iter,
122
+ init_percentile=0.3, schedule='linear',
123
+ mode='soft'):
124
+ """Full curriculum pipeline: compute threshold and weighted loss.
125
+
126
+ Convenience function that combines threshold scheduling and weight
127
+ computation. Use inside ``_compute_loss`` to add curriculum learning
128
+ to any model.
129
+
130
+ Parameters
131
+ ----------
132
+ per_sample_losses : array of shape (n_samples,)
133
+ Per-sample loss values.
134
+ iteration : int or array
135
+ Current training step.
136
+ max_iter : int
137
+ Total training steps.
138
+ init_percentile : float
139
+ Initial fraction of samples to include (by loss quantile).
140
+ Default: 0.3 (include the easiest 30% of samples initially).
141
+ schedule : str
142
+ Threshold growth schedule. Default: 'linear'.
143
+ mode : str
144
+ Weighting mode. Default: 'soft'.
145
+
146
+ Returns
147
+ -------
148
+ weighted_loss : scalar
149
+ Curriculum-weighted mean loss.
150
+ """
151
+ # Estimate threshold from loss statistics
152
+ # Use quantile of current losses as adaptive threshold
153
+ sorted_losses = jnp.sort(per_sample_losses)
154
+ n = per_sample_losses.shape[0]
155
+
156
+ # Progress determines what fraction of samples to include
157
+ progress = jnp.clip(iteration / jnp.maximum(max_iter - 1, 1), 0.0, 1.0)
158
+ # Start with init_percentile fraction, end with all samples
159
+ include_frac = init_percentile + (1.0 - init_percentile) * progress
160
+
161
+ # Convert fraction to threshold via quantile
162
+ idx = jnp.clip((include_frac * n).astype(jnp.int32), 0, n - 1)
163
+ threshold = sorted_losses[idx]
164
+
165
+ weights = curriculum_weights(per_sample_losses, threshold, mode=mode)
166
+
167
+ # Weighted mean loss (avoid division by zero)
168
+ total_weight = jnp.sum(weights)
169
+ weighted_loss = jnp.sum(weights * per_sample_losses) / (total_weight + 1e-10)
170
+
171
+ return weighted_loss
@@ -0,0 +1,217 @@
1
+ """Geodesic interpolation and boundary visualization for Riemannian models.
2
+
3
+ Computes geodesic paths between prototypes on Riemannian manifolds for
4
+ decision boundary visualization and model interpretation.
5
+
6
+ On curved manifolds, the decision boundary (locus where d(x, w_i) = d(x, w_j))
7
+ is not a hyperplane — it's a curved surface. Geodesic interpolation allows
8
+ us to:
9
+ 1. Visualize the path between prototypes along the manifold.
10
+ 2. Find the approximate decision boundary point along geodesics.
11
+ 3. Compute prototype midpoints respecting manifold geometry.
12
+
13
+ References
14
+ ----------
15
+ .. [1] Absil, P.-A., Mahony, R., & Sepulchre, R. (2008). Optimization
16
+ Algorithms on Matrix Manifolds. Princeton University Press.
17
+ """
18
+
19
+ import jax.numpy as jnp
20
+
21
+
22
+ def geodesic_interpolation(manifold, point_a, point_b, n_points=50):
23
+ """Compute a geodesic path between two points on a manifold.
24
+
25
+ Uses the exponential map to trace the shortest path (geodesic)
26
+ between two manifold points:
27
+
28
+ .. math::
29
+
30
+ \\gamma(t) = \\text{Exp}_{w_a}(t \\cdot \\text{Log}_{w_a}(w_b)),
31
+ \\quad t \\in [0, 1]
32
+
33
+ Parameters
34
+ ----------
35
+ manifold : SO, SPD, or Grassmannian
36
+ Riemannian manifold instance with ``exp_map`` and ``log_map``.
37
+ point_a : array
38
+ Starting point on the manifold.
39
+ point_b : array
40
+ End point on the manifold.
41
+ n_points : int
42
+ Number of interpolation points along the geodesic.
43
+ Default: 50.
44
+
45
+ Returns
46
+ -------
47
+ path : array of shape (n_points, ...)
48
+ Points along the geodesic from point_a to point_b.
49
+ Each point lies on the manifold.
50
+ """
51
+ # Compute the tangent vector from a to b
52
+ tangent = manifold.log_map(point_a, point_b)
53
+
54
+ # Interpolate along the geodesic
55
+ t_values = jnp.linspace(0.0, 1.0, n_points)
56
+ path = []
57
+ for t in t_values:
58
+ point = manifold.exp_map(point_a, t * tangent)
59
+ path.append(point)
60
+
61
+ return jnp.stack(path, axis=0)
62
+
63
+
64
+ def geodesic_midpoint(manifold, point_a, point_b):
65
+ """Compute the geodesic midpoint between two manifold points.
66
+
67
+ The midpoint is at t = 0.5 along the geodesic:
68
+
69
+ .. math::
70
+
71
+ m = \\text{Exp}_{w_a}(0.5 \\cdot \\text{Log}_{w_a}(w_b))
72
+
73
+ Parameters
74
+ ----------
75
+ manifold : SO, SPD, or Grassmannian
76
+ Riemannian manifold instance.
77
+ point_a : array
78
+ First manifold point.
79
+ point_b : array
80
+ Second manifold point.
81
+
82
+ Returns
83
+ -------
84
+ midpoint : array
85
+ Geodesic midpoint, guaranteed to lie on the manifold.
86
+ """
87
+ tangent = manifold.log_map(point_a, point_b)
88
+ return manifold.exp_map(point_a, 0.5 * tangent)
89
+
90
+
91
+ def decision_boundary_point(manifold, proto_a, proto_b, n_search=100):
92
+ """Find the decision boundary point along the geodesic between prototypes.
93
+
94
+ On curved manifolds, the equidistant point (where d(x, w_a) = d(x, w_b))
95
+ is not necessarily at t = 0.5. This function searches along the geodesic
96
+ for the point where distances to both prototypes are equal.
97
+
98
+ Parameters
99
+ ----------
100
+ manifold : SO, SPD, or Grassmannian
101
+ Riemannian manifold instance.
102
+ proto_a : array
103
+ First prototype (on manifold).
104
+ proto_b : array
105
+ Second prototype (on manifold).
106
+ n_search : int
107
+ Number of candidate points to evaluate. Default: 100.
108
+
109
+ Returns
110
+ -------
111
+ boundary_point : array
112
+ Point on the geodesic where d(point, proto_a) = d(point, proto_b).
113
+ t_boundary : float
114
+ Parameter value t in [0, 1] where the boundary lies.
115
+ t = 0.5 for symmetric manifolds, may differ on curved spaces.
116
+ """
117
+ tangent = manifold.log_map(proto_a, proto_b)
118
+ t_values = jnp.linspace(0.0, 1.0, n_search)
119
+
120
+ # Compute distance difference at each point along the geodesic
121
+ def distance_diff(t):
122
+ point = manifold.exp_map(proto_a, t * tangent)
123
+ d_a = manifold.distance(point, proto_a)
124
+ d_b = manifold.distance(point, proto_b)
125
+ return jnp.abs(d_a - d_b)
126
+
127
+ diffs = jnp.array([distance_diff(t) for t in t_values])
128
+ best_idx = jnp.argmin(diffs)
129
+ t_boundary = t_values[best_idx]
130
+ boundary_point = manifold.exp_map(proto_a, t_boundary * tangent)
131
+
132
+ return boundary_point, float(t_boundary)
133
+
134
+
135
+ def prototype_geodesic_distances(manifold, prototypes, proto_labels):
136
+ """Compute pairwise geodesic distances between prototypes.
137
+
138
+ Parameters
139
+ ----------
140
+ manifold : SO, SPD, or Grassmannian
141
+ Riemannian manifold instance.
142
+ prototypes : array of shape (n_prototypes, ...)
143
+ Prototype points on the manifold (flattened).
144
+ proto_labels : array of shape (n_prototypes,)
145
+ Class labels for prototypes.
146
+
147
+ Returns
148
+ -------
149
+ distances : array of shape (n_prototypes, n_prototypes)
150
+ Pairwise geodesic distance matrix.
151
+ """
152
+ n = prototypes.shape[0]
153
+ distances = jnp.zeros((n, n))
154
+
155
+ for i in range(n):
156
+ for j in range(i + 1, n):
157
+ d = manifold.distance(prototypes[i], prototypes[j])
158
+ distances = distances.at[i, j].set(d)
159
+ distances = distances.at[j, i].set(d)
160
+
161
+ return distances
162
+
163
+
164
+ def inter_class_geodesics(manifold, prototypes, proto_labels, n_points=50):
165
+ """Compute geodesic paths between all inter-class prototype pairs.
166
+
167
+ Useful for visualizing decision boundaries: the boundary between
168
+ two classes lies somewhere along the geodesic connecting their
169
+ closest prototypes.
170
+
171
+ Parameters
172
+ ----------
173
+ manifold : SO, SPD, or Grassmannian
174
+ Riemannian manifold instance.
175
+ prototypes : array of shape (n_prototypes, ...)
176
+ Prototype positions on manifold.
177
+ proto_labels : array of shape (n_prototypes,)
178
+ Class labels.
179
+ n_points : int
180
+ Points per geodesic path. Default: 50.
181
+
182
+ Returns
183
+ -------
184
+ geodesics : list of dict
185
+ Each dict contains:
186
+ - 'path': array of shape (n_points, ...) — geodesic path
187
+ - 'proto_a_idx': int — index of first prototype
188
+ - 'proto_b_idx': int — index of second prototype
189
+ - 'class_a': int — class of first prototype
190
+ - 'class_b': int — class of second prototype
191
+ - 'boundary_t': float — approximate boundary location
192
+ """
193
+ import numpy as np
194
+
195
+ n = prototypes.shape[0]
196
+ labels_np = np.asarray(proto_labels)
197
+ geodesics = []
198
+
199
+ for i in range(n):
200
+ for j in range(i + 1, n):
201
+ if labels_np[i] != labels_np[j]:
202
+ path = geodesic_interpolation(
203
+ manifold, prototypes[i], prototypes[j], n_points
204
+ )
205
+ _, t_boundary = decision_boundary_point(
206
+ manifold, prototypes[i], prototypes[j]
207
+ )
208
+ geodesics.append({
209
+ 'path': path,
210
+ 'proto_a_idx': i,
211
+ 'proto_b_idx': j,
212
+ 'class_a': int(labels_np[i]),
213
+ 'class_b': int(labels_np[j]),
214
+ 'boundary_t': t_boundary,
215
+ })
216
+
217
+ return geodesics
@@ -1457,6 +1457,121 @@ def _wtac_onnx_nodes(dist_name, proto_labels_name):
1457
1457
  return nodes, []
1458
1458
 
1459
1459
 
1460
+ def _wtac_with_rejection_onnx_nodes(dist_name, proto_labels_name, threshold):
1461
+ """WTAC with reject option: confidence-based rejection.
1462
+
1463
+ Computes:
1464
+ winner_label = proto_labels[argmin(distances, axis=1)]
1465
+ d_plus = min distance to same-class prototype (winner's class)
1466
+ d_minus = min distance to different-class prototype
1467
+ confidence = (d_minus - d_plus) / (d_minus + d_plus + eps)
1468
+ predictions = where(confidence >= threshold, winner_label, -1)
1469
+
1470
+ Outputs two tensors: 'predictions' (INT64) and 'confidence' (FLOAT).
1471
+ """
1472
+ import onnx.helper as oh
1473
+ from onnx import TensorProto
1474
+
1475
+ nodes = []
1476
+ inits = []
1477
+
1478
+ # Constants
1479
+ inits.extend([
1480
+ oh.make_tensor('_rej_eps', TensorProto.FLOAT, [], [1e-10]),
1481
+ oh.make_tensor('_rej_threshold', TensorProto.FLOAT, [], [float(threshold)]),
1482
+ oh.make_tensor('_rej_inf', TensorProto.FLOAT, [], [1e30]),
1483
+ oh.make_tensor('_rej_neg_one', TensorProto.INT64, [], [-1]),
1484
+ ])
1485
+
1486
+ # Step 1: winner_idx = argmin(distances, axis=1) → (batch,)
1487
+ nodes.append(oh.make_node(
1488
+ 'ArgMin', [dist_name], ['_rej_winner_idx'],
1489
+ axis=1, keepdims=0,
1490
+ ))
1491
+ nodes.append(oh.make_node(
1492
+ 'Cast', ['_rej_winner_idx'], ['_rej_winner_idx_i64'],
1493
+ to=TensorProto.INT64,
1494
+ ))
1495
+
1496
+ # Step 2: winner_label = proto_labels[winner_idx] → (batch,)
1497
+ nodes.append(oh.make_node(
1498
+ 'Gather', [proto_labels_name, '_rej_winner_idx_i64'], ['_rej_winner_label'],
1499
+ axis=0,
1500
+ ))
1501
+
1502
+ # Step 3: Build same-class mask
1503
+ # Unsqueeze winner_label to (batch, 1) for broadcasting
1504
+ nodes.append(oh.make_node(
1505
+ 'Unsqueeze', ['_rej_winner_label', '_rej_unsqueeze_axis'], ['_rej_winner_label_2d'],
1506
+ ))
1507
+ inits.append(
1508
+ oh.make_tensor('_rej_unsqueeze_axis', TensorProto.INT64, [1], [1]),
1509
+ )
1510
+ # Unsqueeze proto_labels to (1, n_protos) for broadcasting
1511
+ nodes.append(oh.make_node(
1512
+ 'Unsqueeze', [proto_labels_name, '_rej_unsqueeze_axis0'], ['_rej_proto_labels_2d'],
1513
+ ))
1514
+ inits.append(
1515
+ oh.make_tensor('_rej_unsqueeze_axis0', TensorProto.INT64, [1], [0]),
1516
+ )
1517
+ # same_mask = (winner_label == proto_labels) → (batch, n_protos)
1518
+ nodes.append(oh.make_node(
1519
+ 'Equal', ['_rej_winner_label_2d', '_rej_proto_labels_2d'], ['_rej_same_mask'],
1520
+ ))
1521
+ # diff_mask = NOT same_mask
1522
+ nodes.append(oh.make_node(
1523
+ 'Not', ['_rej_same_mask'], ['_rej_diff_mask'],
1524
+ ))
1525
+
1526
+ # Step 4: d_plus = min(where(same_mask, distances, INF), axis=1)
1527
+ # Use Add with large constant to create INF-filled tensor matching dist shape
1528
+ # Where broadcasts scalars automatically against the mask shape
1529
+ nodes.append(oh.make_node(
1530
+ 'Where', ['_rej_same_mask', dist_name, '_rej_inf'],
1531
+ ['_rej_dist_same'],
1532
+ ))
1533
+ nodes.append(oh.make_node(
1534
+ 'ReduceMin', ['_rej_dist_same'], ['_rej_d_plus'],
1535
+ axes=[1], keepdims=0,
1536
+ ))
1537
+
1538
+ # Step 5: d_minus = min(where(diff_mask, distances, INF), axis=1)
1539
+ nodes.append(oh.make_node(
1540
+ 'Where', ['_rej_diff_mask', dist_name, '_rej_inf'],
1541
+ ['_rej_dist_diff'],
1542
+ ))
1543
+ nodes.append(oh.make_node(
1544
+ 'ReduceMin', ['_rej_dist_diff'], ['_rej_d_minus'],
1545
+ axes=[1], keepdims=0,
1546
+ ))
1547
+
1548
+ # Step 6: confidence = (d_minus - d_plus) / (d_minus + d_plus + eps)
1549
+ nodes.append(oh.make_node(
1550
+ 'Sub', ['_rej_d_minus', '_rej_d_plus'], ['_rej_numerator'],
1551
+ ))
1552
+ nodes.append(oh.make_node(
1553
+ 'Add', ['_rej_d_minus', '_rej_d_plus'], ['_rej_denom_raw'],
1554
+ ))
1555
+ nodes.append(oh.make_node(
1556
+ 'Add', ['_rej_denom_raw', '_rej_eps'], ['_rej_denominator'],
1557
+ ))
1558
+ nodes.append(oh.make_node(
1559
+ 'Div', ['_rej_numerator', '_rej_denominator'], ['confidence'],
1560
+ ))
1561
+
1562
+ # Step 7: predictions = where(confidence >= threshold, winner_label, -1)
1563
+ nodes.append(oh.make_node(
1564
+ 'GreaterOrEqual', ['confidence', '_rej_threshold'], ['_rej_accept_mask'],
1565
+ ))
1566
+ # Where broadcasts scalar _rej_neg_one against _rej_winner_label shape
1567
+ nodes.append(oh.make_node(
1568
+ 'Where', ['_rej_accept_mask', '_rej_winner_label', '_rej_neg_one'],
1569
+ ['predictions'],
1570
+ ))
1571
+
1572
+ return nodes, inits
1573
+
1574
+
1460
1575
  def _argmin_onnx_nodes(dist_name):
1461
1576
  """Unsupervised: predictions = argmin(distances, axis=1)."""
1462
1577
  import onnx.helper as oh
@@ -2374,6 +2489,7 @@ def export_onnx(
2374
2489
  batch_size: int = 1,
2375
2490
  opset_version: int = 17,
2376
2491
  path: str | None = None,
2492
+ reject_threshold: float | None = None,
2377
2493
  ):
2378
2494
  """Export a fitted model's predict function to ONNX format.
2379
2495
 
@@ -2383,7 +2499,7 @@ def export_onnx(
2383
2499
  (reasoning matrices), PLVQ (Gaussian mixture), encoder models
2384
2500
  (MLP/CNN backbone), Differentiating Kernel models (Gaussian,
2385
2501
  relevance-weighted, and exponential kernels), and Riemannian
2386
- models on SO(n)/Grassmannian manifolds (85 of 87 models total).
2502
+ models on SO(n)/Grassmannian manifolds (88 of 102 models total).
2387
2503
 
2388
2504
  Parameters
2389
2505
  ----------
@@ -2396,6 +2512,13 @@ def export_onnx(
2396
2512
  ONNX opset version. Default: 17.
2397
2513
  path : str, optional
2398
2514
  If provided, save the ONNX model to this file path.
2515
+ reject_threshold : float, optional
2516
+ If provided, enables reject option for supervised models.
2517
+ Samples with confidence below this threshold are assigned
2518
+ prediction -1 (rejected). The exported model produces two
2519
+ outputs: ``predictions`` (INT64, with -1 for rejected) and
2520
+ ``confidence`` (FLOAT, in [-1, 1]).
2521
+ Only supported for supervised models (WTAC decision).
2399
2522
 
2400
2523
  Returns
2401
2524
  -------
@@ -2409,6 +2532,8 @@ def export_onnx(
2409
2532
  supported.
2410
2533
  ImportError
2411
2534
  If the ``onnx`` package is not installed.
2535
+ ValueError
2536
+ If ``reject_threshold`` is used with a non-supervised model.
2412
2537
  """
2413
2538
  _check_onnx_installed()
2414
2539
  import onnx
@@ -2649,7 +2774,18 @@ def export_onnx(
2649
2774
  initializers.extend(extra_inits)
2650
2775
 
2651
2776
  # --- Decision / competition nodes ---
2652
- if model_type == 'supervised':
2777
+ use_rejection = reject_threshold is not None
2778
+
2779
+ if use_rejection:
2780
+ if model_type != 'supervised':
2781
+ raise ValueError(
2782
+ "reject_threshold is only supported for supervised models "
2783
+ f"(WTAC decision), got model_type='{model_type}'."
2784
+ )
2785
+ comp_nodes, comp_inits = _wtac_with_rejection_onnx_nodes(
2786
+ dist_out, 'proto_labels', reject_threshold,
2787
+ )
2788
+ elif model_type == 'supervised':
2653
2789
  comp_nodes, comp_inits = _wtac_onnx_nodes(dist_out, 'proto_labels')
2654
2790
  elif model_type == 'unsupervised':
2655
2791
  comp_nodes, comp_inits = _argmin_onnx_nodes(dist_out)
@@ -2687,11 +2823,18 @@ def export_onnx(
2687
2823
  'predictions', output_dtype, [batch_dim],
2688
2824
  )
2689
2825
 
2826
+ outputs = [Y_output]
2827
+ if use_rejection:
2828
+ conf_output = oh.make_tensor_value_info(
2829
+ 'confidence', TensorProto.FLOAT, [batch_dim],
2830
+ )
2831
+ outputs.append(conf_output)
2832
+
2690
2833
  graph = oh.make_graph(
2691
2834
  all_nodes,
2692
2835
  'prosemble_predict',
2693
2836
  [X_input],
2694
- [Y_output],
2837
+ outputs,
2695
2838
  initializer=initializers,
2696
2839
  )
2697
2840