DARTsort 0.5.4__tar.gz → 0.5.6__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 (224) hide show
  1. {dartsort-0.5.4 → dartsort-0.5.6}/PKG-INFO +1 -1
  2. {dartsort-0.5.4 → dartsort-0.5.6}/docs/main_api.md +2 -2
  3. {dartsort-0.5.4 → dartsort-0.5.6}/docs/toolkit_api.md +17 -1
  4. {dartsort-0.5.4 → dartsort-0.5.6}/pyproject.toml +5 -0
  5. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/__init__.py +3 -2
  6. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/clustering.py +21 -1
  7. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/clustering_features.py +7 -3
  8. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/density.py +139 -151
  9. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/kmeans.py +15 -31
  10. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/merge.py +7 -12
  11. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/mixture.py +149 -72
  12. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/config.py +4 -3
  13. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/evaluate/hybrid_util.py +7 -7
  14. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/evaluate/simkit.py +11 -7
  15. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/main.py +19 -15
  16. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/matching.py +4 -3
  17. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/matching_util/pairwise.py +2 -2
  18. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/reduction_template.py +2 -2
  19. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/subtract.py +30 -17
  20. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/templates/postprocess_util.py +10 -2
  21. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/templates/realignment.py +3 -1
  22. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/amortized_localization.py +3 -3
  23. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/pipeline.py +7 -3
  24. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/reduction.py +1 -1
  25. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/single_channel_denoiser.py +2 -2
  26. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/cli_util.py +2 -3
  27. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/data_util.py +22 -41
  28. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/drift_util.py +1 -1
  29. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/internal_config.py +11 -12
  30. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/logging_util.py +43 -24
  31. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/main_util.py +16 -12
  32. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/motion.py +6 -3
  33. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/multiprocessing_util.py +9 -0
  34. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/nn_util.py +2 -0
  35. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/noise_util.py +4 -3
  36. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/peel_util.py +12 -10
  37. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/preprocess_util.py +6 -2
  38. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/py_util.py +11 -4
  39. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/spiketorch.py +32 -9
  40. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/waveform_util.py +1 -41
  41. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/analysis_plots.py +4 -2
  42. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/colors.py +2 -2
  43. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/mixture.py +102 -37
  44. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/scatterplots.py +72 -1
  45. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/unit.py +1 -0
  46. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort.egg-info/PKG-INFO +1 -1
  47. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort.egg-info/SOURCES.txt +1 -0
  48. {dartsort-0.5.4 → dartsort-0.5.6}/tests/conftest.py +6 -6
  49. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_alignment.py +1 -1
  50. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_dartsort.py +1 -1
  51. dartsort-0.5.6/tests/test_density.py +71 -0
  52. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_matching.py +1 -1
  53. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_simkit.py +2 -2
  54. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_templates.py +4 -3
  55. {dartsort-0.5.4 → dartsort-0.5.6}/.github/workflows/ci.yml +0 -0
  56. {dartsort-0.5.4 → dartsort-0.5.6}/.github/workflows/ci_windows.yml +0 -0
  57. {dartsort-0.5.4 → dartsort-0.5.6}/.github/workflows/deploy.yml +0 -0
  58. {dartsort-0.5.4 → dartsort-0.5.6}/.github/workflows/docs.yml +0 -0
  59. {dartsort-0.5.4 → dartsort-0.5.6}/.gitignore +0 -0
  60. {dartsort-0.5.4 → dartsort-0.5.6}/README.md +0 -0
  61. {dartsort-0.5.4 → dartsort-0.5.6}/docs/index.md +0 -0
  62. {dartsort-0.5.4 → dartsort-0.5.6}/environment.yml +0 -0
  63. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/.keep +0 -0
  64. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/basic_integration_test.py +0 -0
  65. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/ci diag.py +0 -0
  66. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/dartsort_dredge_demo.ipynb +0 -0
  67. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/dartsort_dredge_demo.py +0 -0
  68. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/data test.py +0 -0
  69. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/dead reckoning.py +0 -0
  70. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/denoise np1.py +0 -0
  71. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/denoise np2.py +0 -0
  72. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/exploratory.py +0 -0
  73. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/geom debug.py +0 -0
  74. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/hdbscan_clustering_5min.py +0 -0
  75. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/hybrid 10.py +0 -0
  76. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/hybrid cluster.py +0 -0
  77. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/hybrid_11.py +0 -0
  78. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/initial training.py +0 -0
  79. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/make hybrid.py +0 -0
  80. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/mgh.py +0 -0
  81. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/np1 feats.py +0 -0
  82. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/np2 processing version c.py +0 -0
  83. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/paperfigs.py +0 -0
  84. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/pdists.py +0 -0
  85. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/postcheck.py +0 -0
  86. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/ptp vae.py +0 -0
  87. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/re clust.py +0 -0
  88. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/reg fig loop.py +0 -0
  89. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/reg_sim_exp.py +0 -0
  90. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/relocating.py +0 -0
  91. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/relocating_all.py +0 -0
  92. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/restats +0 -0
  93. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/sortStinger1.py +0 -0
  94. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/sortStinger3.py +0 -0
  95. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/sortStinger4.py +0 -0
  96. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/sortStinger5.py +0 -0
  97. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/sortStinger6.py +0 -0
  98. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/spike farm.py +0 -0
  99. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/ss_ensemble_sim.py +0 -0
  100. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/ss_ensemble_sim_drift.py +0 -0
  101. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/ss_sim.py +0 -0
  102. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/stan diagnostics.py +0 -0
  103. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/storytelling c.py +0 -0
  104. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/storytelling.py +0 -0
  105. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/subtract_with_phaseshift_denoiser.py +0 -0
  106. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/debug_matching.py +0 -0
  107. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/test_agreement.py +0 -0
  108. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/test_amplitude_scaling.py +0 -0
  109. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/test_batch_optim.py +0 -0
  110. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/test_ibme.py +0 -0
  111. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/test_mm_initialization.py +0 -0
  112. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/test_normxcorr1d.py +0 -0
  113. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/test_reassignment.py +0 -0
  114. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/tests/test_relocation.py +0 -0
  115. {dartsort-0.5.4 → dartsort-0.5.6}/notebook/thomas.py +0 -0
  116. {dartsort-0.5.4 → dartsort-0.5.6}/setup.cfg +0 -0
  117. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/cli.py +0 -0
  118. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/__init__.py +0 -0
  119. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/agglomerate.py +0 -0
  120. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/cluster_util.py +0 -0
  121. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/forward_backward.py +0 -0
  122. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/modes.py +0 -0
  123. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/clustering/refine_util.py +0 -0
  124. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/detect/__init__.py +0 -0
  125. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/detect/detect.py +0 -0
  126. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/detect/detect_filters.py +0 -0
  127. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/evaluate/__init__.py +0 -0
  128. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/evaluate/analysis.py +0 -0
  129. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/evaluate/comparison.py +0 -0
  130. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/evaluate/noise_recording_tools.py +0 -0
  131. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/evaluate/sim_template_tools.py +0 -0
  132. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/evaluate/simlib.py +0 -0
  133. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/localize/__init__.py +0 -0
  134. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/localize/localize_torch.py +0 -0
  135. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/localize/localize_util.py +0 -0
  136. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/__init__.py +0 -0
  137. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/grab.py +0 -0
  138. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/matching_util/__init__.py +0 -0
  139. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/matching_util/compressed_upsampled.py +0 -0
  140. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/matching_util/drifty.py +0 -0
  141. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/matching_util/matching_base.py +0 -0
  142. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/matching_util/pairwise_util.py +0 -0
  143. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/peel_base.py +0 -0
  144. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/peel_lib.py +0 -0
  145. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/shaver.py +0 -0
  146. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/peel/threshold.py +0 -0
  147. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/default_temporal_kernel.npy +0 -0
  148. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/detect_np1.pt +0 -0
  149. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/detect_np2.pt +0 -0
  150. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/glasbey1024.npz +0 -0
  151. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/offset_trained_denoiser.pt +0 -0
  152. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/offset_trained_denoiser_added_temp.pt +0 -0
  153. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/offset_trained_denoiser_added_temp_boost.pt +0 -0
  154. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/original_single_chan_denoiser.pt +0 -0
  155. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/pretrained/single_chan_denoiser.pt +0 -0
  156. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/templates/__init__.py +0 -0
  157. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/templates/get_templates.py +0 -0
  158. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/templates/template_util.py +0 -0
  159. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/templates/templates.py +0 -0
  160. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/templates/templib.py +0 -0
  161. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/__init__.py +0 -0
  162. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/_multichan_denoiser_kit.py +0 -0
  163. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/all_transformers.py +0 -0
  164. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/amplitudes.py +0 -0
  165. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/decollider.py +0 -0
  166. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/denoising_scorer.py +0 -0
  167. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/enforce_decrease.py +0 -0
  168. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/fixed_prop.py +0 -0
  169. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/interp.py +0 -0
  170. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/localize.py +0 -0
  171. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/matching_denoiser.py +0 -0
  172. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/mixture_classifier.py +0 -0
  173. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/supervised_denoiser.py +0 -0
  174. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/temporal_pca.py +0 -0
  175. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/transform_base.py +0 -0
  176. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/transform/whiten.py +0 -0
  177. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/__init__.py +0 -0
  178. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/interpolation_util.py +0 -0
  179. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/job_util.py +0 -0
  180. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/more_operators.py +0 -0
  181. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/registration_util.py +0 -0
  182. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/residual_util.py +0 -0
  183. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/sparse_util.py +0 -0
  184. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/spikeio.py +0 -0
  185. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/testing_util/__init__.py +0 -0
  186. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/testing_util/matching_debug_util.py +0 -0
  187. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/testing_util/mixture_testing_util.py +0 -0
  188. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/torch_optimization_util.py +0 -0
  189. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/util/torch_util.py +0 -0
  190. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/__init__.py +0 -0
  191. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/gt.py +0 -0
  192. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/layout.py +0 -0
  193. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/metrics_util.py +0 -0
  194. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/recanim.py +0 -0
  195. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/sorting.py +0 -0
  196. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/unit_comparison.py +0 -0
  197. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/versus.py +0 -0
  198. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/vismain.py +0 -0
  199. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort/vis/waveforms.py +0 -0
  200. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort.egg-info/dependency_links.txt +0 -0
  201. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort.egg-info/entry_points.txt +0 -0
  202. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort.egg-info/requires.txt +0 -0
  203. {dartsort-0.5.4 → dartsort-0.5.6}/src/dartsort.egg-info/top_level.txt +0 -0
  204. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_clustering.py +0 -0
  205. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_config.py +0 -0
  206. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_data_util.py +0 -0
  207. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_detect.py +0 -0
  208. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_drift_util.py +0 -0
  209. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_drifty_matching.py +0 -0
  210. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_enforce_decrease.py +0 -0
  211. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_grab_and_featurize.py +0 -0
  212. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_kmeans.py +0 -0
  213. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_linalg.py +0 -0
  214. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_localize_torch.py +0 -0
  215. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_mixture_classifier.py +0 -0
  216. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_modeling.py +0 -0
  217. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_spiketorch.py +0 -0
  218. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_subtract.py +0 -0
  219. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_threshold.py +0 -0
  220. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_torch_optimization_util.py +0 -0
  221. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_transform.py +0 -0
  222. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_util.py +0 -0
  223. {dartsort-0.5.4 → dartsort-0.5.6}/tests/test_waveform_util.py +0 -0
  224. {dartsort-0.5.4 → dartsort-0.5.6}/zensical.toml +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: dartsort
3
- Version: 0.5.4
3
+ Version: 0.5.6
4
4
  Summary: DARTsort
5
5
  Author-email: Charlie Windolf <ciw2048@gmail.com>
6
6
  Project-URL: Homepage, https://github.com/cwindolf/dartsort
@@ -17,9 +17,9 @@ For details on parameters you should think about before running the sorter, see
17
17
  show_signature: true
18
18
  separate_signature: true
19
19
 
20
- The return value from the `dartsort()` function is a DARTsortReturn object, which is a dictionary containing spike trains and motion information:
20
+ The return value from the `dartsort()` function is a `DARTsortResult` object, which is a dictionary containing spike trains and motion information:
21
21
 
22
- ::: dartsort.DARTsortReturn
22
+ ::: dartsort.DARTsortResult
23
23
  options:
24
24
  show_bases: false
25
25
 
@@ -17,7 +17,13 @@ In *dartsort* those kind of workflows are called "peelers" (inspired by the iter
17
17
  All of the peelers accept a featurization pipeline or a [`FeaturizationConfig`](#dartsort.FeaturizationConfig) object which handles the spike featurization; these are discussed [in the next section](#featurization-pipelines).
18
18
 
19
19
  *dartsort* includes high-level functions for running various kinds of peelers and corresponding configuration objects.
20
- These are:
20
+ These are in sections below.
21
+
22
+ ### Template matching
23
+
24
+ [`match()`](#dartsort.match) runs template matching from known templates if its `template_data` parameter is set, or else estimating templates from `sorting` using [`estimate_template_library()`](#dartsort.estimate_template_library).
25
+
26
+ It is configured by the [matching_cfg](#dartsort.MatchingConfig) argument.
21
27
 
22
28
  ::: dartsort.match
23
29
 
@@ -25,18 +31,28 @@ These are:
25
31
  options:
26
32
  show_if_no_docstring: true
27
33
 
34
+ ### Neural-net based collision-cleaned spike detection
35
+
28
36
  ::: dartsort.subtract
29
37
 
30
38
  ::: dartsort.SubtractionConfig
31
39
  options:
32
40
  show_if_no_docstring: true
33
41
 
42
+ ### Thresholding spike detection
43
+
34
44
  ::: dartsort.threshold
35
45
 
36
46
  ::: dartsort.ThresholdingConfig
37
47
  options:
38
48
  show_if_no_docstring: true
39
49
 
50
+
51
+ ### Spike extraction and featurization at known event times
52
+
53
+ ::: dartsort.grab
54
+
55
+
40
56
  ## Featurization pipelines
41
57
 
42
58
  Featurization is configured by building a `FeaturizationConfig`.
@@ -81,6 +81,11 @@ ignore = ["F722"]
81
81
  [tool.ty.src]
82
82
  include = ["src", "tests"]
83
83
 
84
+ [tool.basedpyright]
85
+ include = ["src", "tests"]
86
+ enableTypeIgnoreComments = true
87
+ typeCheckingMode = "standard"
88
+
84
89
  [tool.ty.analysis]
85
90
  replace-imports-with-any = ["hdbscan"]
86
91
 
@@ -24,7 +24,7 @@ from .localize.localize_util import (
24
24
  from .main import (
25
25
  ObjectiveUpdateTemplateMatchingPeeler,
26
26
  SubtractionPeeler,
27
- DARTsortReturn,
27
+ DARTsortResult,
28
28
  check_recording,
29
29
  cluster,
30
30
  dartsort,
@@ -74,11 +74,12 @@ from .util.logging_util import (
74
74
  DARTSORTVERBOSE,
75
75
  DARTsortLogger,
76
76
  get_logger,
77
+ set_log_level,
77
78
  )
78
79
  from .util.motion import MotionInfo, get_motion_info, try_load_motion_info
79
80
  from .util.noise_util import EmbeddedNoise
80
81
  from .util.preprocess_util import preprocess
81
- from .util.py_util import databag, resolve_path
82
+ from .util.py_util import databag, ensure_path
82
83
  from .util.waveform_util import full_channel_index, make_channel_index
83
84
 
84
85
  __version__ = importlib.metadata.version("dartsort")
@@ -352,9 +352,27 @@ class DensityPeaksClusterer(Clusterer):
352
352
  recording: BaseRecording | None,
353
353
  motion: MotionInfo,
354
354
  ) -> np.ndarray:
355
+ labels, _ = self._cluster_extra(
356
+ features=features,
357
+ stable_features=stable_features,
358
+ sorting=sorting,
359
+ recording=recording,
360
+ motion=motion,
361
+ )
362
+ return labels
363
+
364
+ def _cluster_extra(
365
+ self,
366
+ features: SimpleMatrixFeatures,
367
+ stable_features: StableWaveformFeatures | None,
368
+ sorting: DARTsortSorting,
369
+ recording: BaseRecording | None,
370
+ motion: MotionInfo,
371
+ ) -> tuple[np.ndarray, dict]:
355
372
  subsampling, ixs = self.handle_sampling(features)
356
373
  X = features.features
357
374
  X_fit = X[ixs]
375
+ assert self.computation_cfg is not None
358
376
 
359
377
  if not self.uhdversion:
360
378
  res = density.density_peaks(
@@ -369,6 +387,7 @@ class DensityPeaksClusterer(Clusterer):
369
387
  outlier_radius=self.outlier_radius,
370
388
  outlier_neighbor_count=self.outlier_neighbor_count,
371
389
  workers=self.workers,
390
+ device=self.computation_cfg.actual_device(),
372
391
  )
373
392
  else:
374
393
  res = density.density_peaks_fancy(
@@ -403,13 +422,14 @@ class DensityPeaksClusterer(Clusterer):
403
422
  ixs, res["labels"], rest, other_labels
404
423
  )
405
424
  else:
425
+ kdtree = None
406
426
  labels = res["labels"]
407
427
 
408
428
  labels = cluster_util.decrumb(
409
429
  labels, min_size=self.remove_clusters_smaller_than, in_place=True
410
430
  )
411
431
 
412
- return labels
432
+ return labels, dict(dpc_res=res, kdtree=kdtree, ixs=ixs)
413
433
 
414
434
 
415
435
  clustering_strategies["dpc"] = DensityPeaksClusterer
@@ -7,7 +7,11 @@ from torch import Tensor
7
7
 
8
8
  from ..util.data_util import DARTsortSorting
9
9
  from ..util.drift_util import get_stable_channels
10
- from ..util.internal_config import ClusteringFeaturesConfig, ComputationConfig
10
+ from ..util.internal_config import (
11
+ ClusteringFeaturesConfig,
12
+ ComputationConfig,
13
+ default_clustering_features_cfg,
14
+ )
11
15
  from ..util.interpolation_util import (
12
16
  SpikeNeighborhoods,
13
17
  StableFeaturesInterpolator,
@@ -63,8 +67,8 @@ class SimpleMatrixFeatures:
63
67
  *,
64
68
  sorting: DARTsortSorting,
65
69
  motion: MotionInfo,
66
- clustering_features_cfg: ClusteringFeaturesConfig,
67
- computation_cfg: ComputationConfig | None,
70
+ clustering_features_cfg: ClusteringFeaturesConfig = default_clustering_features_cfg,
71
+ computation_cfg: ComputationConfig | None = None,
68
72
  ) -> Self:
69
73
  computation_cfg = ensure_computation_config(computation_cfg)
70
74
  t_s = sorting.times_seconds
@@ -1,21 +1,16 @@
1
1
  from threading import local
2
- from typing import Literal, cast
2
+ from typing import cast
3
3
 
4
4
  import numba
5
5
  import numpy as np
6
6
  import torch
7
- import torch.nn.functional as F
8
7
  from scipy.sparse import coo_array
9
8
  from scipy.sparse.csgraph import connected_components
10
9
  from scipy.spatial import KDTree
11
- from scipy.spatial.distance import pdist, squareform
12
- from torch import Tensor
13
10
 
14
- from ..util.internal_config import ComputationConfig
15
- from ..util.job_util import ensure_computation_config
16
- from ..util.logging_util import get_logger, progbar, progrange
11
+ from ..util.logging_util import DARTSORTVERBOSE, get_logger, progbar, progrange
17
12
  from ..util.multiprocessing_util import get_pool
18
- from ..util.py_util import timer
13
+ from ..util.spiketorch import sqeuc_cdist_known_norm
19
14
  from .cluster_util import decrumb
20
15
 
21
16
  logger = get_logger(__name__)
@@ -266,136 +261,118 @@ def remove_border_points(
266
261
  return new_labels
267
262
 
268
263
 
269
- def guess_mode(
270
- X,
271
- sigma: float | Literal["rule_of_thumb"] = "rule_of_thumb",
272
- outlier_neighbor_count=10,
273
- outlier_sigma=3.0,
274
- kdtree=None,
275
- workers=1,
276
- ):
277
- """Use a KDE to guess the highest density point."""
278
- n = len(X)
279
- sigma0 = np.sqrt(X.var(axis=0).sum())
280
- inliers, kdtree = kdtree_inliers(
281
- X,
282
- kdtree=kdtree,
283
- n_neighbors=outlier_neighbor_count,
284
- distance_upper_bound=outlier_sigma * sigma0,
285
- workers=workers,
286
- )
287
-
288
- if sigma == "rule_of_thumb":
289
- sigma_dens = (
290
- 1.06
291
- * np.linalg.norm(np.std(X[inliers], axis=0))
292
- * np.power(inliers.sum(), -0.2)
293
- )
294
- else:
295
- sigma_dens = sigma
296
-
297
- density = get_smoothed_density(X, inliers=inliers, sigma=sigma_dens)
298
- assert isinstance(density, np.ndarray)
299
- assert density.shape == (n,)
300
-
301
- return np.argmax(density)
302
-
303
-
304
- def bucket_density_ratio(
305
- X: np.ndarray,
306
- scaled_geom: np.ndarray,
307
- sigma: float,
308
- sigma_regional: float,
264
+ def sort_density(
265
+ X: np.ndarray | torch.Tensor,
266
+ sigma0: float,
267
+ sigma1: float,
309
268
  max_sigma: float = 3.0,
310
- computation_cfg: ComputationConfig | None = None,
311
- workers=-1,
312
- batch_size=8192,
313
- ):
314
- """
269
+ device: torch.device = torch.device("cpu"),
270
+ batch_size: int = 2048,
271
+ col_batch_size: int = 1024,
272
+ show_progress: bool = False,
273
+ ) -> np.ndarray:
274
+ """Compute a ratio of Gaussian KDEs by ordering along the first dimension
315
275
 
316
- Algorithm:
317
- - Assign spikes to channels with a geom-kdtree (make sure geom's x
318
- scale matches the clustering feature's x scale.)
319
- - Let Dc be pairwise distances between channels
320
- - Compute sparse distance matrix. If Dx is pairwise distance between
321
- all pairs of spikes, we just want the entries with Dx<max_dist.
322
- - Note
323
- |ci-cj| = |(ci-xi) + (xj-cj) + (xi-xj)|
324
- <= 2C + |xi-xj|
325
- Thus we need not consider i,j s.t. |ci-cj|-2C>max_dist.
326
- """
327
- computation_cfg = ensure_computation_config(computation_cfg)
328
- with timer("geom bucketing"):
329
- geom_kdt = KDTree(scaled_geom, leafsize=1)
330
- dists, channels = geom_kdt.query(X[:, :2], workers=workers)
331
-
332
- max_dist_by_chan = np.zeros(len(scaled_geom))
333
- for c in range(len(scaled_geom)):
334
- inc = np.flatnonzero(channels == c)
335
- if inc.size:
336
- max_dist_by_chan[c] = dists[inc].max()
337
-
338
- larger_sigma = max(sigma, sigma_regional or 0.0)
339
- max_dist = max_sigma * larger_sigma
340
-
341
- lb = squareform(pdist(scaled_geom))
342
- lb -= max_dist_by_chan[:, None]
343
- lb -= max_dist_by_chan
344
- chans_pair_mask = lb <= max_dist
345
-
346
- device = computation_cfg.actual_device()
347
- channels = torch.asarray(channels, device=device, dtype=torch.long)
348
- Xt = torch.asarray(X, device=device, dtype=torch.float)
349
- chans_pair_mask = torch.asarray(chans_pair_mask, device=device, dtype=torch.bool)
350
- density_ratio = torch.full((len(X),), -torch.inf, device=device)
351
-
352
- # for c in trange(len(scaled_geom), desc="bktdens"):
353
- for c in progrange(25, desc="bktdens"):
354
- (inc,) = (channels == c).nonzero(as_tuple=True)
355
- (friends,) = chans_pair_mask[c][channels].nonzero(as_tuple=True)
356
- density_ratio[inc] = _local_sparse_dens_ratio(
357
- Xt, inc, friends, max_dist, sigma, sigma_regional, batch_size=batch_size
358
- )
276
+ This is only useful if the first dimension is much larger than the rest.
277
+ Otherwise, hopefully n is not too big ;).
359
278
 
360
- return density_ratio
279
+ Parameters
280
+ ----------
281
+ X : np.ndarray | torch.Tensor
282
+ Data whose density ratio we want.
283
+ sigma0 : float
284
+ The smaller sigma
285
+ sigma1 : float
286
+ The larger sigma
287
+ max_sigma : float, optional
288
+ The maximum sigma value to use. Default is 3.0.
289
+ device : torch.device, optional
290
+ batch_size : int, optional
291
+ show_progress : bool, optional
361
292
 
293
+ Returns
294
+ -------
295
+ np.ndarray
296
+ The density ratio for each point in X.
297
+ """
298
+ X = torch.asarray(X, device=device)
299
+ order = torch.argsort(X[:, 0])
300
+ X = X[order]
301
+ X0 = X[:, 0].contiguous()
302
+ Xnormsq = torch.linalg.vector_norm(X, dim=1).square_()
303
+ n, dim = X.shape
362
304
 
363
- @torch.jit.script
364
- def _local_sparse_dens_ratio(
365
- X: Tensor,
366
- inc: Tensor,
367
- inf: Tensor,
368
- max_dist: float,
369
- sigma0: float,
370
- sigma1: float,
371
- batch_size: int = 1024,
372
- ):
373
- nc = inc.shape[0]
374
- nf = inf.shape[0]
375
- dens_ratio = X.new_empty(nc)
305
+ radius = max_sigma * max(sigma0, sigma1)
306
+ radius_sq = X.new_tensor(radius * radius)
307
+ dens = X.new_zeros(n)
376
308
 
377
- s0_factor = (1.0 / (torch.sqrt(torch.tensor(2.0)) * sigma0)).to(X)
378
- s1_factor = (1.0 / (torch.sqrt(torch.tensor(2.0)) * sigma1)).to(X)
309
+ k0 = X.new_tensor(-0.5 * sigma0**-2)
310
+ k1 = X.new_tensor(-0.5 * sigma1**-2)
379
311
 
380
- for ic0 in range(0, nc, batch_size):
381
- ic1 = min(nc, ic0 + batch_size)
382
- Xc = X[inc[ic0:ic1]]
383
- for if0 in range(0, nf, batch_size):
384
- if1 = min(nf, if0 + batch_size)
385
- Xf = X[inf[if0:if1]]
312
+ n0 = X.new_tensor(-0.5 * dim * np.log(2.0 * np.pi * sigma0 * sigma0))
313
+ n1 = X.new_tensor(-0.5 * dim * np.log(2.0 * np.pi * sigma1 * sigma1))
386
314
 
387
- d = torch.cdist(Xc, Xf)
388
- F.threshold(d, threshold=max_dist, value=torch.inf, inplace=True)
315
+ if show_progress:
316
+ iter = progrange(0, n, batch_size, desc=f'SortDens:{device.type}')
317
+ else:
318
+ iter = range(0, n, batch_size)
319
+
320
+ i0 = i1 = 0
321
+ dist_buf = X.new_zeros((batch_size * col_batch_size,))
322
+ dist_buf2 = X.new_zeros((batch_size * col_batch_size,))
323
+ dist_buf3 = X.new_zeros((batch_size * col_batch_size,))
324
+ dens0_buf = X.new_zeros((batch_size,))
325
+ dens1_buf = X.new_zeros((batch_size,))
326
+ for q0 in iter:
327
+ q1 = min(n, q0 + batch_size)
328
+ nq = q1 - q0
329
+
330
+ Q = X[q0:q1]
331
+ x0 = Q[0, 0] - radius
332
+ x1 = Q[-1, 0] + radius
333
+
334
+ if i0 == q0:
335
+ assert q0 == 0
336
+ else:
337
+ assert q0 > i0
338
+ i0 = i0 + torch.searchsorted(X0[i0:q0], x0, side="right") - 1
339
+ i0.clamp_(min=0)
340
+ i0 = int(i0.item())
389
341
 
390
- d0 = d * s0_factor
391
- d1 = d * s1_factor
392
- d0.square_().neg_().exp_()
393
- d1.square_().neg_().exp_()
394
- d0 = d0.sum(1)
395
- d1 = d1.sum(1)
396
- torch.divide(d0, d1, out=dens_ratio[ic0:ic1])
342
+ if q1 == n:
343
+ i1 = n
344
+ else:
345
+ i1 = q1 + torch.searchsorted(X0[q1:], x1, side="right")
346
+ i1.clamp_(max=n)
347
+ i1 = int(i1.item())
348
+
349
+ dens0 = dens0_buf[: q1 - q0].zero_()
350
+ dens1 = dens1_buf[: q1 - q0].zero_()
351
+ for j0 in range(i0, i1, col_batch_size):
352
+ j1 = min(j0 + col_batch_size, i1)
353
+ nj = j1 - j0
354
+ nbuf = nq * nj
355
+
356
+ dist = sqeuc_cdist_known_norm(
357
+ X=Q,
358
+ Xnormsq=Xnormsq[q0:q1],
359
+ Y=X[j0:j1],
360
+ Ynormsq=Xnormsq[j0:j1],
361
+ out=dist_buf[:nbuf].view(nq, nj),
362
+ )
363
+ dist.masked_fill_(dist > radius_sq, torch.inf)
364
+ gauss0 = torch.addcmul(
365
+ n0, k0, dist, out=dist_buf2[:nbuf].view(nq, nj)
366
+ ).exp_()
367
+ gauss1 = torch.addcmul(
368
+ n1, k1, dist, out=dist_buf3[:nbuf].view(nq, nj)
369
+ ).exp_()
370
+ dens0 += gauss0.sum(dim=1)
371
+ dens1 += gauss1.sum(dim=1)
372
+ dens[q0:q1] = dens0.div_(dens1)
397
373
 
398
- return dens_ratio.nan_to_num_()
374
+ dens = dens[torch.argsort(order)]
375
+ return dens.numpy(force=True)
399
376
 
400
377
 
401
378
  def kdt_density(
@@ -449,27 +426,34 @@ def _kdtdens_job(i0):
449
426
  jj = sdm["j"]
450
427
  v_local = sdm["v"]
451
428
  v_regional = sdm["v"].copy()
452
- _gausskernel(v_local, sigma)
453
- _gausskernel(v_regional, sigma_regional)
429
+ dim = float(X.shape[1])
430
+ _gausskernel(v_local, sigma, dim)
431
+ _gausskernel(v_regional, sigma_regional, dim)
454
432
  coo_local = coo_array(
455
433
  (v_local, (ii, jj)), shape=(batch_kdt.n, kdtree.n), dtype=v_local.dtype
456
434
  )
457
435
  coo_regional = coo_array(
458
436
  (v_regional, (ii, jj)), shape=(batch_kdt.n, kdtree.n), dtype=v_local.dtype
459
437
  )
460
- dens_local = coo_local.sum(axis=1)
438
+ dens = coo_local.sum(axis=1)
461
439
  dens_regional = coo_regional.sum(axis=1)
462
- dens_local /= dens_regional
463
- return i0, i1, dens_local
440
+ dens /= dens_regional
441
+
442
+ return i0, i1, dens
443
+
444
+
445
+ _log_2pi = np.log(2.0 * np.pi)
464
446
 
465
447
 
466
448
  @numba.jit(
467
449
  parallel=False, nopython=True, boundscheck=False, nogil=True, error_model="numpy"
468
450
  )
469
- def _gausskernel(v: np.ndarray, sigma: float):
470
- const = 1.0 / (np.sqrt(2.0) * sigma)
451
+ def _gausskernel(v: np.ndarray, sigma: float, dim: float):
452
+ h = np.sqrt(2.0) * sigma
453
+ const = 1.0 / h
454
+ addconst = (dim / 2) * np.log(2.0 * np.pi * sigma * sigma)
471
455
  for i in range(v.size):
472
- v[i] = np.exp(-np.square(v[i] * const))
456
+ v[i] = np.exp(-np.square(v[i] * const) - addconst)
473
457
 
474
458
 
475
459
  def knn_density(
@@ -507,8 +491,6 @@ def density_peaks(
507
491
  kdtree=None,
508
492
  density=None,
509
493
  knn_k=None,
510
- use_knn=False,
511
- use_histograms=False,
512
494
  sigma_local=5.0,
513
495
  sigma_regional=None,
514
496
  outlier_neighbor_count: int | None = 10,
@@ -522,7 +504,9 @@ def density_peaks(
522
504
  border_search_neighbors=3,
523
505
  inlier_dims=slice(0, 2),
524
506
  leafsize=24,
507
+ density_strategy="sort",
525
508
  workers=-1,
509
+ device: torch.device = torch.device("cpu"),
526
510
  ):
527
511
  """Density peaks clustering as described by Rodriguez and Laio, but...
528
512
 
@@ -536,6 +520,7 @@ def density_peaks(
536
520
  radius_search = radius_search * np.sqrt(X.shape[1])
537
521
  dim_arange = np.arange(X.shape[1])
538
522
  inlier_dim_ixs = dim_arange[inlier_dims]
523
+
539
524
  if outlier_radius is not None:
540
525
  outlier_radius = outlier_radius * np.sqrt(inlier_dim_ixs.size)
541
526
  inliers, kdtree = kdtree_inliers(
@@ -550,7 +535,8 @@ def density_peaks(
550
535
  kdtree = KDTree(X)
551
536
 
552
537
  if density is None:
553
- if sigma_regional and not use_histograms:
538
+ if density_strategy == "kdt":
539
+ assert sigma_regional is not None
554
540
  density = kdt_density(
555
541
  kdtree,
556
542
  X,
@@ -558,24 +544,31 @@ def density_peaks(
558
544
  sigma_regional=sigma_regional,
559
545
  n_threads=workers,
560
546
  )
561
- elif use_knn:
562
- density = knn_density(
563
- kdtree, X, k=knn_k, distance_upper_bound=radius_search, workers=workers
547
+ elif density_strategy == "sort":
548
+ assert sigma_regional is not None
549
+ density = sort_density(
550
+ X,
551
+ sigma0=sigma_local,
552
+ sigma1=sigma_regional,
553
+ device=device,
554
+ show_progress=logger.isEnabledFor(DARTSORTVERBOSE),
564
555
  )
565
- elif use_histograms:
556
+ elif density_strategy == "hist":
566
557
  sigmas = [sigma_local]
567
558
  if sigma_regional is not None:
568
559
  sigmas.append(sigma_regional)
569
560
  density = get_smoothed_density_ratio(X, inliers=inliers, sigmas=sigmas)
570
- else:
561
+ elif density_strategy == "knn":
571
562
  density = knn_density(
572
563
  kdtree,
573
564
  X,
574
- k=n_neighbors_search,
565
+ k=knn_k,
575
566
  distance_upper_bound=radius_search,
576
567
  workers=workers,
577
568
  sigma=sigma_local,
578
569
  )
570
+ else:
571
+ raise ValueError(f"Unknown density strategy: {density_strategy}")
579
572
 
580
573
  nhdn = nearest_higher_density_neighbor(
581
574
  kdtree,
@@ -606,12 +599,7 @@ def density_peaks(
606
599
  if remove_clusters_smaller_than:
607
600
  labels = decrumb(labels, min_size=remove_clusters_smaller_than, in_place=True)
608
601
 
609
- return dict(
610
- density=density,
611
- nhdn=nhdn,
612
- labels=labels,
613
- kdtree=kdtree,
614
- )
602
+ return dict(density=density, nhdn=nhdn, labels=labels, kdtree=kdtree)
615
603
 
616
604
 
617
605
  def nearest_neighbor_assign(
@@ -21,8 +21,7 @@ from ..util.sparse_util import (
21
21
  logsumexp_coo,
22
22
  sparse_centroid_distsq,
23
23
  )
24
- from ..util.spiketorch import spawn_torch_rg
25
- from .density import guess_mode
24
+ from ..util.spiketorch import spawn_torch_rg, sqeuc_cdist_known_norm
26
25
 
27
26
  logger = get_logger(__name__)
28
27
 
@@ -57,20 +56,11 @@ def kmeanspp(
57
56
  elif kmeanspp_initial == "mean":
58
57
  closest = torch.cdist(X, X.mean(0, keepdim=True)).argmax()
59
58
  centroid_ixs[0] = closest.item()
60
- elif kmeanspp_initial == "mode":
61
- Xm = X
62
- if Xm.shape[1] > mode_dim:
63
- q = min(mode_dim + 10, *Xm.shape)
64
- u, s, v = torch.pca_lowrank(Xm, q=q, niter=7)
65
- Xm = u[:, :mode_dim].mul_(s[:mode_dim])
66
- cixs = guess_mode(Xm.numpy(force=True))
67
- cixs = torch.asarray(cixs, dtype=centroid_ixs.dtype, device=centroid_ixs.device)
68
- centroid_ixs.copy_(cixs)
69
59
  else:
70
60
  assert False
71
61
 
72
62
  if Xnormsq is None:
73
- Xnormsq = torch.square(X).sum(1)
63
+ Xnormsq = torch.linalg.vector_norm(X, dim=1).square_()
74
64
 
75
65
  dists_ = X.new_zeros(len(X))[:, None]
76
66
  dists = _sqeuc(X, Xnormsq[:, None], Y=X[centroid_ixs[0]][None], out=dists_)[:, 0]
@@ -119,12 +109,15 @@ def kmeanspp(
119
109
  if invalid.all():
120
110
  break
121
111
  p.masked_fill_(invalid, 0.0)
122
- centroid_ixs[j] = torch.multinomial(p, 1, generator=gen)
112
+ ci = torch.multinomial(p, 1, generator=gen)
113
+ centroid_ixs[j] = ci
123
114
 
124
- # newdists = torch.subtract(X, X[centroid_ixs[j]], out=diff_buffer).square_()
125
- # newdists = torch.sum(newdists, dim=1, out=p)
126
- newdists = _sqeuc(
127
- X, Xnormsq[:, None], X[centroid_ixs[j]][None], out=p[:, None]
115
+ newdists = sqeuc_cdist_known_norm(
116
+ X,
117
+ Xnormsq,
118
+ X[ci],
119
+ Xnormsq[ci],
120
+ out=p[:, None],
128
121
  )[:, 0]
129
122
  if not skip_assignment:
130
123
  closer = newdists < dists
@@ -142,6 +135,8 @@ def kmeanspp(
142
135
  else:
143
136
  phi = dists.mul_(weights).mean() / weights.mean()
144
137
 
138
+ dists.relu_()
139
+
145
140
  return centroid_ixs, assignments, dists, phi
146
141
 
147
142
 
@@ -378,7 +373,7 @@ def kmeans_inner(
378
373
  centroids = X[centroid_ixs]
379
374
 
380
375
  if X_normsq is None:
381
- X_normsq = torch.linalg.norm(X, dim=1).square_()
376
+ X_normsq = torch.linalg.vector_norm(X, dim=1).square_()
382
377
  assert X_normsq is not None
383
378
  X_normsq = X_normsq[:, None]
384
379
  dists_out = X.new_empty((X.shape[0], centroids.shape[0]))
@@ -482,15 +477,6 @@ def _sqeuc(X: Tensor, Xnormsq: Tensor, Y: Tensor, out: Tensor):
482
477
  return out
483
478
 
484
479
 
485
- @torch.jit.script
486
- def _sqeuc_ysqknown(
487
- X: Tensor, Xnormsq: Tensor, Y: Tensor, Ynormsq: Tensor, out: Tensor
488
- ):
489
- torch.addmm(Xnormsq, X, Y.transpose(0, 1), alpha=-2.0, out=out)
490
- out.add_(Ynormsq)
491
- return out
492
-
493
-
494
480
  @torch.jit.script
495
481
  def _kmeanspp_simple_loop(
496
482
  *,
@@ -505,7 +491,6 @@ def _kmeanspp_simple_loop(
505
491
  ):
506
492
  buf = torch.empty_like(p)
507
493
  buf_ = buf[:, None]
508
- Xnormsq = Xnormsq[:, None]
509
494
  closer = X.new_zeros(X.shape[:1], dtype=torch.bool)
510
495
  cent = torch.zeros_like(X[:1])
511
496
  cnormsq = torch.zeros_like(Xnormsq[:1])
@@ -521,7 +506,7 @@ def _kmeanspp_simple_loop(
521
506
 
522
507
  torch.index_select(X, dim=0, index=cj, out=cent)
523
508
  torch.index_select(Xnormsq, dim=0, index=cj, out=cnormsq)
524
- newdists = _sqeuc_ysqknown(X, Xnormsq, cent, cnormsq, out=buf_).view(-1)
509
+ newdists = sqeuc_cdist_known_norm(X, Xnormsq, cent, cnormsq, out=buf_).view(-1)
525
510
 
526
511
  torch.lt(newdists, dists, out=closer)
527
512
  assignments.masked_fill_(closer, j)
@@ -541,7 +526,6 @@ def _kmeanspp_noassign_loop(
541
526
  ):
542
527
  buf = torch.empty_like(p)
543
528
  buf_ = buf[:, None]
544
- Xnormsq = Xnormsq[:, None]
545
529
  cent = torch.zeros_like(X[:1])
546
530
  cnormsq = torch.zeros_like(Xnormsq[:1])
547
531
  for j in range(1, centroid_ixs.shape[0]):
@@ -556,7 +540,7 @@ def _kmeanspp_noassign_loop(
556
540
 
557
541
  torch.index_select(X, dim=0, index=cj, out=cent)
558
542
  torch.index_select(Xnormsq, dim=0, index=cj, out=cnormsq)
559
- newdists = _sqeuc_ysqknown(X, Xnormsq, cent, cnormsq, out=buf_).view(-1)
543
+ newdists = sqeuc_cdist_known_norm(X, Xnormsq, cent, cnormsq, out=buf_).view(-1)
560
544
 
561
545
  torch.minimum(dists, newdists, out=dists)
562
546