ergodic 0.3.2__tar.gz → 0.4.0__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 (424) hide show
  1. {ergodic-0.3.2 → ergodic-0.4.0}/CHANGELOG.md +25 -1
  2. {ergodic-0.3.2 → ergodic-0.4.0}/PKG-INFO +1 -1
  3. ergodic-0.4.0/docsite/pages/continuous-treatments.md +159 -0
  4. ergodic-0.4.0/docsite/pages/decisions.md +150 -0
  5. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/estimate-average-effect.md +2 -0
  6. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/install.md +1 -1
  7. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/site.toml +4 -1
  8. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/__init__.py +9 -1
  9. ergodic-0.4.0/src/ergodic/decision/__init__.py +7 -0
  10. ergodic-0.4.0/src/ergodic/decision/_categorical.py +229 -0
  11. ergodic-0.4.0/src/ergodic/decision/_pipeline.py +692 -0
  12. ergodic-0.4.0/src/ergodic/decision/_ranking.py +154 -0
  13. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/__init__.py +11 -1
  14. ergodic-0.4.0/src/ergodic/inference/_contrast.py +83 -0
  15. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_estimate.py +36 -0
  16. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/__init__.py +4 -0
  17. ergodic-0.4.0/src/ergodic/inference/estimators/_continuous.py +385 -0
  18. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_dml.py +114 -4
  19. ergodic-0.4.0/tests/test_continuous_effects.py +415 -0
  20. ergodic-0.4.0/tests/test_decision_pipeline.py +262 -0
  21. ergodic-0.4.0/tests/test_uplift_ranking.py +471 -0
  22. {ergodic-0.3.2 → ergodic-0.4.0}/.gitignore +0 -0
  23. {ergodic-0.3.2 → ergodic-0.4.0}/CONTRIBUTING.md +0 -0
  24. {ergodic-0.3.2 → ergodic-0.4.0}/LICENSE +0 -0
  25. {ergodic-0.3.2 → ergodic-0.4.0}/NOTICE +0 -0
  26. {ergodic-0.3.2 → ergodic-0.4.0}/README.md +0 -0
  27. {ergodic-0.3.2 → ergodic-0.4.0}/SECURITY.md +0 -0
  28. {ergodic-0.3.2 → ergodic-0.4.0}/docs/case-studies/01-birthweight.py +0 -0
  29. {ergodic-0.3.2 → ergodic-0.4.0}/docs/case-studies/02-berkeley.py +0 -0
  30. {ergodic-0.3.2 → ergodic-0.4.0}/docs/case-studies/03-pension401k.py +0 -0
  31. {ergodic-0.3.2 → ergodic-0.4.0}/docs/case-studies/04-synthetic-control.py +0 -0
  32. {ergodic-0.3.2 → ergodic-0.4.0}/docs/case-studies/05-discovery.py +0 -0
  33. {ergodic-0.3.2 → ergodic-0.4.0}/docs/case-studies/06-clv.py +0 -0
  34. {ergodic-0.3.2 → ergodic-0.4.0}/docs/case-studies/07-churn.py +0 -0
  35. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/01-why-causality.py +0 -0
  36. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/02-good-and-bad-controls.py +0 -0
  37. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/03-estimating-effects.py +0 -0
  38. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/04-heterogeneous-effects.py +0 -0
  39. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/05-hidden-confounding.py +0 -0
  40. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/06-discovering-structure.py +0 -0
  41. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/07-quasi-experiments.py +0 -0
  42. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/08-process-mining.py +0 -0
  43. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/09-process-intelligence.py +0 -0
  44. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/10-decision-points.py +0 -0
  45. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/11-forecasting.py +0 -0
  46. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/12-structural-causal-models.py +0 -0
  47. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/13-hidden-regimes.py +0 -0
  48. {ergodic-0.3.2 → ergodic-0.4.0}/docs/learn/14-causal-forests-and-policy-trees.py +0 -0
  49. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/docs.css +0 -0
  50. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/docs.js +0 -0
  51. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/ergodic-mark-dark.svg +0 -0
  52. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/ergodic-mark-light.svg +0 -0
  53. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/fonts/OFL-Geist-Mono.txt +0 -0
  54. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/fonts/OFL-Geist.txt +0 -0
  55. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/fonts/geist-latin-ext-wght-normal.woff2 +0 -0
  56. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/fonts/geist-latin-wght-normal.woff2 +0 -0
  57. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/fonts/geist-mono-latin-ext-wght-normal.woff2 +0 -0
  58. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/fonts/geist-mono-latin-wght-normal.woff2 +0 -0
  59. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/site.css +0 -0
  60. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/assets/tokens.css +0 -0
  61. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/adjustment-sets.md +0 -0
  62. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/assumptions.md +0 -0
  63. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/backtest-a-forecast.md +0 -0
  64. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/bootstrap-stability.md +0 -0
  65. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/certify-and-calibrate.md +0 -0
  66. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/changelog.md +0 -0
  67. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/choose-a-method.md +0 -0
  68. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/compatibility.md +0 -0
  69. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/conformance-and-decisions.md +0 -0
  70. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/decode-a-graph.md +0 -0
  71. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/difference-in-differences.md +0 -0
  72. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/discover-a-graph.md +0 -0
  73. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/domain-knowledge.md +0 -0
  74. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/edge-probabilities.md +0 -0
  75. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/evaluate-a-graph.md +0 -0
  76. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/fit-an-scm.md +0 -0
  77. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/forecast-with-drivers.md +0 -0
  78. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/forecast-with-welt.md +0 -0
  79. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/glossary.md +0 -0
  80. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/graphs-and-marks.md +0 -0
  81. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/heterogeneous-effects.md +0 -0
  82. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/identify-an-effect.md +0 -0
  83. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/interventions-and-counterfactuals.md +0 -0
  84. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/map-a-process.md +0 -0
  85. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/mediation.md +0 -0
  86. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/reconcile-a-hierarchy.md +0 -0
  87. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/recover-hidden-regimes.md +0 -0
  88. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/refute-an-estimate.md +0 -0
  89. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/run-scenarios.md +0 -0
  90. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/start.md +0 -0
  91. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/synthetic-control.md +0 -0
  92. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/target-a-policy.md +0 -0
  93. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/test-a-graph.md +0 -0
  94. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/time-series-discovery.md +0 -0
  95. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/pages/welt.md +0 -0
  96. {ergodic-0.3.2 → ergodic-0.4.0}/docsite/provenance.json +0 -0
  97. {ergodic-0.3.2 → ergodic-0.4.0}/examples/README.md +0 -0
  98. {ergodic-0.3.2 → ergodic-0.4.0}/examples/adjustment_birthweight.py +0 -0
  99. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/01_cate_basics.py +0 -0
  100. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/02_meta_learners_compared.py +0 -0
  101. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/03_is_there_heterogeneity.py +0 -0
  102. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/04_where_is_the_effect_higher.py +0 -0
  103. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/05_targeting_with_rate.py +0 -0
  104. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/06_per_unit_uncertainty.py +0 -0
  105. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/07_cate_vs_ite.py +0 -0
  106. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/08_identification_gating.py +0 -0
  107. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/09_case_study_uplift_end_to_end.py +0 -0
  108. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/10_causal_forest.py +0 -0
  109. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/11_policy_trees.py +0 -0
  110. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/12_model_selection.py +0 -0
  111. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/13_case_study_phase2.py +0 -0
  112. {ergodic-0.3.2 → ergodic-0.4.0}/examples/cate/README.md +0 -0
  113. {ergodic-0.3.2 → ergodic-0.4.0}/examples/confounding_berkeley.py +0 -0
  114. {ergodic-0.3.2 → ergodic-0.4.0}/examples/discovery_asia.py +0 -0
  115. {ergodic-0.3.2 → ergodic-0.4.0}/examples/double_ml_pension.py +0 -0
  116. {ergodic-0.3.2 → ergodic-0.4.0}/examples/hidden_regimes.py +0 -0
  117. {ergodic-0.3.2 → ergodic-0.4.0}/examples/latent_switch.py +0 -0
  118. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/01_three_rungs.py +0 -0
  119. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/02_honesty_and_tipping_point.py +0 -0
  120. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/03_causal_price_optimization.py +0 -0
  121. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/04_counterfactual_recourse.py +0 -0
  122. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/05_counterfactual_fairness.py +0 -0
  123. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/06_causal_attribution.py +0 -0
  124. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/07_but_for_accounting.py +0 -0
  125. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/08_off_policy_evaluation.py +0 -0
  126. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/09_root_cause_attribution.py +0 -0
  127. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/10_frontdoor_unmeasured_confounder.py +0 -0
  128. {ergodic-0.3.2 → ergodic-0.4.0}/examples/scm/README.md +0 -0
  129. {ergodic-0.3.2 → ergodic-0.4.0}/examples/synthetic_control_smoking.py +0 -0
  130. {ergodic-0.3.2 → ergodic-0.4.0}/public/ergodic_logomark_dark.svg +0 -0
  131. {ergodic-0.3.2 → ergodic-0.4.0}/public/ergodic_logomark_light.svg +0 -0
  132. {ergodic-0.3.2 → ergodic-0.4.0}/pyproject.toml +0 -0
  133. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/audits/qualify_aipw.py +0 -0
  134. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/audits/qualify_ci_discovery.py +0 -0
  135. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/audits/qualify_gcm.py +0 -0
  136. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/__init__.py +0 -0
  137. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/__main__.py +0 -0
  138. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/build.py +0 -0
  139. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/config.py +0 -0
  140. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/examples.py +0 -0
  141. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/graphite.py +0 -0
  142. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/links.py +0 -0
  143. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/notebooks.py +0 -0
  144. {ergodic-0.3.2 → ergodic-0.4.0}/scripts/docsite/reference.py +0 -0
  145. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/__init__.py +0 -0
  146. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_dataset.py +0 -0
  147. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_eventlog.py +0 -0
  148. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_frame.py +0 -0
  149. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_hierarchy.py +0 -0
  150. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_hts.py +0 -0
  151. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_panel.py +0 -0
  152. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_schema.py +0 -0
  153. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_tabular.py +0 -0
  154. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/data/_timeseries.py +0 -0
  155. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/__init__.py +0 -0
  156. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/_bayesnet.py +0 -0
  157. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/_bunch.py +0 -0
  158. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/_bundled.py +0 -0
  159. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/_tables.py +0 -0
  160. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/data/birthweight.csv +0 -0
  161. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/data/germany.csv +0 -0
  162. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/data/pension401k.csv +0 -0
  163. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/datasets/data/smoking.csv +0 -0
  164. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/__init__.py +0 -0
  165. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_average.py +0 -0
  166. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_bootstrap.py +0 -0
  167. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_ci.py +0 -0
  168. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_decode.py +0 -0
  169. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_discover.py +0 -0
  170. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_encode.py +0 -0
  171. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_evaluate.py +0 -0
  172. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_exact.py +0 -0
  173. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_granger.py +0 -0
  174. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_graphtest.py +0 -0
  175. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_lagged.py +0 -0
  176. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_mcmc.py +0 -0
  177. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_posterior.py +0 -0
  178. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_probabilities.py +0 -0
  179. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_registry.py +0 -0
  180. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_result.py +0 -0
  181. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_score.py +0 -0
  182. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_stats.py +0 -0
  183. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_target.py +0 -0
  184. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/_tsresult.py +0 -0
  185. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/__init__.py +0 -0
  186. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_astar.py +0 -0
  187. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_boss.py +0 -0
  188. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_fci.py +0 -0
  189. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_ges.py +0 -0
  190. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_glasso.py +0 -0
  191. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_grasp.py +0 -0
  192. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_hc.py +0 -0
  193. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_lingam.py +0 -0
  194. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_mmpc.py +0 -0
  195. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_pc.py +0 -0
  196. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_pcmci.py +0 -0
  197. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_rfci.py +0 -0
  198. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/algorithms/_skeleton.py +0 -0
  199. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/ci/__init__.py +0 -0
  200. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/ci/_discrete.py +0 -0
  201. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/ci/_gcm.py +0 -0
  202. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/ci/_linear.py +0 -0
  203. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/ci/_mixed.py +0 -0
  204. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/scores/__init__.py +0 -0
  205. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/scores/_bdeu.py +0 -0
  206. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/scores/_bge.py +0 -0
  207. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/scores/_bic.py +0 -0
  208. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/scores/_discrete_bic.py +0 -0
  209. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/discovery/scores/_mixed_bic.py +0 -0
  210. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/__init__.py +0 -0
  211. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_attribution.py +0 -0
  212. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_backtest.py +0 -0
  213. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_baselines.py +0 -0
  214. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_facade.py +0 -0
  215. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_fit.py +0 -0
  216. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_mechanism.py +0 -0
  217. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_metrics.py +0 -0
  218. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_protocol.py +0 -0
  219. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_quantiles.py +0 -0
  220. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_reconcile.py +0 -0
  221. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_result.py +0 -0
  222. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_scalers.py +0 -0
  223. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_schedule.py +0 -0
  224. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_staged.py +0 -0
  225. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/_structure.py +0 -0
  226. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/mechanisms/__init__.py +0 -0
  227. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/mechanisms/_bass.py +0 -0
  228. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/mechanisms/_ou.py +0 -0
  229. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/forecast/mechanisms/_structural.py +0 -0
  230. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/__init__.py +0 -0
  231. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/_graph.py +0 -0
  232. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/_io.py +0 -0
  233. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/_marks.py +0 -0
  234. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/_meek.py +0 -0
  235. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/_pag.py +0 -0
  236. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/_primitives.py +0 -0
  237. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/_separation.py +0 -0
  238. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/graph/_surgery.py +0 -0
  239. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/__init__.py +0 -0
  240. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_criteria.py +0 -0
  241. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_distribution.py +0 -0
  242. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_evaluate.py +0 -0
  243. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_functional.py +0 -0
  244. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_id.py +0 -0
  245. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_ida.py +0 -0
  246. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_mediation.py +0 -0
  247. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_recognize.py +0 -0
  248. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/identification/_record.py +0 -0
  249. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_crossfit.py +0 -0
  250. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_design.py +0 -0
  251. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_estimator.py +0 -0
  252. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_learner.py +0 -0
  253. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_overlap.py +0 -0
  254. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_perunit.py +0 -0
  255. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_refute.py +0 -0
  256. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_registry.py +0 -0
  257. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_result.py +0 -0
  258. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_scores.py +0 -0
  259. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_select.py +0 -0
  260. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_stats.py +0 -0
  261. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/_target.py +0 -0
  262. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_did.py +0 -0
  263. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_doubly_robust.py +0 -0
  264. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_forest.py +0 -0
  265. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_frontdoor.py +0 -0
  266. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_its.py +0 -0
  267. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_iv.py +0 -0
  268. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_mediation.py +0 -0
  269. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_meta.py +0 -0
  270. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_null.py +0 -0
  271. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_plugin.py +0 -0
  272. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_regression.py +0 -0
  273. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_synthetic_control.py +0 -0
  274. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/estimators/_weighting.py +0 -0
  275. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/__init__.py +0 -0
  276. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/_blp.py +0 -0
  277. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/_calibration.py +0 -0
  278. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/_clan.py +0 -0
  279. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/_gates.py +0 -0
  280. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/_linalg.py +0 -0
  281. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/_policy.py +0 -0
  282. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/_rate.py +0 -0
  283. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/explain/_record.py +0 -0
  284. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/learners/__init__.py +0 -0
  285. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/learners/_linear.py +0 -0
  286. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/learners/_neighbors.py +0 -0
  287. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/inference/learners/_pymc.py +0 -0
  288. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/knowledge.py +0 -0
  289. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/__init__.py +0 -0
  290. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/_engine.py +0 -0
  291. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/_facade.py +0 -0
  292. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/_identify.py +0 -0
  293. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/_likelihood.py +0 -0
  294. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/_marginalize.py +0 -0
  295. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/_result.py +0 -0
  296. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/likelihoods/__init__.py +0 -0
  297. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/likelihoods/_ar.py +0 -0
  298. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/latent/likelihoods/_spectral.py +0 -0
  299. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/__init__.py +0 -0
  300. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_case_table.py +0 -0
  301. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_cases.py +0 -0
  302. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_conformance.py +0 -0
  303. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_decisions.py +0 -0
  304. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_filters.py +0 -0
  305. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_map.py +0 -0
  306. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_mine.py +0 -0
  307. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_net.py +0 -0
  308. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_panel.py +0 -0
  309. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_performance.py +0 -0
  310. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_soundness.py +0 -0
  311. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_tree.py +0 -0
  312. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/process/_variants.py +0 -0
  313. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/py.typed +0 -0
  314. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/__init__.py +0 -0
  315. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/_bounds.py +0 -0
  316. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/_compose.py +0 -0
  317. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/_district.py +0 -0
  318. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/_mechanism.py +0 -0
  319. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/_noise.py +0 -0
  320. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/_result.py +0 -0
  321. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/_sampling.py +0 -0
  322. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/scm/_scm.py +0 -0
  323. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/welt/__init__.py +0 -0
  324. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/welt/_client.py +0 -0
  325. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/welt/_data.py +0 -0
  326. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/welt/_discovery.py +0 -0
  327. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/welt/_forecast.py +0 -0
  328. {ergodic-0.3.2 → ergodic-0.4.0}/src/ergodic/welt/_learner.py +0 -0
  329. {ergodic-0.3.2 → ergodic-0.4.0}/tests/conftest.py +0 -0
  330. {ergodic-0.3.2 → ergodic-0.4.0}/tests/data/exp0626_decoder_parity.json +0 -0
  331. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_average_effect.py +0 -0
  332. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_bayes.py +0 -0
  333. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_case_table.py +0 -0
  334. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_cate.py +0 -0
  335. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_cate_forest.py +0 -0
  336. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_cate_phase2_stress.py +0 -0
  337. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_cate_policy.py +0 -0
  338. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_cate_select.py +0 -0
  339. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_cate_stress.py +0 -0
  340. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_conformance.py +0 -0
  341. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_data.py +0 -0
  342. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_datasets.py +0 -0
  343. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_decisions.py +0 -0
  344. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_decode.py +0 -0
  345. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_decode_knowledge.py +0 -0
  346. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_did.py +0 -0
  347. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery.py +0 -0
  348. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_bootstrap.py +0 -0
  349. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_boss.py +0 -0
  350. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_ci.py +0 -0
  351. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_fci.py +0 -0
  352. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_ges.py +0 -0
  353. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_glasso.py +0 -0
  354. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_graphtest.py +0 -0
  355. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_grasp.py +0 -0
  356. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_hc.py +0 -0
  357. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_lingam.py +0 -0
  358. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_mmpc.py +0 -0
  359. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_pc.py +0 -0
  360. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_pcmci.py +0 -0
  361. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_rfci.py +0 -0
  362. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_discovery_score.py +0 -0
  363. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_docs_welt_forecast_page.py +0 -0
  364. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_docs_welt_page.py +0 -0
  365. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_docsite.py +0 -0
  366. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_estimators_e2.py +0 -0
  367. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_evaluate_graph.py +0 -0
  368. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_backtest.py +0 -0
  369. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_backtest_mcmc.py +0 -0
  370. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_core.py +0 -0
  371. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_explain.py +0 -0
  372. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_facade.py +0 -0
  373. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_mechanisms.py +0 -0
  374. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_metrics.py +0 -0
  375. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_quantiles.py +0 -0
  376. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_reconcile.py +0 -0
  377. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_staged.py +0 -0
  378. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_structure.py +0 -0
  379. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_forecast_zoo.py +0 -0
  380. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_granger.py +0 -0
  381. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_graph.py +0 -0
  382. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_graph_format.py +0 -0
  383. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_graph_posterior.py +0 -0
  384. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_graph_primitives.py +0 -0
  385. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_id_algorithm.py +0 -0
  386. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_id_integration.py +0 -0
  387. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_identification.py +0 -0
  388. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_identification_classes.py +0 -0
  389. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_inductive_miner.py +0 -0
  390. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_inference.py +0 -0
  391. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_inference_guardrails.py +0 -0
  392. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_its.py +0 -0
  393. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_knowledge.py +0 -0
  394. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_knowledge_soft.py +0 -0
  395. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_kpi_panel.py +0 -0
  396. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_latent_calibrate.py +0 -0
  397. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_latent_certify.py +0 -0
  398. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_latent_facade.py +0 -0
  399. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_latent_golden.py +0 -0
  400. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_latent_identify.py +0 -0
  401. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_latent_marginalize.py +0 -0
  402. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_mediation.py +0 -0
  403. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_native_learners.py +0 -0
  404. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_pag.py +0 -0
  405. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_plugin.py +0 -0
  406. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_process_filters.py +0 -0
  407. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_process_map.py +0 -0
  408. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_refute.py +0 -0
  409. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_release_cate_regressions.py +0 -0
  410. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_reprs.py +0 -0
  411. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_review_fixes.py +0 -0
  412. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_scm.py +0 -0
  413. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_scm_pymc.py +0 -0
  414. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_score_posterior.py +0 -0
  415. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_sharp_edges.py +0 -0
  416. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_sklearn_interop.py +0 -0
  417. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_smoke.py +0 -0
  418. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_synthetic_control.py +0 -0
  419. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_temporal_wiring.py +0 -0
  420. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_user_facing_wording.py +0 -0
  421. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_welt.py +0 -0
  422. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_welt_forecast.py +0 -0
  423. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_welt_isolation.py +0 -0
  424. {ergodic-0.3.2 → ergodic-0.4.0}/tests/test_workflow_net.py +0 -0
@@ -7,6 +7,29 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
7
7
 
8
8
  ## [Unreleased]
9
9
 
10
+ ## [0.4.0] - 2026-11-18
11
+
12
+ ### Added
13
+ - Explicit decision v2: per-variable plugin/S/DR ranking with `plugin` (fold-averaged
14
+ difference of the cross-fit arm outcome models) as the binary-treatment,
15
+ binary-target default, DR population estimates, binary event/complement
16
+ coherence, separate learner slots and conservative independent randomized
17
+ validation with disjoint unit IDs; `auto` keeps `plugin` unless S or DR is
18
+ proven better on validation. v1 remains the implicit protocol; these semantics
19
+ do not establish a universal quality winner.
20
+ - `ergodic.decision.decide`: graph-identified automatic per-variable contrasts,
21
+ multiarm doubly robust estimation and ranked population/query-row effects for
22
+ continuous, binary and multicategory treatments and targets.
23
+ - Explicit numerical treatment contrasts through `estimate_contrast`,
24
+ `DML.contrast` and `DML.effects`, without changing the existing unit-slope API.
25
+ - `ContinuousDR`: cross-fitted kernel dose contrasts, pluggable treatment density,
26
+ support diagnostics and cached query-row point estimates. The new continuous
27
+ treatment guide includes an executable example and the inference limitations.
28
+
29
+ ### Fixed
30
+ - DML rejects non-finite inputs, invalid fold counts and absent residual treatment
31
+ variation. A failed refit cannot return stale estimates.
32
+
10
33
  ## [0.3.2] - 2026-10-24
11
34
 
12
35
  ### Fixed
@@ -1006,7 +1029,8 @@ inference, discovery, process, SCM, and forecast pillars.
1006
1029
  silently with single-graph semantics; it now dispatches on the face, and
1007
1030
  mixed graphs raise instead of being misread.
1008
1031
 
1009
- [Unreleased]: https://github.com/ergodic-ai/ergodic/compare/v0.3.2...HEAD
1032
+ [Unreleased]: https://github.com/ergodic-ai/ergodic/compare/v0.4.0...HEAD
1033
+ [0.4.0]: https://github.com/ergodic-ai/ergodic/compare/v0.3.2...v0.4.0
1010
1034
  [0.3.2]: https://github.com/ergodic-ai/ergodic/compare/v0.3.1...v0.3.2
1011
1035
  [0.3.1]: https://github.com/ergodic-ai/ergodic/compare/v0.3.0...v0.3.1
1012
1036
  [0.3.0]: https://github.com/ergodic-ai/ergodic/compare/v0.2.1...v0.3.0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: ergodic
3
- Version: 0.3.2
3
+ Version: 0.4.0
4
4
  Summary: Causal AI: causal discovery, causal inference, and process intelligence.
5
5
  Project-URL: Homepage, https://docs.ergodic.dev
6
6
  Project-URL: Documentation, https://docs.ergodic.dev
@@ -0,0 +1,159 @@
1
+ # Estimate effects of numerical changes
2
+
3
+ Compare two values of a numerical treatment, such as a dose or a price. Choose a
4
+ partially linear model for a constant effect per unit, or continuous DR for a
5
+ nonlinear dose response. Both methods identify the adjustment set from your graph.
6
+
7
+ !!! need "You'll need"
8
+ - ergodic 0.4.0 or later.
9
+ - Independent rows with a numerical treatment, a numerical or binary outcome,
10
+ and an adjustment-identifiable graph.
11
+ - Enough observations near both treatment values, across the relevant covariates.
12
+
13
+ ## Run the program
14
+
15
+ This simulation has a nonlinear response and an effect that varies with `z`.
16
+ Changing the dose from -0.5 to 1 has a true conditional effect of `3 + 0.75*z`.
17
+ The population effect is 3 because `z` is centered at zero in the generating model.
18
+
19
+ ```python file="continuous_treatments.py"
20
+ import numpy as np
21
+ from ergodic import dag, estimate_contrast, identify_effect, tabular
22
+ from ergodic.inference import ContinuousDR
23
+
24
+ rng = np.random.default_rng(17)
25
+ n = 6000
26
+ z = rng.uniform(-1, 1, n)
27
+ dose = 0.3 * z + rng.normal(size=n)
28
+ response = 1.5 * dose + dose**2 + 0.5 * dose * z + z + rng.normal(0, 0.15, n)
29
+ context = tabular({"dose": dose, "z": z, "response": response})
30
+ graph = dag(["z -> dose", "z -> response", "dose -> response"])
31
+
32
+ linear = estimate_contrast(
33
+ graph, context, "dose", "response",
34
+ baseline=-0.5, value=1, method="partially_linear", random_state=0,
35
+ )
36
+ print(f"partially linear: {linear.value:.2f}")
37
+
38
+ identified = identify_effect(graph, "dose", "response")
39
+ fitted = ContinuousDR(random_state=0).fit(identified, context)
40
+ average = fitted.contrast(baseline=-0.5, value=1)
41
+ print(f"continuous DR: {average.value:.2f}")
42
+ print(f"bandwidth: {average.diagnostics['bandwidth']:.3f}")
43
+
44
+ # Query rows need only the adjustment covariates, not dose or response.
45
+ rows = tabular({"z": [-0.5, 0.0, 0.5]})
46
+ effects = fitted.effects(rows, baseline=-0.5, value=1)
47
+ print("conditional effects:", np.round(effects.values, 2))
48
+ print("true conditional effects:", np.array([2.625, 3.0, 3.375]))
49
+ print("query interval available:", effects.std_error is not None)
50
+ ```
51
+
52
+ ```output from="continuous_treatments.py"
53
+ partially linear: 2.24
54
+ continuous DR: 2.99
55
+ bandwidth: 0.116
56
+ conditional effects: [2.58 2.99 3.4 ]
57
+ true conditional effects: [2.625 3. 3.375]
58
+ query interval available: False
59
+ ```
60
+
61
+ The partially linear fit summarizes a slope, so multiplying it by the dose change
62
+ doesn't recover this nonlinear contrast. Continuous DR estimates the two dose
63
+ responses separately. This example illustrates the distinction; it isn't a
64
+ benchmark showing one method always wins.
65
+
66
+ ## Choose the method for each variable
67
+
68
+ `estimate_contrast(graph, context, treatment, outcome, baseline=..., value=...,
69
+ method=...)` handles one treatment variable per call. Choose independently for
70
+ another variable; there is no global method selection.
71
+
72
+ | Method | Assumption | Population result | Query-row result |
73
+ |---|---|---|---|
74
+ | `partially_linear` | Homogeneous linear treatment effect, flexible covariate response | Slope times the dose change, with scaled uncertainty | The same effect per row, with shared slope uncertainty |
75
+ | `continuous_dr` | Smooth dose response and conditional treatment density | Difference of paired kernel DR scores, with a pointwise asymptotic interval | Conditional expected-effect predictions; no per-row interval |
76
+
77
+ For repeated queries, use `DML(...).fit(identified, context)` or
78
+ `ContinuousDR(...).fit(identified, context)`. Both expose
79
+ `contrast(baseline=..., value=...)` and `effects(rows, baseline=..., value=...)`.
80
+ Continuous DR reuses the final model for each dose pair across query batches.
81
+ Refitting the context clears its caches.
82
+
83
+ The existing `DML.effect()` and `DML.cate()` still report a **unit slope**.
84
+ `DML.contrast()` scales that slope by the dose change; reversing the contrast
85
+ reverses its sign, not its standard error. Averaging its row effects preserves the
86
+ shared uncertainty. Duplicating query rows cannot make the slope more precise.
87
+
88
+ ## Numerical and classification outcomes
89
+
90
+ For a numerical outcome, the effect is a change in its expected value. For a
91
+ binary outcome, encode the desired event as 1 and the other event as 0: the effect
92
+ is a probability difference. An effect of 0.08 means eight percentage points.
93
+ Explicit contrasts preserve that meaning even when categorical labels are stored
94
+ in reversed order. DML's legacy slope API retains its original category coding;
95
+ new contrasts use raw numerical treatment values and outcome event 1.
96
+
97
+ Multiclass outcomes, joint changes to several treatments, and categorical
98
+ multi-arm treatments aren't implemented by these methods. Query effects are
99
+ conditional averages, not individual structural counterfactuals. Heterogeneity
100
+ is modeled over the graph-selected adjustment covariates.
101
+
102
+ ## Prediction and density models
103
+
104
+ Pass an outcome learner or an `outcome`/`final` learner map through
105
+ `ContinuousDR(learners=...)`. By default, the outcome learner is OLS for numerical
106
+ outcomes or logistic regression for binary outcomes. The final effect regressor
107
+ is OLS. More flexible learners can model richer responses; the final learner must
108
+ always predict numerical effect scores, even for a classification outcome.
109
+
110
+ The separate `density=` argument requires a `TreatmentDensity` with
111
+ `fit(X, treatment)` and `pdf(X, treatment)`. The latter returns one conditional
112
+ density per row in inverse treatment units. A regressor that only predicts the
113
+ mean treatment is not a density estimator. The default `GaussianTreatmentDensity`
114
+ assumes a linear conditional mean and homoskedastic Gaussian residuals. Replace
115
+ it when that assumption is inappropriate. Each fold fits fresh nuisance models.
116
+ No hosted model is called by default, and this guide doesn't qualify Kumo or any
117
+ other remote learner for causal estimation.
118
+
119
+ ## Bandwidth, support and uncertainty
120
+
121
+ Continuous DR uses the cross-fitted kernel score from
122
+ [Colangelo and Lee, equation (1)](https://arxiv.org/html/2004.03036v7):
123
+
124
+ ```text
125
+ m(a, X) + K_h(T-a) / f(a|X) * (Y-m(a,X))
126
+ ```
127
+
128
+ Both nuisances are evaluated at the requested dose `a`, using fits that exclude
129
+ the scored row's fold. Contrast scores are differenced before averaging and
130
+ computing the standard error, retaining dependence between the dose estimates.
131
+
132
+ The Epanechnikov bandwidth defaults to `sd(T) * n**(-1/4)` using the context sample
133
+ standard deviation. An explicit positive `bandwidth` is in treatment units. This
134
+ is an undersmoothing heuristic, not a tuned optimum. Results record bandwidth,
135
+ local effective sample sizes, minimum estimated density and procedure version.
136
+
137
+ Continuous DR refuses doses outside the observed range, within a bandwidth of
138
+ its boundaries, with density below `min_density` (default `1e-6` in inverse
139
+ treatment units), or with effective local sample size below `min_effective_n`
140
+ (default 10). DML checks the observed range. These checks diagnose some support
141
+ failures; they don't prove positivity. Neither method silently clips effects or
142
+ weights. Numeric query covariates retain their values; categorical query
143
+ covariates use context encoding and reject unseen levels.
144
+
145
+ Double robustness for point doses is **asymptotic** as bandwidth shrinks. At a
146
+ finite bandwidth, smoothing bias remains. `average.interval()` is a pointwise
147
+ normal approximation without bias correction: it needs appropriate
148
+ undersmoothing and nuisance convergence rates, not merely one correct nuisance.
149
+ It doesn't account for graph selection or simultaneous comparisons. A large
150
+ user-selected bandwidth can invalidate the approximation.
151
+
152
+ `effects.interval()` raises because query-row uncertainty isn't available.
153
+ `effects.ate()` averages those predictions over the query rows without an
154
+ interval; it isn't the fitted population's DR estimate or a transportability
155
+ guarantee. Both methods rely on the graph's causal assumptions, which the data
156
+ alone cannot establish.
157
+
158
+ Continue with [average effects](estimate-average-effect.md) or
159
+ [heterogeneous effects](heterogeneous-effects.md).
@@ -0,0 +1,150 @@
1
+ # Rank interventions
2
+
3
+ Use a graph and context observations to compare actions on one target. The
4
+ intervention map names what can change. Defaults choose the reference and
5
+ alternative values from context; query rows only request personalised rankings.
6
+ This API is unreleased and requires the decision-pipeline source revision.
7
+
8
+ ## A three-category treatment
9
+
10
+ This synthetic example has three actions and a continuous target. The effect of
11
+ an action depends on the baseline variable `z`. The graph is supplied here;
12
+ Welt's hosted decision pipeline also runs discovery and records its analysis graph.
13
+
14
+ ```python file="rank_decisions.py"
15
+ import numpy as np
16
+ import pandas as pd
17
+ from ergodic import dag
18
+ from ergodic.decision import decide
19
+
20
+ rng = np.random.default_rng(8)
21
+ n = 4000
22
+ z = rng.uniform(-1, 1, n)
23
+ a = rng.integers(0, 3, n)
24
+ labels = np.array(["none", "small", "large"])[a]
25
+ y = z + a * (1 + .5 * z) + rng.normal(0, .2, n)
26
+ context = pd.DataFrame({"z": z, "action": labels, "outcome": y})
27
+ result = decide(
28
+ dag(["z -> outcome", "action -> outcome"]),
29
+ context,
30
+ target="outcome",
31
+ interventions={"action": {"baseline": "none", "values": ("small", "large")}},
32
+ rows=pd.DataFrame({"z": [-.5, .5]}),
33
+ seed=8,
34
+ )
35
+ for action in result["population"]:
36
+ print(action["value"], round(action["effect"], 2))
37
+ print("query best actions:", [row["interventions"][0]["value"] for row in result["individual"]])
38
+ print("skipped:", len(result["skipped"]))
39
+ ```
40
+
41
+ ```output from="rank_decisions.py"
42
+ large 2.0
43
+ small 1.01
44
+ query best actions: ['large', 'large']
45
+ skipped: 0
46
+ ```
47
+
48
+ ## Choose the target and treatment types
49
+
50
+ Binary and multicategory treatments use cross-fitted, multi-arm DR scores over
51
+ the full context population. Numerical treatments default to `partially_linear`,
52
+ a homogeneous slope assumption. Select `continuous_dr` per variable when a
53
+ nonlinear dose response is required; see [numerical contrasts](continuous-treatments.md).
54
+ Only the implemented method names are accepted.
55
+
56
+ Numerical targets report differences in their units. Binary and multicategory
57
+ targets report class-probability differences. Set `target_kind="categorical"`
58
+ and `target_class` to the event you want to optimise. Boolean and 0/1 binary
59
+ targets default to True/1. Integer-coded categories need an explicit categorical
60
+ kind: class identifiers do not define a numerical utility.
61
+
62
+ For categorical outcomes the true class contrasts sum to zero. We enforce this
63
+ identity by projecting joint row-aligned scores before calculating means and
64
+ standard errors, including their cross-class covariance. Query predictions use
65
+ the same linear projection. This is not clipping, calibration or a guarantee
66
+ that finite-sample effects lie in [-1, 1].
67
+
68
+ ## Defaults and interpretation
69
+
70
+ - Boolean/0–1 treatment: false to true.
71
+ - Categorical treatment: most frequent label to every other label; ties use stable label order.
72
+ - Numerical treatment: median to the 25th and 75th percentiles.
73
+
74
+ Every returned action records the actual baseline, destination, method and
75
+ adjustment set. `objective="minimize"` reverses the ranking direction. Rankings
76
+ compare one variable at a time; they do not combine actions or account for costs.
77
+ A row compares the same baseline/destination as the population. It does not use
78
+ the row's observed treatment as an implicit baseline.
79
+
80
+ Identification is conditional on the graph. The engine accepts adjustment or
81
+ null effects and reports other identification strategies as skipped. A valid
82
+ pretreatment adjustment set supplies the query features; unused columns do not
83
+ become evidence about an individual's potential outcome. Numeric range and
84
+ query-propensity checks refuse obvious extrapolation, but do not certify joint
85
+ support. A scientific refusal is atomic for the variable, never a zero effect.
86
+
87
+ Population intervals are pointwise and asymptotic. They exclude graph discovery,
88
+ ranking selection, multiple comparisons and continuous-DR smoothing bias. Query
89
+ effects are conditional expectations with no individual intervals or structural
90
+ ITE claim. All encodings, folds, defaults and nuisance fits use context only.
91
+
92
+ Pass regression learners in the `outcome` and `final` slots and a classifier
93
+ with normalised arm probabilities in `propensity`. Each fold gets a fresh model.
94
+ Welt's server uses its admitted native models; local ergodic defaults are simple
95
+ working models, not a guarantee of correct nuisance specification.
96
+
97
+ ## Candidate binary ranking protocol
98
+
99
+ Explicit `protocol="decision-contrasts_v2"` keeps population DR estimation and
100
+ uses `plugin` as the default for binary-treatment, binary-target query rankings:
101
+ each query effect is the difference between the cross-fitted arm outcome models
102
+ (the same nuisances that build the DR scores), averaged over folds. The original
103
+ protocol remains the default when the argument is omitted. `method` is still per
104
+ intervention: `plugin`, `s` (one outcome model with the treatment toggled), `dr`
105
+ (a final regression on the paired DR scores) or, with independent randomized
106
+ validation, `auto`. All other treatment/target defaults are unchanged. The
107
+ default is a predeclared choice, not a proven universal winner. It is the
108
+ classic variance trade-off: separately fitted arm models (`plugin`) have higher
109
+ variance, which shows where the untreated outcome varies a lot, while one model
110
+ with the treatment toggled (`s`) shrinks effects toward zero.
111
+
112
+ For discrete treatments and a binary target, v2 fits the selected event once and
113
+ derives the opposite class contrast by negation. An outcome classifier supplies
114
+ probability means, not labels; DR's final learner remains a regressor. Pass
115
+ `learners_by_variable={"action": {"outcome": classifier, "final": regressor}}`
116
+ to override the shared learner slots. Population `method="dr"` and query
117
+ `method="plugin"` or `"s"` intentionally identify different estimators. The
118
+ population ATE is not the average of the query effects.
119
+
120
+ Optional `validation` is a separate labeled dataframe with the context schema.
121
+ Supply `unit_id` and `validation_design="independent_bernoulli"`; context and
122
+ validation identifiers must be unique and disjoint, and are excluded from all
123
+ learning. The declaration asserts independent noninterfering units, truthful IDs,
124
+ pretreatment covariates and no validation-driven tuning or reuse. Merely renaming
125
+ the same units does not satisfy it. No validation row fits a model or preprocessing.
126
+
127
+ The initial gate supports binary treatment/outcome only. Each intervention supplies
128
+ `validation_probability`, the known assignment probability of its requested value
129
+ in the validation experiment, strictly between .01 and .99. It is not the learned
130
+ context propensity. For a fixed top-20% targeting policy, fractional boundary ties
131
+ preserve exactly 20% selected mass. Hoeffding bounds target the finite validation
132
+ cohort, conditional on frozen models/features and binary potential outcomes.
133
+ Every bound that may be reported receives Bonferroni correction, counted from the
134
+ request before any validation outcome is read: five per `auto` contrast (three
135
+ candidates against random and two paired differences against the default), one per
136
+ validated explicit-method contrast.
137
+
138
+ Auto evaluates `plugin`, `s` and `dr` and keeps `plugin` unless another candidate's
139
+ paired gain over `plugin` and its own gain over random both have strictly positive
140
+ lower bounds; among such candidates the largest paired lower bound wins. Otherwise
141
+ it reports `plugin`'s own evidence. Because `auto` chooses the ranking method
142
+ from validation outcomes, that validation cohort is spent: the selected method's
143
+ reported evidence holds within the simultaneous family, but reusing the cohort to
144
+ judge the chosen ranking again falls outside the guarantee.
145
+ `ranking_validation.status` is `supported`, `evidence_against`, `inconclusive`, or
146
+ `not_assessed`. No failed gate silently zeroes estimates. Supported targeting means
147
+ an advantage over random selection of the same fraction, not benefit versus treating
148
+ nobody, accurate individual causal effects, a correct discovered graph, or future
149
+ population performance. Repeated selection using the same validation cohort is
150
+ outside the guarantee. Other assignment designs are not silently approximated.
@@ -166,3 +166,5 @@ instruments; weak-IV robust intervals are not implemented here.
166
166
  - [Python reference](reference-inference.md): every option of `estimate_effect`.
167
167
 
168
168
  </div>
169
+
170
+ For explicit numerical changes, see [effects of numerical changes](continuous-treatments.md).
@@ -72,7 +72,7 @@ for module, use in optional.items():
72
72
  ```
73
73
 
74
74
  ```output from="install_check.py" lines="1"
75
- ergodic 0.3.2 writes ergodic.graph/1
75
+ ergodic 0.4.0 writes ergodic.graph/1
76
76
  ```
77
77
 
78
78
  - **Step 1.** The version, and `ergodic.graph/1`: the format `MixedGraph.to_dict()`
@@ -31,7 +31,7 @@ commands = [
31
31
 
32
32
  [versions]
33
33
  # Published versions older than this build that stay in the picker (v<version>/), oldest first.
34
- previous = ["0.2.0", "0.2.1", "0.3.0", "0.3.1"]
34
+ previous = ["0.2.0", "0.2.1", "0.3.0", "0.3.1", "0.3.2"]
35
35
 
36
36
  [[products]]
37
37
  name = "ergodic"
@@ -122,6 +122,8 @@ pages = [
122
122
  label = "Effects"
123
123
  pages = [
124
124
  ["Estimate an average effect", "estimate-average-effect"],
125
+ ["Estimate effects of numerical changes", "continuous-treatments"],
126
+ ["Rank interventions", "decisions"],
125
127
  ["Refute an estimate", "refute-an-estimate"],
126
128
  ["Estimate heterogeneous effects", "heterogeneous-effects"],
127
129
  ["Target a policy", "target-a-policy"],
@@ -235,6 +237,7 @@ modules = [
235
237
  ["identification", "Identification"],
236
238
  ["discovery", "Discovery"],
237
239
  ["inference", "Effect estimation"],
240
+ ["decision", "Ranked decisions"],
238
241
  ["process", "Process intelligence"],
239
242
  ["scm", "Structural causal models"],
240
243
  ["forecast", "Forecasting"],
@@ -91,6 +91,9 @@ from .identification import (
91
91
  from .inference import (
92
92
  CATEResult,
93
93
  CausalForest,
94
+ ContinuousDR,
95
+ DoseContrast,
96
+ DoseEffects,
94
97
  EffectEstimate,
95
98
  EffectTarget,
96
99
  ITEResult,
@@ -103,6 +106,7 @@ from .inference import (
103
106
  cate,
104
107
  did,
105
108
  estimate,
109
+ estimate_contrast,
106
110
  estimate_effect,
107
111
  estimate_mediation,
108
112
  event_study,
@@ -144,7 +148,7 @@ from .scm import (
144
148
  Sensitivity,
145
149
  )
146
150
 
147
- __version__ = "0.3.2"
151
+ __version__ = "0.4.0"
148
152
 
149
153
  __all__ = [
150
154
  "GRAPH_FORMAT",
@@ -157,9 +161,12 @@ __all__ = [
157
161
  "CausalForest",
158
162
  "CertifyResult",
159
163
  "Consensus",
164
+ "ContinuousDR",
160
165
  "Dataset",
161
166
  "DecodeResult",
162
167
  "DomainKnowledge",
168
+ "DoseContrast",
169
+ "DoseEffects",
163
170
  "Edge",
164
171
  "EdgeProbabilities",
165
172
  "EffectEstimate",
@@ -226,6 +233,7 @@ __all__ = [
226
233
  "discover",
227
234
  "discover_model",
228
235
  "estimate",
236
+ "estimate_contrast",
229
237
  "estimate_effect",
230
238
  "estimate_mediation",
231
239
  "evaluate_graph",
@@ -0,0 +1,7 @@
1
+ # SPDX-License-Identifier: Apache-2.0
2
+ """Rank explicit single-variable actions using graph-identified causal contrasts."""
3
+
4
+ from ._categorical import ArmDR
5
+ from ._pipeline import PROTOCOL, PROTOCOLS, RANKING_METHODS, Intervention, decide
6
+
7
+ __all__ = ["PROTOCOL", "PROTOCOLS", "RANKING_METHODS", "ArmDR", "Intervention", "decide"]