cowmrp 0.1.0__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.
- cowmrp-0.1.0/CITATION.cff +31 -0
- cowmrp-0.1.0/LICENSE +21 -0
- cowmrp-0.1.0/MANIFEST.in +10 -0
- cowmrp-0.1.0/PKG-INFO +277 -0
- cowmrp-0.1.0/README.md +230 -0
- cowmrp-0.1.0/examples/cow_comparison_example.py +18 -0
- cowmrp-0.1.0/examples/keep_cull_example.py +33 -0
- cowmrp-0.1.0/examples/whole_farm_example.py +39 -0
- cowmrp-0.1.0/examples/youngstock_example.py +25 -0
- cowmrp-0.1.0/guide/README.md +46 -0
- cowmrp-0.1.0/guide/assets/F01_milk_curves.png +0 -0
- cowmrp-0.1.0/guide/assets/F08_steady_state_herd.png +0 -0
- cowmrp-0.1.0/guide/assets/F11_rewards_per_day.png +0 -0
- cowmrp-0.1.0/guide/assets/fig_farm_lactation.png +0 -0
- cowmrp-0.1.0/guide/assets/fig_farm_origin.png +0 -0
- cowmrp-0.1.0/guide/assets/fig_farm_status.png +0 -0
- cowmrp-0.1.0/guide/assets/fig_perf_memory.png +0 -0
- cowmrp-0.1.0/guide/assets/fig_perf_per_cow.png +0 -0
- cowmrp-0.1.0/guide/assets/fig_perf_runtime.png +0 -0
- cowmrp-0.1.0/guide/assets/fig_perf_stages.png +0 -0
- cowmrp-0.1.0/guide/assets/fig_time_resolution.png +0 -0
- cowmrp-0.1.0/guide/build.py +337 -0
- cowmrp-0.1.0/guide/build_all.py +38 -0
- cowmrp-0.1.0/guide/check_links.py +20 -0
- cowmrp-0.1.0/guide/requirements-docs.txt +9 -0
- cowmrp-0.1.0/guide/resolution_worker.py +37 -0
- cowmrp-0.1.0/guide/sync.py +490 -0
- cowmrp-0.1.0/guide/tutorial.html +210 -0
- cowmrp-0.1.0/guide/web/README.md +23 -0
- cowmrp-0.1.0/guide/web/build_web.py +771 -0
- cowmrp-0.1.0/guide/web/check_methods.py +150 -0
- cowmrp-0.1.0/guide/web/content/defaults.html +43 -0
- cowmrp-0.1.0/guide/web/content/getting-started.html +44 -0
- cowmrp-0.1.0/guide/web/content/home.html +63 -0
- cowmrp-0.1.0/guide/web/content/keepcull-workbook.html +61 -0
- cowmrp-0.1.0/guide/web/content/notebooks.html +47 -0
- cowmrp-0.1.0/guide/web/content/performance.html +97 -0
- cowmrp-0.1.0/guide/web/content/quickstart.py +10 -0
- cowmrp-0.1.0/guide/web/content/science.html +54 -0
- cowmrp-0.1.0/guide/web/content/time_resolution.py +15 -0
- cowmrp-0.1.0/guide/web/performance_page.py +343 -0
- cowmrp-0.1.0/guide/web/theme/fonts/OFL.txt +103 -0
- cowmrp-0.1.0/guide/web/theme/fonts/cmu-serif-500-italic.woff2 +0 -0
- cowmrp-0.1.0/guide/web/theme/fonts/cmu-serif-500-roman.woff2 +0 -0
- cowmrp-0.1.0/guide/web/theme/fonts/cmu-serif-700-italic.woff2 +0 -0
- cowmrp-0.1.0/guide/web/theme/fonts/cmu-serif-700-roman.woff2 +0 -0
- cowmrp-0.1.0/guide/web/theme/fonts/cmu-typewriter-text-500-roman.woff2 +0 -0
- cowmrp-0.1.0/guide/web/theme/web.css +128 -0
- cowmrp-0.1.0/guide/web/theme/web.js +60 -0
- cowmrp-0.1.0/pyproject.toml +59 -0
- cowmrp-0.1.0/runners/README.md +63 -0
- cowmrp-0.1.0/runners/__init__.py +0 -0
- cowmrp-0.1.0/runners/_shared/__init__.py +0 -0
- cowmrp-0.1.0/runners/_shared/checks/__init__.py +0 -0
- cowmrp-0.1.0/runners/_shared/checks/check_documentation.py +266 -0
- cowmrp-0.1.0/runners/_shared/checks/validate_model_documentation.py +157 -0
- cowmrp-0.1.0/runners/_shared/checks/verify_baseline.py +80 -0
- cowmrp-0.1.0/runners/_shared/checks/verify_migration_parity.py +94 -0
- cowmrp-0.1.0/runners/_shared/checks/verify_mrp_boundary.py +39 -0
- cowmrp-0.1.0/runners/_shared/common.py +125 -0
- cowmrp-0.1.0/runners/_shared/engine_run.py +57 -0
- cowmrp-0.1.0/runners/_shared/new_domain.py +111 -0
- cowmrp-0.1.0/runners/_shared/notebook/mrp_engine.ipynb +519 -0
- cowmrp-0.1.0/runners/_shared/review.py +231 -0
- cowmrp-0.1.0/runners/_shared/run_notebook.py +46 -0
- cowmrp-0.1.0/runners/_shared/scale_benchmark.py +494 -0
- cowmrp-0.1.0/runners/_shared/scale_report.py +231 -0
- cowmrp-0.1.0/runners/_shared/views.py +509 -0
- cowmrp-0.1.0/runners/cow/__init__.py +0 -0
- cowmrp-0.1.0/runners/cow/checks/__init__.py +0 -0
- cowmrp-0.1.0/runners/cow/checks/audit_conception_sensitivity.py +108 -0
- cowmrp-0.1.0/runners/cow/checks/audit_exits.py +103 -0
- cowmrp-0.1.0/runners/cow/checks/audit_later_lactation_sensitivity.py +115 -0
- cowmrp-0.1.0/runners/cow/checks/audit_methods.py +80 -0
- cowmrp-0.1.0/runners/cow/checks/audit_metrics.py +226 -0
- cowmrp-0.1.0/runners/cow/checks/audit_milk.py +31 -0
- cowmrp-0.1.0/runners/cow/checks/audit_registry.py +10 -0
- cowmrp-0.1.0/runners/cow/checks/audit_repro.py +112 -0
- cowmrp-0.1.0/runners/cow/checks/audit_state_timing.py +118 -0
- cowmrp-0.1.0/runners/cow/checks/audit_tables.py +29 -0
- cowmrp-0.1.0/runners/cow/checks/compare_state_conventions.py +137 -0
- cowmrp-0.1.0/runners/cow/checks/verify_biological_consistency.py +227 -0
- cowmrp-0.1.0/runners/cow/checks/verify_farm.py +120 -0
- cowmrp-0.1.0/runners/cow/checks/verify_lactationcurve_milk_income_reward.py +273 -0
- cowmrp-0.1.0/runners/cow/checks/verify_lookup_export.py +231 -0
- cowmrp-0.1.0/runners/cow/checks/verify_reward_vector.py +258 -0
- cowmrp-0.1.0/runners/cow/checks/verify_state_semantics.py +227 -0
- cowmrp-0.1.0/runners/cow/checks/verify_state_space.py +261 -0
- cowmrp-0.1.0/runners/cow/checks/verify_state_sufficiency.py +109 -0
- cowmrp-0.1.0/runners/cow/checks/verify_time_discretization.py +144 -0
- cowmrp-0.1.0/runners/cow/checks/verify_value_comparisons.py +250 -0
- cowmrp-0.1.0/runners/cow/farm/__init__.py +0 -0
- cowmrp-0.1.0/runners/cow/farm/build.py +97 -0
- cowmrp-0.1.0/runners/cow/farm/generate.py +86 -0
- cowmrp-0.1.0/runners/cow/farm/keepcull.py +120 -0
- cowmrp-0.1.0/runners/cow/farm/simulate.py +103 -0
- cowmrp-0.1.0/runners/cow/plots/__init__.py +0 -0
- cowmrp-0.1.0/runners/cow/plots/_plot_data.py +186 -0
- cowmrp-0.1.0/runners/cow/plots/export_lookup_tables.py +356 -0
- cowmrp-0.1.0/runners/cow/plots/plot_validation.py +392 -0
- cowmrp-0.1.0/runners/cow/plots/plot_validation_plotly.py +460 -0
- cowmrp-0.1.0/runners/cow/profile_build.py +88 -0
- cowmrp-0.1.0/runners/cow/projection.py +135 -0
- cowmrp-0.1.0/runners/cow/run.py +104 -0
- cowmrp-0.1.0/runners/cow/scale_inputs.py +21 -0
- cowmrp-0.1.0/runners/run_all.py +50 -0
- cowmrp-0.1.0/runners/youngstock/__init__.py +0 -0
- cowmrp-0.1.0/runners/youngstock/checks/__init__.py +0 -0
- cowmrp-0.1.0/runners/youngstock/plots/__init__.py +0 -0
- cowmrp-0.1.0/runners/youngstock/run.py +121 -0
- cowmrp-0.1.0/runners/youngstock/scale_inputs.py +19 -0
- cowmrp-0.1.0/setup.cfg +4 -0
- cowmrp-0.1.0/src/cowmrp/__init__.py +14 -0
- cowmrp-0.1.0/src/cowmrp/data/__init__.py +18 -0
- cowmrp-0.1.0/src/cowmrp/data/example_herd.csv +201 -0
- cowmrp-0.1.0/src/cowmrp/data/example_herd.meta.json +10 -0
- cowmrp-0.1.0/src/cowmrp/data/example_herd_raw.csv +201 -0
- cowmrp-0.1.0/src/cowmrp/domain.py +123 -0
- cowmrp-0.1.0/src/cowmrp/domains/__init__.py +0 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/__init__.py +0 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/curves.py +52 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/data/README.md +24 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/economics.py +205 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/evidence/__init__.py +8 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/evidence/miller2008.py +22 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/evidence/pinedo2010.py +119 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/farm.py +538 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/flows.py +28 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/keepcull.py +104 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/keepcull_excel.py +217 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/lookup_tables.py +233 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/lookup_workbook.py +224 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/notebook_hooks.py +28 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/parameters.py +779 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/probabilities/__init__.py +123 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/probabilities/abortion.py +279 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/probabilities/base.py +117 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/probabilities/calving.py +150 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/probabilities/conception.py +234 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/probabilities/culling.py +205 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/probabilities/hazard_tables.py +108 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/probabilities/mortality.py +170 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/projection.py +197 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/registry.py +203 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/review.py +113 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/__init__.py +179 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/base.py +152 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/cull_reward.py +40 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/__init__.py +213 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/co2.py +123 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/enteric_methane.py +134 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/manure_methane.py +117 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/n2o.py +167 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/nitrogen.py +161 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/reward_enteric_methane.py +26 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/reward_manure_methane.py +33 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/reward_n2o.py +33 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/environment/reward_nitrogen.py +37 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/feed_cost.py +49 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/lactationcurve_milk_income.py +330 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/milk_income.py +112 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/milk_produced.py +33 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/mortality_cost.py +16 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/pregnancy_reward.py +25 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/replacement_cost.py +61 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/rewards/reproduction_cost.py +57 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/state_checks.py +67 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/state_constraints.py +181 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/states.py +70 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/tables.py +209 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/transitions.py +110 -0
- cowmrp-0.1.0/src/cowmrp/domains/cow/value_comparisons.py +317 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/__init__.py +0 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/notebook_hooks.py +4 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/parameters.py +368 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/probabilities/__init__.py +61 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/probabilities/birth_outcomes.py +137 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/probabilities/heifer_repro.py +188 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/probabilities/mortality.py +109 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/review.py +129 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/states.py +69 -0
- cowmrp-0.1.0/src/cowmrp/domains/youngstock/transitions.py +93 -0
- cowmrp-0.1.0/src/cowmrp/io/__init__.py +4 -0
- cowmrp-0.1.0/src/cowmrp/io/export.py +36 -0
- cowmrp-0.1.0/src/cowmrp/model.py +173 -0
- cowmrp-0.1.0/src/cowmrp/mrp/__init__.py +1 -0
- cowmrp-0.1.0/src/cowmrp/mrp/core.py +520 -0
- cowmrp-0.1.0/src/cowmrp/mrp/engine.py +221 -0
- cowmrp-0.1.0/src/cowmrp/mrp/simulator.py +221 -0
- cowmrp-0.1.0/src/cowmrp/mrp/utils.py +36 -0
- cowmrp-0.1.0/src/cowmrp/runner.py +8 -0
- cowmrp-0.1.0/src/cowmrp/viz/__init__.py +1 -0
- cowmrp-0.1.0/src/cowmrp/viz/distribution.py +44 -0
- cowmrp-0.1.0/src/cowmrp/viz/farm.py +266 -0
- cowmrp-0.1.0/src/cowmrp/viz/projection.py +118 -0
- cowmrp-0.1.0/src/cowmrp.egg-info/PKG-INFO +277 -0
- cowmrp-0.1.0/src/cowmrp.egg-info/SOURCES.txt +229 -0
- cowmrp-0.1.0/src/cowmrp.egg-info/dependency_links.txt +1 -0
- cowmrp-0.1.0/src/cowmrp.egg-info/requires.txt +31 -0
- cowmrp-0.1.0/src/cowmrp.egg-info/top_level.txt +1 -0
- cowmrp-0.1.0/tests/README.md +40 -0
- cowmrp-0.1.0/tests/__init__.py +0 -0
- cowmrp-0.1.0/tests/baselines/farm_golden.json +1 -0
- cowmrp-0.1.0/tests/baselines/golden_month.json +21 -0
- cowmrp-0.1.0/tests/baselines/keepcull_golden.json +1 -0
- cowmrp-0.1.0/tests/baselines/projection_golden.json +593 -0
- cowmrp-0.1.0/tests/boundary/test_domain_independence.py +75 -0
- cowmrp-0.1.0/tests/conftest.py +47 -0
- cowmrp-0.1.0/tests/install/test_clean_install.py +125 -0
- cowmrp-0.1.0/tests/integration/test_cow_economics.py +290 -0
- cowmrp-0.1.0/tests/integration/test_cow_farm.py +260 -0
- cowmrp-0.1.0/tests/integration/test_cow_keepcull.py +144 -0
- cowmrp-0.1.0/tests/integration/test_cow_keepcull_excel.py +126 -0
- cowmrp-0.1.0/tests/integration/test_cow_projection.py +132 -0
- cowmrp-0.1.0/tests/integration/test_doc_sync.py +75 -0
- cowmrp-0.1.0/tests/integration/test_environment_outputs.py +21 -0
- cowmrp-0.1.0/tests/integration/test_group_probs.py +107 -0
- cowmrp-0.1.0/tests/integration/test_model.py +106 -0
- cowmrp-0.1.0/tests/integration/test_runners_use_package.py +36 -0
- cowmrp-0.1.0/tests/machine_domain/__init__.py +2 -0
- cowmrp-0.1.0/tests/machine_domain/notebook_hooks.py +7 -0
- cowmrp-0.1.0/tests/machine_domain/parameters.py +6 -0
- cowmrp-0.1.0/tests/machine_domain/probabilities/__init__.py +7 -0
- cowmrp-0.1.0/tests/machine_domain/rewards.py +5 -0
- cowmrp-0.1.0/tests/machine_domain/states.py +4 -0
- cowmrp-0.1.0/tests/machine_domain/transitions.py +13 -0
- cowmrp-0.1.0/tests/performance/test_performance.py +66 -0
- cowmrp-0.1.0/tests/regression/test_scientific_reference.py +45 -0
- cowmrp-0.1.0/tests/unit/test_core.py +95 -0
- cowmrp-0.1.0/tests/unit/test_domain.py +55 -0
- cowmrp-0.1.0/tests/unit/test_licence_consistency.py +34 -0
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
# Citation metadata. ORCIDs, funding, the release date and the DOI are added when available.
|
|
2
|
+
cff-version: 1.2.0
|
|
3
|
+
message: "If you use this software, please cite it using the metadata in this file."
|
|
4
|
+
title: "cowmrp: Markov reward process models of dairy cattle"
|
|
5
|
+
abstract: >-
|
|
6
|
+
A Python package of Markov reward process (MRP) models of dairy cattle: a domain-free MRP engine with a cow domain
|
|
7
|
+
(state: lactation, days in milk, days pregnant) and a youngstock domain. It values cows (keep or replace), projects a herd
|
|
8
|
+
forward with the expected distribution, stochastic paths or Monte Carlo, and reports environmental outputs beside economic ones.
|
|
9
|
+
type: software
|
|
10
|
+
version: 0.1.0
|
|
11
|
+
license: MIT
|
|
12
|
+
repository-code: "https://github.com/Bovi-analytics/cow-mrp"
|
|
13
|
+
url: "https://bovi-analytics.github.io/cow-mrp/"
|
|
14
|
+
keywords:
|
|
15
|
+
- markov reward process
|
|
16
|
+
- dairy cattle
|
|
17
|
+
- replacement
|
|
18
|
+
- simulation
|
|
19
|
+
- livestock
|
|
20
|
+
authors:
|
|
21
|
+
- family-names: Sharma
|
|
22
|
+
given-names: Sumit
|
|
23
|
+
affiliation: "Department of Animal Science, Cornell University, Ithaca, New York, USA"
|
|
24
|
+
# orcid: pending (not provided yet; add as "https://orcid.org/0000-0000-0000-0000")
|
|
25
|
+
- family-names: Hansen
|
|
26
|
+
given-names: Edward
|
|
27
|
+
# orcid: pending (not provided yet; add as "https://orcid.org/0000-0000-0000-0000")
|
|
28
|
+
- family-names: Hostens
|
|
29
|
+
given-names: Miel
|
|
30
|
+
affiliation: "Department of Animal Science, Cornell University, Ithaca, New York, USA"
|
|
31
|
+
# orcid: pending (not provided yet; add as "https://orcid.org/0000-0000-0000-0000")
|
cowmrp-0.1.0/LICENSE
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 Bovi-analytics Lab, Cornell University
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE.
|
cowmrp-0.1.0/MANIFEST.in
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
include LICENSE README.md CITATION.cff
|
|
2
|
+
recursive-include examples *.py
|
|
3
|
+
recursive-include tests *.py *.md *.json *.csv
|
|
4
|
+
recursive-include runners *.py *.md *.ipynb *.json *.csv
|
|
5
|
+
recursive-include guide *.py *.md *.txt *.html *.css *.js *.png *.woff2 OFL.txt
|
|
6
|
+
prune runners/*/outputs
|
|
7
|
+
prune guide/site
|
|
8
|
+
prune guide/web/site
|
|
9
|
+
prune guide/web/outputs
|
|
10
|
+
global-exclude __pycache__ *.py[cod]
|
cowmrp-0.1.0/PKG-INFO
ADDED
|
@@ -0,0 +1,277 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: cowmrp
|
|
3
|
+
Version: 0.1.0
|
|
4
|
+
Summary: Markov reward process (MRP) models of dairy cattle: a domain-free MRP engine with cow and youngstock domains
|
|
5
|
+
Author: Sumit Sharma, Edward Hansen, Miel Hostens
|
|
6
|
+
License-Expression: MIT
|
|
7
|
+
Project-URL: Homepage, https://github.com/Bovi-analytics/cow-mrp
|
|
8
|
+
Project-URL: Source, https://github.com/Bovi-analytics/cow-mrp
|
|
9
|
+
Project-URL: Documentation, https://bovi-analytics.github.io/cow-mrp/
|
|
10
|
+
Project-URL: Issues, https://github.com/Bovi-analytics/cow-mrp/issues
|
|
11
|
+
Keywords: markov reward process,dairy cattle,replacement,simulation,livestock
|
|
12
|
+
Classifier: Development Status :: 3 - Alpha
|
|
13
|
+
Classifier: Intended Audience :: Science/Research
|
|
14
|
+
Classifier: Programming Language :: Python :: 3
|
|
15
|
+
Classifier: Programming Language :: Python :: 3.10
|
|
16
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
17
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
18
|
+
Classifier: Programming Language :: Python :: 3.13
|
|
19
|
+
Classifier: Topic :: Scientific/Engineering
|
|
20
|
+
Requires-Python: >=3.10
|
|
21
|
+
Description-Content-Type: text/markdown
|
|
22
|
+
License-File: LICENSE
|
|
23
|
+
Requires-Dist: numpy
|
|
24
|
+
Requires-Dist: pandas
|
|
25
|
+
Requires-Dist: scipy
|
|
26
|
+
Provides-Extra: plot
|
|
27
|
+
Requires-Dist: matplotlib; extra == "plot"
|
|
28
|
+
Requires-Dist: plotly; extra == "plot"
|
|
29
|
+
Provides-Extra: excel
|
|
30
|
+
Requires-Dist: openpyxl; extra == "excel"
|
|
31
|
+
Provides-Extra: parquet
|
|
32
|
+
Requires-Dist: pyarrow; extra == "parquet"
|
|
33
|
+
Provides-Extra: bench
|
|
34
|
+
Requires-Dist: psutil; extra == "bench"
|
|
35
|
+
Provides-Extra: notebook
|
|
36
|
+
Requires-Dist: jupyter; extra == "notebook"
|
|
37
|
+
Provides-Extra: milkbot
|
|
38
|
+
Requires-Dist: lactationcurve; extra == "milkbot"
|
|
39
|
+
Provides-Extra: dev
|
|
40
|
+
Requires-Dist: pytest; extra == "dev"
|
|
41
|
+
Requires-Dist: ruff; extra == "dev"
|
|
42
|
+
Requires-Dist: build; extra == "dev"
|
|
43
|
+
Requires-Dist: twine; extra == "dev"
|
|
44
|
+
Provides-Extra: all
|
|
45
|
+
Requires-Dist: cowmrp[bench,excel,notebook,parquet,plot]; extra == "all"
|
|
46
|
+
Dynamic: license-file
|
|
47
|
+
|
|
48
|
+
# CowMRP (`cowmrp`)
|
|
49
|
+
|
|
50
|
+
Markov reward process (MRP) models of dairy cattle, as a Python package. A **domain-free MRP engine** (`cowmrp.mrp`) plus two animal domains:
|
|
51
|
+
the **cow** (state `(lactation, dim, dp)`: lactation number, days in milk, days pregnant, in model-step units) and the **youngstock** chain
|
|
52
|
+
(calf, heifer, first calving). The engine knows no animal, so you can write your own domain.
|
|
53
|
+
|
|
54
|
+
It is for researchers and farm advisers who want to value cows (keep or replace), project a herd forward, and look at environmental outputs
|
|
55
|
+
beside economic ones. The scientific model is the reference implementation; the package changes how it is reached, not what it computes.
|
|
56
|
+
|
|
57
|
+
## Install
|
|
58
|
+
|
|
59
|
+
```
|
|
60
|
+
pip install cowmrp # engine + domains (numpy, pandas, scipy)
|
|
61
|
+
pip install "cowmrp[plot]" # + matplotlib, plotly: dashboards and projection figures
|
|
62
|
+
pip install -e ".[dev,plot,parquet]" # from a clone of this repository (Python 3.10 or newer)
|
|
63
|
+
```
|
|
64
|
+
|
|
65
|
+
The installed package contains the engine, the domains, the plotting helpers and the 200-cow tutorial herd. The runner scripts (`runners/`), `examples/`,
|
|
66
|
+
`user_examples/` (the example notebooks), the tests and the documentation sources are in the source repository, https://github.com/Bovi-analytics/cow-mrp; the commands in
|
|
67
|
+
"Running the scripts" and "Tests" below are for a clone.
|
|
68
|
+
|
|
69
|
+
## Quick start: a whole farm in a few lines
|
|
70
|
+
|
|
71
|
+
```python
|
|
72
|
+
from cowmrp.data import example_herd_path # 200 synthetic cows shipped with the package
|
|
73
|
+
from cowmrp.model import build_model
|
|
74
|
+
from cowmrp.domains.cow.farm import load_herd, project_herd
|
|
75
|
+
from cowmrp.viz.farm import write_farm_outputs
|
|
76
|
+
|
|
77
|
+
model = build_model("cow") # month resolution (see "Time resolution")
|
|
78
|
+
herd = load_herd(example_herd_path(), model) # or your own CSV / DataFrame
|
|
79
|
+
result = project_herd(herd, model, mode="expected", steps=24)
|
|
80
|
+
write_farm_outputs(result, "farm_results") # CSV tables + dashboard.html + dashboard_plotly.html
|
|
81
|
+
```
|
|
82
|
+
|
|
83
|
+
**Example notebooks.** The folder [`user_examples/`](user_examples/) holds five Jupyter notebooks that start with `pip install "cowmrp[plot,excel]"`, import only from the installed package and run on a CSV that you replace with your own: cows from a CSV with keep/cull, a farm projection with the interactive Plotly dashboard, individual rewards, a youngstock farm, and a domain of your own on the generic engine. They are not part of the installed package; get them with `git clone https://github.com/Bovi-analytics/cow-mrp.git` (or Download ZIP on GitHub) and see `user_examples/README.md` and the Examples page of the documentation site.
|
|
84
|
+
|
|
85
|
+
A complete script is `examples/whole_farm_example.py`; the same steps are walked through in the [tutorial](https://bovi-analytics.github.io/cow-mrp/tutorial.html) of the documentation site (online once the repository is public; the source is `guide/tutorial.html`).
|
|
86
|
+
|
|
87
|
+
## What the package does
|
|
88
|
+
|
|
89
|
+
| Task | Entry point | Notes |
|
|
90
|
+
|---|---|---|
|
|
91
|
+
| Build and solve a model | `cowmrp.model.build_model("cow" \| "youngstock")`, `evaluate`, `evolve_distribution` | Any domain that follows the interface in `cowmrp.domain` |
|
|
92
|
+
| **One cow, or cow A against cow B** | `cowmrp.domains.cow.projection.compare_cows(scenarios, years=2)` | Rewards and distributions over time for cows given as `lactation`, `dim_days`, `dp_days`; figures in `cowmrp.viz.projection` |
|
|
93
|
+
| **Keep or cull** | `cowmrp.domains.cow.keepcull.keep_cull_table(herd, model, years=(1,3,5), offer=900)` | Break-even sale value of every cow = value of keeping her minus value of replacing her now; a KEEP/CULL call only when you give an `offer`. Salvage is not credited in the model's own comparison (a documented limitation). `cowmrp.domains.cow.keepcull_excel.write_keepcull_workbook` (`--excel`, needs `cowmrp[excel]`) writes an Excel workbook that shows the sale of the cow now as its own column; see below |
|
|
94
|
+
| **Whole-farm projection** | `cowmrp.domains.cow.farm.project_herd(herd, model, mode, steps)` | Open / pregnant / dry, lactation, DIM stage; cows never replaced versus replaced; per-cow values |
|
|
95
|
+
| **Load farm records** | `cowmrp.domains.cow.farm.load_herd(csv_or_dataframe, model)` | Raw records or already-mapped `state_id`; every problem reported with row and `cow_id` |
|
|
96
|
+
| Environmental outputs | `cowmrp.domains.cow.rewards.environment` | The physical quantities (enteric and manure methane, N2O, nitrogen, CO2) are always computed by `compute_environment()` and do not depend on any price; the monetary environmental rewards are off by default and enter the economic total only if you enable them |
|
|
97
|
+
| Synthetic herd | `cowmrp.domains.cow.farm.generate_herd(model, mode, n_cows, seed)` | `equilibrium` (the model's steady-state herd), `mix`, `exact` |
|
|
98
|
+
|
|
99
|
+
### Farm data
|
|
100
|
+
|
|
101
|
+
One row per cow, described in **real days** (never health or DNB flags, which are not state variables):
|
|
102
|
+
|
|
103
|
+
| Column | Meaning |
|
|
104
|
+
|---|---|
|
|
105
|
+
| `cow_id` | optional, unique (1..n when absent) |
|
|
106
|
+
| `lactation` | 1 to 10 |
|
|
107
|
+
| `dim_days` | days in milk, 0 or more |
|
|
108
|
+
| `dp_days` | days pregnant, 0 for an open cow (default 0) |
|
|
109
|
+
| `state_id` | alternative: the model state of the cow, as written by `generate_herd` (if given together with the days they must agree) |
|
|
110
|
+
| `uim`, `up` | alternative to the days columns: the model's step indices |
|
|
111
|
+
|
|
112
|
+
A cow is mapped to a state exactly as the original runner `runners/cow/farm/generate.py` always did. `load_herd` adds checks only: missing columns, non-numeric
|
|
113
|
+
values, out-of-range lactation, negative days, duplicate `cow_id`, a combination the model has no state for (the message lists the nearest valid states), and a herd
|
|
114
|
+
file made under another time resolution. A DataFrame is accepted wherever a CSV path is.
|
|
115
|
+
|
|
116
|
+
### Tutorial dataset
|
|
117
|
+
|
|
118
|
+
`cowmrp.data.example_herd_path()` is a 200-cow **synthetic** herd (cows drawn from the model's own steady-state herd, seed 42, month resolution) in the model's schema;
|
|
119
|
+
`example_raw_herd_path()` is the same cows as a farmer would export them (`cow_id, lactation, dim_days, dp_days`). No real farm records are included.
|
|
120
|
+
|
|
121
|
+
### Time resolution
|
|
122
|
+
|
|
123
|
+
The model steps in `month` (30 d, default), `week` (7 d) or `day` (1 d). Set it **before Python starts**: `COW_TIME_UNIT=week python my_script.py`.
|
|
124
|
+
The resolution is not a display setting: it rebuilds the state space, the transition probabilities (rescaled to the step), the reward of every state (the rate integrated over the days
|
|
125
|
+
one step spans) and so the expected values. `dim` and `dp` are step indices of the chosen resolution, not calendar days; durations such as gestation stay in real days.
|
|
126
|
+
One process holds one resolution (the cow domain reads it once, when imported); `compare_cows` runs scenarios of another resolution in worker processes for you.
|
|
127
|
+
A herd file made at one resolution is refused at another; regenerate it. The day resolution is large: use it deliberately.
|
|
128
|
+
|
|
129
|
+
<!-- GEN:resolution_table -->
|
|
130
|
+
| Resolution | Step | States | Transitions | Example cow as `(dim, dp)` |
|
|
131
|
+
|---|---:|---:|---:|---:|
|
|
132
|
+
| month | 30 d | 2,970 | 9,040 | (4, 2) |
|
|
133
|
+
| week | 7 d | 42,290 | 131,197 | (20, 9) |
|
|
134
|
+
| day | 1 d | 1,952,060 | 6,109,989 | (149, 60) |
|
|
135
|
+
|
|
136
|
+
The example cow is in her second lactation, 150 days in milk and 60 days pregnant; `dim` and `dp` count model steps, so the same cow has different state coordinates at each resolution.
|
|
137
|
+
<!-- /GEN -->
|
|
138
|
+
|
|
139
|
+
<!-- GEN:state_counts -->
|
|
140
|
+
State counts (month resolution): 2,970 states, of which 1,210 are reachable from the replacement state (the default is the full table), 9,040 transitions.
|
|
141
|
+
<!-- /GEN -->
|
|
142
|
+
|
|
143
|
+
### Expected versus stochastic and Monte Carlo
|
|
144
|
+
|
|
145
|
+
`project_herd(..., mode=...)`:
|
|
146
|
+
|
|
147
|
+
| Mode | What it computes | When to use |
|
|
148
|
+
|---|---|---|
|
|
149
|
+
| `expected` (default) | The exact distribution of the herd after each step, split into cows never replaced and cows replaced at least once. Deterministic, fast | Almost always: planning, comparing scenarios, dashboards |
|
|
150
|
+
| `stochastic` | One sampled path per cow (`seed`) | To see a single plausible herd history |
|
|
151
|
+
| `monte_carlo` | Many sampled herds (`sims`), reported as mean, standard deviation, 5th and 95th percentile | When the spread, not only the mean, matters (small herds) |
|
|
152
|
+
|
|
153
|
+
For a large herd the three agree on the mean; Monte Carlo is the slow one, so it is never run unless you select it.
|
|
154
|
+
|
|
155
|
+
## Repository layout
|
|
156
|
+
|
|
157
|
+
| Folder | Purpose |
|
|
158
|
+
|---|---|
|
|
159
|
+
| `src/cowmrp/mrp/` | Domain-free MRP engine: `core.py` (math), `engine.py` (spec orchestration and validation), `simulator.py`, `utils.py` |
|
|
160
|
+
| `src/cowmrp/domain.py`, `model.py` | The domain interface (`load_domain`, `validate_domain`) and the generic build/solve API |
|
|
161
|
+
| `src/cowmrp/domains/` | One folder per domain: `cow/` (parameters, states, probabilities, transitions, rewards, lookup tables, evidence, `farm.py`, `keepcull.py`, `projection.py`) and `youngstock/` |
|
|
162
|
+
| `src/cowmrp/viz/`, `io/`, `data/` | Figures and dashboards, model-file export, the shipped example herd |
|
|
163
|
+
| `runners/` | Scripts that use the package: one module per domain (`cow/`, `youngstock/`) with `run.py`, `checks/`, `plots/`, `outputs/<kind>/`; `_shared/` common code |
|
|
164
|
+
| `examples/` | Four runnable scripts (`whole_farm_example.py`, ...) |
|
|
165
|
+
| `user_examples/` | Five example notebooks with their synthetic CSV files, a custom-domain folder and the installed-wheel test (`check_installed.py`); obtained from GitHub, not in the package |
|
|
166
|
+
| `tests/` | Fast default suite plus opt-in `slow` and `install` tests (see `tests/README.md`) |
|
|
167
|
+
| `guide/` | Documentation: the tutorial, a file guide and API reference built by `guide/build.py`, and the website in `guide/web/` |
|
|
168
|
+
|
|
169
|
+
Dependency direction is one-way: `cowmrp.mrp` <- `cowmrp.domain` <- `cowmrp.model` <- `cowmrp.domains` <- `runners`
|
|
170
|
+
(check: `python runners/_shared/checks/verify_mrp_boundary.py`). `runners/README.md` explains the runner modules.
|
|
171
|
+
|
|
172
|
+
**Add a domain:** write a package with `parameters`, `states`, `probabilities`, `transitions` (and optional `rewards`) following `cowmrp.domain`, then
|
|
173
|
+
`cowmrp.domain.load_domain("your_package")`. `python runners/_shared/new_domain.py <name>` creates the stubs and the runner module.
|
|
174
|
+
|
|
175
|
+
## Running the scripts (research workflows)
|
|
176
|
+
|
|
177
|
+
All scripts run from any directory. They are kept and call the package.
|
|
178
|
+
|
|
179
|
+
```
|
|
180
|
+
python runners/run_all.py # every module -> runners/<module>/outputs/
|
|
181
|
+
python runners/cow/run.py --review # review pages for a domain expert
|
|
182
|
+
python runners/cow/projection.py # edit SCENARIOS at the top first (cow A vs cow B)
|
|
183
|
+
python runners/cow/farm/generate.py # a synthetic herd -> runners/cow/outputs/farm/farm_data.csv
|
|
184
|
+
python runners/cow/farm/simulate.py --mode expected --steps 24 # herd projection and dashboards
|
|
185
|
+
python runners/cow/farm/keepcull.py --years 1,3,5 --offer 900 # keep/cull table
|
|
186
|
+
python runners/cow/farm/keepcull.py --excel # + keepcull_accounting.xlsx: KEEP, REPLACE, salvage, break-even per cow (needs cowmrp[excel])
|
|
187
|
+
python runners/cow/farm/keepcull.py --set REARING_COST=2800 # scenario run: override an economic setting for this run only (see below)
|
|
188
|
+
python runners/cow/checks/verify_farm.py # and the other verify_* scripts in runners/*/checks/
|
|
189
|
+
python runners/_shared/scale_benchmark.py --quick # cost and accuracy against herd size
|
|
190
|
+
```
|
|
191
|
+
|
|
192
|
+
### Keeping the documentation in step with the code
|
|
193
|
+
|
|
194
|
+
Numbers and tables that the documents repeat (state counts, reward components, overridable economic settings) are read from the package, not typed:
|
|
195
|
+
|
|
196
|
+
```
|
|
197
|
+
python guide/build_all.py # rewrite the generated blocks, build both sites, then check that nothing is stale
|
|
198
|
+
python guide/sync.py check # exit 1 if a block, token, number or built site is out of date
|
|
199
|
+
python guide/sync.py impact <file> # which documents and outputs depend on a source file
|
|
200
|
+
```
|
|
201
|
+
|
|
202
|
+
Generated blocks sit between `<!-- GEN:name -->` and `<!-- /GEN -->` in this file; do not edit inside them. `{{NAME}}` tokens in the tutorial and site pages are expanded at build time.
|
|
203
|
+
|
|
204
|
+
### Keep/cull accounting workbook
|
|
205
|
+
|
|
206
|
+
`python runners/cow/farm/keepcull.py --excel` also writes `keepcull_accounting.xlsx`: for every cow and horizon, the KEEP value, the REPLACE value, the immediate salvage (an indicative estimate, or the net offer you type for that cow), the break-even offer, ΔQ and the decision, as Excel formulas. It is a reporting output; the model, the reward vector and `keep_cull_table` are unchanged. The estimate is indicative (state-based body weight; the model's dressing percentage, beef price and culling fee) and a typed offer is the net cash received. The `Checks` sheet shows that the break-even column reproduces `keep_cull_table`. Terms, every column and formula, the treatment of future salvage and replacement costs, and the limitations are on the documentation site page *Keep/cull workbook* (`guide/web/content/keepcull-workbook.html`). The column table below is generated from the package:
|
|
207
|
+
|
|
208
|
+
<!-- GEN:excel_columns -->
|
|
209
|
+
| Col | Header | Kind | Formula (row r) | Meaning |
|
|
210
|
+
|---|---|---|---|---|
|
|
211
|
+
| A | cow_id | from the herd file | | Cow identifier from the herd file (row number if the herd has none). |
|
|
212
|
+
| B | state_id | from the herd file | | Model state of the cow (index into the state table of the current time resolution). |
|
|
213
|
+
| C | lactation | from the herd file | | Lactation number (copied when the herd has it). |
|
|
214
|
+
| D | dim_days | from the herd file | | Days in milk (copied when the herd has it). |
|
|
215
|
+
| E | dp_days | from the herd file | | Days pregnant (copied when the herd has it). |
|
|
216
|
+
| F | body_weight_kg (model, state-based) | model value (pasted) | | Body weight of the state in kg, computed as cull_reward computes it for the salvage value. A model estimate, not the weight on the day of sale. |
|
|
217
|
+
| G | V_total(s) | model value (pasted) | | V_total(s): expected discounted total reward of the cow over the horizon (milk, pregnancy, cull, mortality, feed, reproduction), future cull salvage included. |
|
|
218
|
+
| H | PV future replacement costs (s) | model value (pasted) | | PV_rc(s): expected discounted cost of the future replacements of this cow and of her replacements over the horizon. |
|
|
219
|
+
| I | KEEP = V_total - PV_rc | Excel formula | `=Gr-Hr` | KEEP: expected future value of keeping the cow, every future replacement cost paid. |
|
|
220
|
+
| J | estimated salvage S_est | Excel formula | `=Fr*Inputs!$B$3*Inputs!$B$2-Inputs!$B$4` | Estimated salvage now: body weight x dressing percentage x beef price - culling fee (Inputs B3, B2, B4). |
|
|
221
|
+
| K | actual net offer (type here, optional) | input (yellow) | | Actual net offer for this cow, typed by the user (optional). Net cash received: the culling fee is not subtracted again. |
|
|
222
|
+
| L | S used | Excel formula | `=IF(Kr="",Jr,Kr)` | Salvage used: the typed offer if there is one, otherwise the estimate. |
|
|
223
|
+
| M | V_keep(heifer) | Excel formula | `=Inputs!$E$<horizon row>` | V_keep(h): value of a fresh replacement heifer (state (1, 0, 0)) over the same horizon (Inputs sheet). |
|
|
224
|
+
| N | rearing cost C | Excel formula | `=Inputs!$B$<horizon row>` | C: rearing cost of one replacement, charged once, now (Inputs sheet). |
|
|
225
|
+
| O | REPLACE = S + V_keep(h) - C | Excel formula | `=Lr+Mr-Nr` | REPLACE: salvage used + value of the replacement heifer - rearing cost. |
|
|
226
|
+
| P | Delta Q = KEEP - REPLACE | Excel formula | `=Ir-Or` | Delta Q = KEEP - REPLACE. Positive: keeping is better; negative: replacing is better. |
|
|
227
|
+
| Q | break-even offer | Excel formula | `=Ir-Mr+Nr` | Break-even offer: the salvage at which KEEP and REPLACE are equal. |
|
|
228
|
+
| R | decision | Excel formula | `=IF(Lr>Qr,"CULL","KEEP")` | Decision: CULL when the salvage used is above the break-even offer, otherwise KEEP. |
|
|
229
|
+
| S | S source | Excel formula | `=IF(Kr="","estimate","offer")` | Which salvage was used: the estimate or a typed offer. |
|
|
230
|
+
| T | value from keep_cull_table (check) | check | | value_<N>y from keep_cull_table for the same cow and horizon (pasted from the model). |
|
|
231
|
+
| U | difference to check | check | `=Qr-Tr` | Break-even offer minus the reference value; must be 0 (to rounding). |
|
|
232
|
+
<!-- /GEN -->
|
|
233
|
+
|
|
234
|
+
### Economic scenarios
|
|
235
|
+
|
|
236
|
+
Prices and costs of the cow rewards (milk, feed, beef, calf, rearing, culling fee and the service costs) can be overridden for one run, without touching the model's defaults:
|
|
237
|
+
|
|
238
|
+
```python
|
|
239
|
+
from cowmrp.model import build_model
|
|
240
|
+
from cowmrp.domains.cow.economics import economic_overrides, reprice_model
|
|
241
|
+
from cowmrp.domains.cow.keepcull import keep_cull_table
|
|
242
|
+
|
|
243
|
+
model = build_model("cow")
|
|
244
|
+
with economic_overrides(MILK_PRICE_PER_KG=0.52, REARING_COST=2800):
|
|
245
|
+
scenario = reprice_model(model) # rewards recomputed; states, probabilities and the transition matrix are the same objects
|
|
246
|
+
table = keep_cull_table(herd, scenario) # keep/cull inside the block, so REARING_COST applies
|
|
247
|
+
```
|
|
248
|
+
|
|
249
|
+
Settings not named keep their defaults, and every value is restored when the block ends. The discount rate and the sale offer are already arguments of the keep/cull functions.
|
|
250
|
+
The override replaces module-level values, so it is **not thread-safe**: use it in sequential scripts and notebooks. The supported settings, with unit, currency and the source recorded in the repository, are in `cowmrp.domains.cow.economics.ECONOMIC_SETTINGS`. Biological and management values are not overridable.
|
|
251
|
+
|
|
252
|
+
## Tests
|
|
253
|
+
|
|
254
|
+
```
|
|
255
|
+
pytest # fast set, about 1 minute: expected-distribution checks, month and week
|
|
256
|
+
pytest -m slow # day resolution and weekly Monte Carlo parity, about 13 minutes
|
|
257
|
+
pytest -m install # builds the wheel and installs it in a clean virtual environment (needs network)
|
|
258
|
+
```
|
|
259
|
+
|
|
260
|
+
## Documentation
|
|
261
|
+
|
|
262
|
+
The site is built from the code by `python guide/build.py` and published to GitHub Pages by `.github/workflows/guide.yml` when `src/`, `guide/`,
|
|
263
|
+
`examples/`, `user_examples/`, `README.md` or `pyproject.toml` change on `main`.
|
|
264
|
+
|
|
265
|
+
## License and citation
|
|
266
|
+
|
|
267
|
+
Authors: Sumit Sharma (first author, primary developer), Edward Hansen and Miel Hostens (supervisor), Department of Animal Science, Cornell University, Ithaca, New York, USA.
|
|
268
|
+
CowMRP is released under the MIT licence (SPDX `MIT`), see `LICENSE`. Copyright (c) 2026 Bovi-analytics Lab, Cornell University. How to cite: see `CITATION.cff` (a Zenodo DOI will be added after the first public release).
|
|
269
|
+
|
|
270
|
+
Third-party material keeps its own licence: the Computer Modern (CMU) web fonts bundled with the documentation theme (`guide/web/theme/fonts/`, not in the installed package) are under the SIL Open Font License 1.1, whose text (`OFL.txt`) is kept with them; the interactive dashboards embed plotly.js (MIT), which keeps its own notice in the generated files. Dependencies such as numpy, pandas, scipy, matplotlib, plotly and openpyxl are installed from PyPI under their own licences and are not redistributed here.
|
|
271
|
+
|
|
272
|
+
## Notes
|
|
273
|
+
|
|
274
|
+
- The state reward vector does not contain `replacement_cost`; it is charged in the cow-versus-heifer comparison
|
|
275
|
+
(`replacement_comparison`, used by the keep/cull table), see the module docstring of `cowmrp.domains.cow.rewards`.
|
|
276
|
+
- Lookup tables can be overridden by CSV files (`src/cowmrp/domains/cow/data/README.md`).
|
|
277
|
+
- All generated files are written to `runners/<module>/outputs/`, which is not tracked.
|
cowmrp-0.1.0/README.md
ADDED
|
@@ -0,0 +1,230 @@
|
|
|
1
|
+
# CowMRP (`cowmrp`)
|
|
2
|
+
|
|
3
|
+
Markov reward process (MRP) models of dairy cattle, as a Python package. A **domain-free MRP engine** (`cowmrp.mrp`) plus two animal domains:
|
|
4
|
+
the **cow** (state `(lactation, dim, dp)`: lactation number, days in milk, days pregnant, in model-step units) and the **youngstock** chain
|
|
5
|
+
(calf, heifer, first calving). The engine knows no animal, so you can write your own domain.
|
|
6
|
+
|
|
7
|
+
It is for researchers and farm advisers who want to value cows (keep or replace), project a herd forward, and look at environmental outputs
|
|
8
|
+
beside economic ones. The scientific model is the reference implementation; the package changes how it is reached, not what it computes.
|
|
9
|
+
|
|
10
|
+
## Install
|
|
11
|
+
|
|
12
|
+
```
|
|
13
|
+
pip install cowmrp # engine + domains (numpy, pandas, scipy)
|
|
14
|
+
pip install "cowmrp[plot]" # + matplotlib, plotly: dashboards and projection figures
|
|
15
|
+
pip install -e ".[dev,plot,parquet]" # from a clone of this repository (Python 3.10 or newer)
|
|
16
|
+
```
|
|
17
|
+
|
|
18
|
+
The installed package contains the engine, the domains, the plotting helpers and the 200-cow tutorial herd. The runner scripts (`runners/`), `examples/`,
|
|
19
|
+
`user_examples/` (the example notebooks), the tests and the documentation sources are in the source repository, https://github.com/Bovi-analytics/cow-mrp; the commands in
|
|
20
|
+
"Running the scripts" and "Tests" below are for a clone.
|
|
21
|
+
|
|
22
|
+
## Quick start: a whole farm in a few lines
|
|
23
|
+
|
|
24
|
+
```python
|
|
25
|
+
from cowmrp.data import example_herd_path # 200 synthetic cows shipped with the package
|
|
26
|
+
from cowmrp.model import build_model
|
|
27
|
+
from cowmrp.domains.cow.farm import load_herd, project_herd
|
|
28
|
+
from cowmrp.viz.farm import write_farm_outputs
|
|
29
|
+
|
|
30
|
+
model = build_model("cow") # month resolution (see "Time resolution")
|
|
31
|
+
herd = load_herd(example_herd_path(), model) # or your own CSV / DataFrame
|
|
32
|
+
result = project_herd(herd, model, mode="expected", steps=24)
|
|
33
|
+
write_farm_outputs(result, "farm_results") # CSV tables + dashboard.html + dashboard_plotly.html
|
|
34
|
+
```
|
|
35
|
+
|
|
36
|
+
**Example notebooks.** The folder [`user_examples/`](user_examples/) holds five Jupyter notebooks that start with `pip install "cowmrp[plot,excel]"`, import only from the installed package and run on a CSV that you replace with your own: cows from a CSV with keep/cull, a farm projection with the interactive Plotly dashboard, individual rewards, a youngstock farm, and a domain of your own on the generic engine. They are not part of the installed package; get them with `git clone https://github.com/Bovi-analytics/cow-mrp.git` (or Download ZIP on GitHub) and see `user_examples/README.md` and the Examples page of the documentation site.
|
|
37
|
+
|
|
38
|
+
A complete script is `examples/whole_farm_example.py`; the same steps are walked through in the [tutorial](https://bovi-analytics.github.io/cow-mrp/tutorial.html) of the documentation site (online once the repository is public; the source is `guide/tutorial.html`).
|
|
39
|
+
|
|
40
|
+
## What the package does
|
|
41
|
+
|
|
42
|
+
| Task | Entry point | Notes |
|
|
43
|
+
|---|---|---|
|
|
44
|
+
| Build and solve a model | `cowmrp.model.build_model("cow" \| "youngstock")`, `evaluate`, `evolve_distribution` | Any domain that follows the interface in `cowmrp.domain` |
|
|
45
|
+
| **One cow, or cow A against cow B** | `cowmrp.domains.cow.projection.compare_cows(scenarios, years=2)` | Rewards and distributions over time for cows given as `lactation`, `dim_days`, `dp_days`; figures in `cowmrp.viz.projection` |
|
|
46
|
+
| **Keep or cull** | `cowmrp.domains.cow.keepcull.keep_cull_table(herd, model, years=(1,3,5), offer=900)` | Break-even sale value of every cow = value of keeping her minus value of replacing her now; a KEEP/CULL call only when you give an `offer`. Salvage is not credited in the model's own comparison (a documented limitation). `cowmrp.domains.cow.keepcull_excel.write_keepcull_workbook` (`--excel`, needs `cowmrp[excel]`) writes an Excel workbook that shows the sale of the cow now as its own column; see below |
|
|
47
|
+
| **Whole-farm projection** | `cowmrp.domains.cow.farm.project_herd(herd, model, mode, steps)` | Open / pregnant / dry, lactation, DIM stage; cows never replaced versus replaced; per-cow values |
|
|
48
|
+
| **Load farm records** | `cowmrp.domains.cow.farm.load_herd(csv_or_dataframe, model)` | Raw records or already-mapped `state_id`; every problem reported with row and `cow_id` |
|
|
49
|
+
| Environmental outputs | `cowmrp.domains.cow.rewards.environment` | The physical quantities (enteric and manure methane, N2O, nitrogen, CO2) are always computed by `compute_environment()` and do not depend on any price; the monetary environmental rewards are off by default and enter the economic total only if you enable them |
|
|
50
|
+
| Synthetic herd | `cowmrp.domains.cow.farm.generate_herd(model, mode, n_cows, seed)` | `equilibrium` (the model's steady-state herd), `mix`, `exact` |
|
|
51
|
+
|
|
52
|
+
### Farm data
|
|
53
|
+
|
|
54
|
+
One row per cow, described in **real days** (never health or DNB flags, which are not state variables):
|
|
55
|
+
|
|
56
|
+
| Column | Meaning |
|
|
57
|
+
|---|---|
|
|
58
|
+
| `cow_id` | optional, unique (1..n when absent) |
|
|
59
|
+
| `lactation` | 1 to 10 |
|
|
60
|
+
| `dim_days` | days in milk, 0 or more |
|
|
61
|
+
| `dp_days` | days pregnant, 0 for an open cow (default 0) |
|
|
62
|
+
| `state_id` | alternative: the model state of the cow, as written by `generate_herd` (if given together with the days they must agree) |
|
|
63
|
+
| `uim`, `up` | alternative to the days columns: the model's step indices |
|
|
64
|
+
|
|
65
|
+
A cow is mapped to a state exactly as the original runner `runners/cow/farm/generate.py` always did. `load_herd` adds checks only: missing columns, non-numeric
|
|
66
|
+
values, out-of-range lactation, negative days, duplicate `cow_id`, a combination the model has no state for (the message lists the nearest valid states), and a herd
|
|
67
|
+
file made under another time resolution. A DataFrame is accepted wherever a CSV path is.
|
|
68
|
+
|
|
69
|
+
### Tutorial dataset
|
|
70
|
+
|
|
71
|
+
`cowmrp.data.example_herd_path()` is a 200-cow **synthetic** herd (cows drawn from the model's own steady-state herd, seed 42, month resolution) in the model's schema;
|
|
72
|
+
`example_raw_herd_path()` is the same cows as a farmer would export them (`cow_id, lactation, dim_days, dp_days`). No real farm records are included.
|
|
73
|
+
|
|
74
|
+
### Time resolution
|
|
75
|
+
|
|
76
|
+
The model steps in `month` (30 d, default), `week` (7 d) or `day` (1 d). Set it **before Python starts**: `COW_TIME_UNIT=week python my_script.py`.
|
|
77
|
+
The resolution is not a display setting: it rebuilds the state space, the transition probabilities (rescaled to the step), the reward of every state (the rate integrated over the days
|
|
78
|
+
one step spans) and so the expected values. `dim` and `dp` are step indices of the chosen resolution, not calendar days; durations such as gestation stay in real days.
|
|
79
|
+
One process holds one resolution (the cow domain reads it once, when imported); `compare_cows` runs scenarios of another resolution in worker processes for you.
|
|
80
|
+
A herd file made at one resolution is refused at another; regenerate it. The day resolution is large: use it deliberately.
|
|
81
|
+
|
|
82
|
+
<!-- GEN:resolution_table -->
|
|
83
|
+
| Resolution | Step | States | Transitions | Example cow as `(dim, dp)` |
|
|
84
|
+
|---|---:|---:|---:|---:|
|
|
85
|
+
| month | 30 d | 2,970 | 9,040 | (4, 2) |
|
|
86
|
+
| week | 7 d | 42,290 | 131,197 | (20, 9) |
|
|
87
|
+
| day | 1 d | 1,952,060 | 6,109,989 | (149, 60) |
|
|
88
|
+
|
|
89
|
+
The example cow is in her second lactation, 150 days in milk and 60 days pregnant; `dim` and `dp` count model steps, so the same cow has different state coordinates at each resolution.
|
|
90
|
+
<!-- /GEN -->
|
|
91
|
+
|
|
92
|
+
<!-- GEN:state_counts -->
|
|
93
|
+
State counts (month resolution): 2,970 states, of which 1,210 are reachable from the replacement state (the default is the full table), 9,040 transitions.
|
|
94
|
+
<!-- /GEN -->
|
|
95
|
+
|
|
96
|
+
### Expected versus stochastic and Monte Carlo
|
|
97
|
+
|
|
98
|
+
`project_herd(..., mode=...)`:
|
|
99
|
+
|
|
100
|
+
| Mode | What it computes | When to use |
|
|
101
|
+
|---|---|---|
|
|
102
|
+
| `expected` (default) | The exact distribution of the herd after each step, split into cows never replaced and cows replaced at least once. Deterministic, fast | Almost always: planning, comparing scenarios, dashboards |
|
|
103
|
+
| `stochastic` | One sampled path per cow (`seed`) | To see a single plausible herd history |
|
|
104
|
+
| `monte_carlo` | Many sampled herds (`sims`), reported as mean, standard deviation, 5th and 95th percentile | When the spread, not only the mean, matters (small herds) |
|
|
105
|
+
|
|
106
|
+
For a large herd the three agree on the mean; Monte Carlo is the slow one, so it is never run unless you select it.
|
|
107
|
+
|
|
108
|
+
## Repository layout
|
|
109
|
+
|
|
110
|
+
| Folder | Purpose |
|
|
111
|
+
|---|---|
|
|
112
|
+
| `src/cowmrp/mrp/` | Domain-free MRP engine: `core.py` (math), `engine.py` (spec orchestration and validation), `simulator.py`, `utils.py` |
|
|
113
|
+
| `src/cowmrp/domain.py`, `model.py` | The domain interface (`load_domain`, `validate_domain`) and the generic build/solve API |
|
|
114
|
+
| `src/cowmrp/domains/` | One folder per domain: `cow/` (parameters, states, probabilities, transitions, rewards, lookup tables, evidence, `farm.py`, `keepcull.py`, `projection.py`) and `youngstock/` |
|
|
115
|
+
| `src/cowmrp/viz/`, `io/`, `data/` | Figures and dashboards, model-file export, the shipped example herd |
|
|
116
|
+
| `runners/` | Scripts that use the package: one module per domain (`cow/`, `youngstock/`) with `run.py`, `checks/`, `plots/`, `outputs/<kind>/`; `_shared/` common code |
|
|
117
|
+
| `examples/` | Four runnable scripts (`whole_farm_example.py`, ...) |
|
|
118
|
+
| `user_examples/` | Five example notebooks with their synthetic CSV files, a custom-domain folder and the installed-wheel test (`check_installed.py`); obtained from GitHub, not in the package |
|
|
119
|
+
| `tests/` | Fast default suite plus opt-in `slow` and `install` tests (see `tests/README.md`) |
|
|
120
|
+
| `guide/` | Documentation: the tutorial, a file guide and API reference built by `guide/build.py`, and the website in `guide/web/` |
|
|
121
|
+
|
|
122
|
+
Dependency direction is one-way: `cowmrp.mrp` <- `cowmrp.domain` <- `cowmrp.model` <- `cowmrp.domains` <- `runners`
|
|
123
|
+
(check: `python runners/_shared/checks/verify_mrp_boundary.py`). `runners/README.md` explains the runner modules.
|
|
124
|
+
|
|
125
|
+
**Add a domain:** write a package with `parameters`, `states`, `probabilities`, `transitions` (and optional `rewards`) following `cowmrp.domain`, then
|
|
126
|
+
`cowmrp.domain.load_domain("your_package")`. `python runners/_shared/new_domain.py <name>` creates the stubs and the runner module.
|
|
127
|
+
|
|
128
|
+
## Running the scripts (research workflows)
|
|
129
|
+
|
|
130
|
+
All scripts run from any directory. They are kept and call the package.
|
|
131
|
+
|
|
132
|
+
```
|
|
133
|
+
python runners/run_all.py # every module -> runners/<module>/outputs/
|
|
134
|
+
python runners/cow/run.py --review # review pages for a domain expert
|
|
135
|
+
python runners/cow/projection.py # edit SCENARIOS at the top first (cow A vs cow B)
|
|
136
|
+
python runners/cow/farm/generate.py # a synthetic herd -> runners/cow/outputs/farm/farm_data.csv
|
|
137
|
+
python runners/cow/farm/simulate.py --mode expected --steps 24 # herd projection and dashboards
|
|
138
|
+
python runners/cow/farm/keepcull.py --years 1,3,5 --offer 900 # keep/cull table
|
|
139
|
+
python runners/cow/farm/keepcull.py --excel # + keepcull_accounting.xlsx: KEEP, REPLACE, salvage, break-even per cow (needs cowmrp[excel])
|
|
140
|
+
python runners/cow/farm/keepcull.py --set REARING_COST=2800 # scenario run: override an economic setting for this run only (see below)
|
|
141
|
+
python runners/cow/checks/verify_farm.py # and the other verify_* scripts in runners/*/checks/
|
|
142
|
+
python runners/_shared/scale_benchmark.py --quick # cost and accuracy against herd size
|
|
143
|
+
```
|
|
144
|
+
|
|
145
|
+
### Keeping the documentation in step with the code
|
|
146
|
+
|
|
147
|
+
Numbers and tables that the documents repeat (state counts, reward components, overridable economic settings) are read from the package, not typed:
|
|
148
|
+
|
|
149
|
+
```
|
|
150
|
+
python guide/build_all.py # rewrite the generated blocks, build both sites, then check that nothing is stale
|
|
151
|
+
python guide/sync.py check # exit 1 if a block, token, number or built site is out of date
|
|
152
|
+
python guide/sync.py impact <file> # which documents and outputs depend on a source file
|
|
153
|
+
```
|
|
154
|
+
|
|
155
|
+
Generated blocks sit between `<!-- GEN:name -->` and `<!-- /GEN -->` in this file; do not edit inside them. `{{NAME}}` tokens in the tutorial and site pages are expanded at build time.
|
|
156
|
+
|
|
157
|
+
### Keep/cull accounting workbook
|
|
158
|
+
|
|
159
|
+
`python runners/cow/farm/keepcull.py --excel` also writes `keepcull_accounting.xlsx`: for every cow and horizon, the KEEP value, the REPLACE value, the immediate salvage (an indicative estimate, or the net offer you type for that cow), the break-even offer, ΔQ and the decision, as Excel formulas. It is a reporting output; the model, the reward vector and `keep_cull_table` are unchanged. The estimate is indicative (state-based body weight; the model's dressing percentage, beef price and culling fee) and a typed offer is the net cash received. The `Checks` sheet shows that the break-even column reproduces `keep_cull_table`. Terms, every column and formula, the treatment of future salvage and replacement costs, and the limitations are on the documentation site page *Keep/cull workbook* (`guide/web/content/keepcull-workbook.html`). The column table below is generated from the package:
|
|
160
|
+
|
|
161
|
+
<!-- GEN:excel_columns -->
|
|
162
|
+
| Col | Header | Kind | Formula (row r) | Meaning |
|
|
163
|
+
|---|---|---|---|---|
|
|
164
|
+
| A | cow_id | from the herd file | | Cow identifier from the herd file (row number if the herd has none). |
|
|
165
|
+
| B | state_id | from the herd file | | Model state of the cow (index into the state table of the current time resolution). |
|
|
166
|
+
| C | lactation | from the herd file | | Lactation number (copied when the herd has it). |
|
|
167
|
+
| D | dim_days | from the herd file | | Days in milk (copied when the herd has it). |
|
|
168
|
+
| E | dp_days | from the herd file | | Days pregnant (copied when the herd has it). |
|
|
169
|
+
| F | body_weight_kg (model, state-based) | model value (pasted) | | Body weight of the state in kg, computed as cull_reward computes it for the salvage value. A model estimate, not the weight on the day of sale. |
|
|
170
|
+
| G | V_total(s) | model value (pasted) | | V_total(s): expected discounted total reward of the cow over the horizon (milk, pregnancy, cull, mortality, feed, reproduction), future cull salvage included. |
|
|
171
|
+
| H | PV future replacement costs (s) | model value (pasted) | | PV_rc(s): expected discounted cost of the future replacements of this cow and of her replacements over the horizon. |
|
|
172
|
+
| I | KEEP = V_total - PV_rc | Excel formula | `=Gr-Hr` | KEEP: expected future value of keeping the cow, every future replacement cost paid. |
|
|
173
|
+
| J | estimated salvage S_est | Excel formula | `=Fr*Inputs!$B$3*Inputs!$B$2-Inputs!$B$4` | Estimated salvage now: body weight x dressing percentage x beef price - culling fee (Inputs B3, B2, B4). |
|
|
174
|
+
| K | actual net offer (type here, optional) | input (yellow) | | Actual net offer for this cow, typed by the user (optional). Net cash received: the culling fee is not subtracted again. |
|
|
175
|
+
| L | S used | Excel formula | `=IF(Kr="",Jr,Kr)` | Salvage used: the typed offer if there is one, otherwise the estimate. |
|
|
176
|
+
| M | V_keep(heifer) | Excel formula | `=Inputs!$E$<horizon row>` | V_keep(h): value of a fresh replacement heifer (state (1, 0, 0)) over the same horizon (Inputs sheet). |
|
|
177
|
+
| N | rearing cost C | Excel formula | `=Inputs!$B$<horizon row>` | C: rearing cost of one replacement, charged once, now (Inputs sheet). |
|
|
178
|
+
| O | REPLACE = S + V_keep(h) - C | Excel formula | `=Lr+Mr-Nr` | REPLACE: salvage used + value of the replacement heifer - rearing cost. |
|
|
179
|
+
| P | Delta Q = KEEP - REPLACE | Excel formula | `=Ir-Or` | Delta Q = KEEP - REPLACE. Positive: keeping is better; negative: replacing is better. |
|
|
180
|
+
| Q | break-even offer | Excel formula | `=Ir-Mr+Nr` | Break-even offer: the salvage at which KEEP and REPLACE are equal. |
|
|
181
|
+
| R | decision | Excel formula | `=IF(Lr>Qr,"CULL","KEEP")` | Decision: CULL when the salvage used is above the break-even offer, otherwise KEEP. |
|
|
182
|
+
| S | S source | Excel formula | `=IF(Kr="","estimate","offer")` | Which salvage was used: the estimate or a typed offer. |
|
|
183
|
+
| T | value from keep_cull_table (check) | check | | value_<N>y from keep_cull_table for the same cow and horizon (pasted from the model). |
|
|
184
|
+
| U | difference to check | check | `=Qr-Tr` | Break-even offer minus the reference value; must be 0 (to rounding). |
|
|
185
|
+
<!-- /GEN -->
|
|
186
|
+
|
|
187
|
+
### Economic scenarios
|
|
188
|
+
|
|
189
|
+
Prices and costs of the cow rewards (milk, feed, beef, calf, rearing, culling fee and the service costs) can be overridden for one run, without touching the model's defaults:
|
|
190
|
+
|
|
191
|
+
```python
|
|
192
|
+
from cowmrp.model import build_model
|
|
193
|
+
from cowmrp.domains.cow.economics import economic_overrides, reprice_model
|
|
194
|
+
from cowmrp.domains.cow.keepcull import keep_cull_table
|
|
195
|
+
|
|
196
|
+
model = build_model("cow")
|
|
197
|
+
with economic_overrides(MILK_PRICE_PER_KG=0.52, REARING_COST=2800):
|
|
198
|
+
scenario = reprice_model(model) # rewards recomputed; states, probabilities and the transition matrix are the same objects
|
|
199
|
+
table = keep_cull_table(herd, scenario) # keep/cull inside the block, so REARING_COST applies
|
|
200
|
+
```
|
|
201
|
+
|
|
202
|
+
Settings not named keep their defaults, and every value is restored when the block ends. The discount rate and the sale offer are already arguments of the keep/cull functions.
|
|
203
|
+
The override replaces module-level values, so it is **not thread-safe**: use it in sequential scripts and notebooks. The supported settings, with unit, currency and the source recorded in the repository, are in `cowmrp.domains.cow.economics.ECONOMIC_SETTINGS`. Biological and management values are not overridable.
|
|
204
|
+
|
|
205
|
+
## Tests
|
|
206
|
+
|
|
207
|
+
```
|
|
208
|
+
pytest # fast set, about 1 minute: expected-distribution checks, month and week
|
|
209
|
+
pytest -m slow # day resolution and weekly Monte Carlo parity, about 13 minutes
|
|
210
|
+
pytest -m install # builds the wheel and installs it in a clean virtual environment (needs network)
|
|
211
|
+
```
|
|
212
|
+
|
|
213
|
+
## Documentation
|
|
214
|
+
|
|
215
|
+
The site is built from the code by `python guide/build.py` and published to GitHub Pages by `.github/workflows/guide.yml` when `src/`, `guide/`,
|
|
216
|
+
`examples/`, `user_examples/`, `README.md` or `pyproject.toml` change on `main`.
|
|
217
|
+
|
|
218
|
+
## License and citation
|
|
219
|
+
|
|
220
|
+
Authors: Sumit Sharma (first author, primary developer), Edward Hansen and Miel Hostens (supervisor), Department of Animal Science, Cornell University, Ithaca, New York, USA.
|
|
221
|
+
CowMRP is released under the MIT licence (SPDX `MIT`), see `LICENSE`. Copyright (c) 2026 Bovi-analytics Lab, Cornell University. How to cite: see `CITATION.cff` (a Zenodo DOI will be added after the first public release).
|
|
222
|
+
|
|
223
|
+
Third-party material keeps its own licence: the Computer Modern (CMU) web fonts bundled with the documentation theme (`guide/web/theme/fonts/`, not in the installed package) are under the SIL Open Font License 1.1, whose text (`OFL.txt`) is kept with them; the interactive dashboards embed plotly.js (MIT), which keeps its own notice in the generated files. Dependencies such as numpy, pandas, scipy, matplotlib, plotly and openpyxl are installed from PyPI under their own licences and are not redistributed here.
|
|
224
|
+
|
|
225
|
+
## Notes
|
|
226
|
+
|
|
227
|
+
- The state reward vector does not contain `replacement_cost`; it is charged in the cow-versus-heifer comparison
|
|
228
|
+
(`replacement_comparison`, used by the keep/cull table), see the module docstring of `cowmrp.domains.cow.rewards`.
|
|
229
|
+
- Lookup tables can be overridden by CSV files (`src/cowmrp/domains/cow/data/README.md`).
|
|
230
|
+
- All generated files are written to `runners/<module>/outputs/`, which is not tracked.
|
|
@@ -0,0 +1,18 @@
|
|
|
1
|
+
"""One cow, or cow A against cow B -- expected value over two years, and where each cow is likely to be.
|
|
2
|
+
|
|
3
|
+
python examples/cow_comparison_example.py
|
|
4
|
+
|
|
5
|
+
Two second-lactation cows 150 days in milk, one open and one already 60 days pregnant, are projected forward with the model's own
|
|
6
|
+
transition matrix and rewards. Change the scenarios (lactation, dim_days, dp_days, time_unit) to compare other cows.
|
|
7
|
+
"""
|
|
8
|
+
from cowmrp.domains.cow.projection import compare_cows
|
|
9
|
+
|
|
10
|
+
scenarios = [
|
|
11
|
+
dict(label="open", time_unit="month", lactation=2, dim_days=150, dp_days=0),
|
|
12
|
+
dict(label="pregnant", time_unit="month", lactation=2, dim_days=150, dp_days=60),
|
|
13
|
+
]
|
|
14
|
+
results = compare_cows(scenarios, years=2) # {scenario index: result}; n_trajectories=100 adds Monte Carlo bands
|
|
15
|
+
|
|
16
|
+
print("Expected value over the next 2 years (USD, discounted):")
|
|
17
|
+
for i, sc in enumerate(scenarios):
|
|
18
|
+
print(f" {sc['label']:9s} {results[i]['finite_horizon_value']:8.0f}")
|
|
@@ -0,0 +1,33 @@
|
|
|
1
|
+
"""Keep or cull -- the break-even sale value of every cow in the example herd.
|
|
2
|
+
|
|
3
|
+
python examples/keep_cull_example.py
|
|
4
|
+
|
|
5
|
+
keep = expected future value of the cow, her future replacement costs included; replace = value of a heifer bought now, after one replacement
|
|
6
|
+
cost; value = keep - replace = the sale value above which culling and replacing her beats keeping her. No sale price is built into the model:
|
|
7
|
+
the offer below is an input you choose. The salvage value of a sale is not credited in the model's own comparison (documented limitation).
|
|
8
|
+
The last part repeats the table under a different rearing cost to show how a default is overridden for one block of code and then restored.
|
|
9
|
+
"""
|
|
10
|
+
from cowmrp.data import example_herd_path
|
|
11
|
+
from cowmrp.domains.cow.economics import ECONOMIC_SETTINGS, economic_overrides, reprice_model
|
|
12
|
+
from cowmrp.domains.cow.farm import load_herd
|
|
13
|
+
from cowmrp.domains.cow.keepcull import keep_cull_summary, keep_cull_table
|
|
14
|
+
from cowmrp.model import build_model
|
|
15
|
+
|
|
16
|
+
model = build_model("cow")
|
|
17
|
+
herd = load_herd(example_herd_path(), model)
|
|
18
|
+
table = keep_cull_table(herd, model, years=(1, 3, 5), offer=900) # offer = USD per cow
|
|
19
|
+
|
|
20
|
+
print("Weakest five cows at the 3-year horizon (value_3y = break-even sale value, USD):")
|
|
21
|
+
print(table.nsmallest(5, "value_3y")[["cow_id", "lactation", "dim_days", "value_3y", "decision_3y", "stable"]].round(0).to_string(index=False))
|
|
22
|
+
print("\nSummary at an offer of 900 USD:")
|
|
23
|
+
summary = keep_cull_summary(table, (1, 3, 5))
|
|
24
|
+
print(summary[summary["lactation"] == "all"].round(0).to_string(index=False))
|
|
25
|
+
|
|
26
|
+
# Scenario: the defaults are untouched outside the block. Anything not named keeps its default. 2,800 USD is an illustrative value, not a recommendation.
|
|
27
|
+
default_cost = ECONOMIC_SETTINGS["REARING_COST"]["default"]
|
|
28
|
+
with economic_overrides(REARING_COST=2800):
|
|
29
|
+
scenario = reprice_model(model) # same states and transition matrix; rewards recomputed
|
|
30
|
+
table_s = keep_cull_table(herd, scenario, years=(1, 3, 5), offer=900) # the keep/cull layer reads REARING_COST inside the block
|
|
31
|
+
n_default, n_scenario = int((table["decision_3y"] == "CULL").sum()), int((table_s["decision_3y"] == "CULL").sum())
|
|
32
|
+
print(f"\nCows to cull at 900 USD, 3-year horizon: {n_default} with the default rearing cost ({default_cost:,.0f} USD), {n_scenario} with 2,800 USD")
|
|
33
|
+
print(f"Median break-even value (3-year): {table['value_3y'].median():,.0f} -> {table_s['value_3y'].median():,.0f} USD")
|