bfflearn 0.2.1__tar.gz → 0.3.0__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (152) hide show
  1. {bfflearn-0.2.1 → bfflearn-0.3.0}/CHANGELOG.md +41 -0
  2. {bfflearn-0.2.1 → bfflearn-0.3.0}/CITATION.cff +2 -2
  3. {bfflearn-0.2.1 → bfflearn-0.3.0}/PKG-INFO +12 -2
  4. {bfflearn-0.2.1 → bfflearn-0.3.0}/README.md +10 -0
  5. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/__init__.py +1 -1
  6. bfflearn-0.3.0/bff/bayes/__init__.py +21 -0
  7. bfflearn-0.3.0/bff/bayes/effective_observations.py +31 -0
  8. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/gaussian_process.py +24 -7
  9. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/learning.py +21 -38
  10. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/likelihoods.py +62 -39
  11. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/cli.py +1 -118
  12. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/plotting.py +233 -0
  13. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/qoi/data.py +10 -3
  14. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/qoi/rdf.py +3 -3
  15. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/qoi/routines.py +5 -3
  16. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/topology.py +38 -11
  17. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/_shared/config.py +1 -22
  18. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/build/config.py +4 -4
  19. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/fit/config.py +6 -7
  20. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/fit/main.py +0 -4
  21. bfflearn-0.3.0/bff/workflows/learn/config.py +210 -0
  22. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/learn/main.py +90 -9
  23. {bfflearn-0.2.1 → bfflearn-0.3.0}/bfflearn.egg-info/PKG-INFO +12 -2
  24. {bfflearn-0.2.1 → bfflearn-0.3.0}/bfflearn.egg-info/SOURCES.txt +2 -1
  25. {bfflearn-0.2.1 → bfflearn-0.3.0}/bfflearn.egg-info/requires.txt +1 -1
  26. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/architecture.md +3 -2
  27. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/changelog.md +24 -1
  28. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/cli.md +11 -9
  29. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/configuration/analyze.md +3 -0
  30. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/configuration/build.md +3 -1
  31. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/configuration/fit.md +7 -4
  32. bfflearn-0.3.0/docs/configuration/learn.md +127 -0
  33. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/development.md +1 -1
  34. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/examples/acetate.md +1 -0
  35. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/examples/arbitrary-data.md +9 -1
  36. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/examples/neon-mie-lgpmd.md +12 -3
  37. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/installation.md +16 -0
  38. {bfflearn-0.2.1 → bfflearn-0.3.0}/pyproject.toml +2 -2
  39. bfflearn-0.3.0/tests/bayes/test_effective_observations.py +28 -0
  40. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/bayes/test_gaussian_process.py +61 -3
  41. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/bayes/test_learning_problem.py +9 -14
  42. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/bayes/test_likelihoods.py +73 -3
  43. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/qoi/test_data.py +34 -1
  44. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/test_examples.py +26 -0
  45. bfflearn-0.3.0/tests/test_plotting.py +128 -0
  46. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/test_topology.py +29 -1
  47. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/workflows/test_configs.py +206 -4
  48. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/workflows/test_regressions.py +48 -7
  49. bfflearn-0.2.1/bff/bayes/__init__.py +0 -11
  50. bfflearn-0.2.1/bff/data.py +0 -19
  51. bfflearn-0.2.1/bff/workflows/learn/config.py +0 -114
  52. bfflearn-0.2.1/docs/configuration/learn.md +0 -78
  53. bfflearn-0.2.1/tests/test_plotting.py +0 -53
  54. {bfflearn-0.2.1 → bfflearn-0.3.0}/CODE_OF_CONDUCT.md +0 -0
  55. {bfflearn-0.2.1 → bfflearn-0.3.0}/CONTRIBUTING.md +0 -0
  56. {bfflearn-0.2.1 → bfflearn-0.3.0}/LICENSE +0 -0
  57. {bfflearn-0.2.1 → bfflearn-0.3.0}/MANIFEST.in +0 -0
  58. {bfflearn-0.2.1 → bfflearn-0.3.0}/SECURITY.md +0 -0
  59. {bfflearn-0.2.1 → bfflearn-0.3.0}/SUPPORT.md +0 -0
  60. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/kernels.py +0 -0
  61. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/means.py +0 -0
  62. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/posterior.py +0 -0
  63. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/priors.py +0 -0
  64. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/results.py +0 -0
  65. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/bayes/utils.py +0 -0
  66. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/domain/__init__.py +0 -0
  67. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/domain/bias.py +0 -0
  68. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/domain/sample.py +0 -0
  69. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/domain/specs.py +0 -0
  70. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/__init__.py +0 -0
  71. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/colvars.py +0 -0
  72. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/commands.py +0 -0
  73. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/cp2k.py +0 -0
  74. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/extxyz.py +0 -0
  75. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/logs.py +0 -0
  76. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/mdp.py +0 -0
  77. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/plumed.py +0 -0
  78. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/progress.py +0 -0
  79. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/schedulers.py +0 -0
  80. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/io/utils.py +0 -0
  81. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/mcmc/__init__.py +0 -0
  82. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/mcmc/convergence.py +0 -0
  83. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/mcmc/proposal.py +0 -0
  84. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/mcmc/sampler.py +0 -0
  85. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/qoi/__init__.py +0 -0
  86. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/qoi/analysis.py +0 -0
  87. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/qoi/hbonds.py +0 -0
  88. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/tools.py +0 -0
  89. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/__init__.py +0 -0
  90. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/_shared/__init__.py +0 -0
  91. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/_shared/campaign.py +0 -0
  92. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/_shared/preparation.py +0 -0
  93. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/_shared/scheduler.py +0 -0
  94. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/analyze/__init__.py +0 -0
  95. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/analyze/__main__.py +0 -0
  96. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/analyze/config.py +0 -0
  97. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/analyze/main.py +0 -0
  98. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/build/__init__.py +0 -0
  99. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/build/__main__.py +0 -0
  100. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/build/main.py +0 -0
  101. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/evaluate_snapshots/__init__.py +0 -0
  102. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/evaluate_snapshots/__main__.py +0 -0
  103. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/evaluate_snapshots/config.py +0 -0
  104. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/evaluate_snapshots/main.py +0 -0
  105. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/examples.py +0 -0
  106. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/fit/__init__.py +0 -0
  107. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/fit/__main__.py +0 -0
  108. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/learn/__init__.py +0 -0
  109. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/learn/__main__.py +0 -0
  110. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/md/__init__.py +0 -0
  111. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/md/__main__.py +0 -0
  112. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/md/config.py +0 -0
  113. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/md/main.py +0 -0
  114. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/prepare_assets/__init__.py +0 -0
  115. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/prepare_assets/__main__.py +0 -0
  116. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/prepare_assets/config.py +0 -0
  117. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/prepare_assets/main.py +0 -0
  118. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/sample/__init__.py +0 -0
  119. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/sample/__main__.py +0 -0
  120. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/sample/config.py +0 -0
  121. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/sample/main.py +0 -0
  122. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/validate/__init__.py +0 -0
  123. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/validate/__main__.py +0 -0
  124. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/validate/config.py +0 -0
  125. {bfflearn-0.2.1 → bfflearn-0.3.0}/bff/workflows/validate/main.py +0 -0
  126. {bfflearn-0.2.1 → bfflearn-0.3.0}/bfflearn.egg-info/dependency_links.txt +0 -0
  127. {bfflearn-0.2.1 → bfflearn-0.3.0}/bfflearn.egg-info/entry_points.txt +0 -0
  128. {bfflearn-0.2.1 → bfflearn-0.3.0}/bfflearn.egg-info/top_level.txt +0 -0
  129. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/assets/bff-logo.pdf +0 -0
  130. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/assets/bff-logo.svg +0 -0
  131. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/configuration/evaluate-snapshots.md +0 -0
  132. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/configuration/prepare-assets.md +0 -0
  133. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/configuration/sample.md +0 -0
  134. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/configuration/validate.md +0 -0
  135. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/examples/index.md +0 -0
  136. {bfflearn-0.2.1 → bfflearn-0.3.0}/docs/index.md +0 -0
  137. {bfflearn-0.2.1 → bfflearn-0.3.0}/mkdocs.yml +0 -0
  138. {bfflearn-0.2.1 → bfflearn-0.3.0}/setup.cfg +0 -0
  139. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/bayes/test_priors_posterior.py +0 -0
  140. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/bayes/test_results.py +0 -0
  141. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/bayes/test_utils.py +0 -0
  142. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/domain/test_specs.py +0 -0
  143. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/io/_fake_mdanalysis.py +0 -0
  144. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/io/test_cp2k.py +0 -0
  145. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/io/test_cp2k_collect.py +0 -0
  146. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/io/test_logs.py +0 -0
  147. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/io/test_mdp.py +0 -0
  148. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/mcmc/test_convergence.py +0 -0
  149. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/mcmc/test_proposal.py +0 -0
  150. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/mcmc/test_sampler.py +0 -0
  151. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/workflows/test_charge_constraints.py +0 -0
  152. {bfflearn-0.2.1 → bfflearn-0.3.0}/tests/workflows/test_scheduler.py +0 -0
@@ -2,6 +2,47 @@
2
2
 
3
3
  ## Unreleased
4
4
 
5
+ ## `0.3.0` - 2026-06-11
6
+
7
+ ### Breaking Changes
8
+
9
+ - Moved effective-observation configuration from `bff fit` to each model under
10
+ `bff learn`. Model entries now require `model_path` and exactly one strategy:
11
+ explicit `n_eff`, `independent_observations: true`, or curve inference with
12
+ `tolerance`.
13
+ - Changed `.lgp` serialization to store reference-curve metadata. Models
14
+ written by earlier releases must be refitted before use with 0.3.0.
15
+
16
+ ### Added
17
+
18
+ - Added prominence-based effective-observation estimates for smooth,
19
+ curve-valued QoIs while retaining direct counts for independent scalar data.
20
+ - Added `qoi-marginals.pdf`, which colors posterior parameter marginals by the
21
+ local contribution of each QoI and shows priors and parameter bounds.
22
+ - Added per-QoI Gaussian log-likelihood evaluation for post-processing thinned
23
+ posterior samples.
24
+
25
+ ### Changed
26
+
27
+ - Made the Gaussian likelihood insensitive to curve binning by applying
28
+ `n_eff` to a mean squared residual rather than treating every bin as an
29
+ independent observation.
30
+ - Allowed surrogate models for different QoIs to use different numbers of
31
+ training rows, provided their parameter dimensions agree.
32
+ - Simplified the `bff` command entry point and shell completion while retaining
33
+ subcommands such as `bff build`.
34
+ - Made build templates optional for systems that do not require molecule
35
+ templates.
36
+ - Updated the packaged examples and documentation for effective observations
37
+ and QoI-attributed posterior plots.
38
+ - Included `pytest` in the standard installation so packaged test and example
39
+ checks are available without installing the developer extra.
40
+
41
+ ### Fixed
42
+
43
+ - Fixed multiprocessing imports for analysis routines loaded from user Python
44
+ files.
45
+
5
46
  ## `0.2.1` - 2026-06-01
6
47
 
7
48
  ### Fixed
@@ -1,8 +1,8 @@
1
1
  cff-version: 1.2.0
2
2
  message: "If you use BFF, please cite the repository and the associated paper."
3
3
  title: "Bayesic Force Fields"
4
- version: "0.2.1"
5
- date-released: "2026-06-01"
4
+ version: "0.3.0"
5
+ date-released: "2026-06-11"
6
6
  authors:
7
7
  - family-names: Kostal
8
8
  given-names: Vojtech
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: bfflearn
3
- Version: 0.2.1
3
+ Version: 0.3.0
4
4
  Summary: Bayesian workflows for fixed-charge molecular force fields
5
5
  Author-email: Vojtech Kostal <vojtakostal8@gmail.com>
6
6
  License-Expression: GPL-3.0-only
@@ -28,6 +28,7 @@ Requires-Dist: gmxtopology>=0.1.1
28
28
  Requires-Dist: matplotlib>=3.8
29
29
  Requires-Dist: MDAnalysis>=2.7
30
30
  Requires-Dist: numpy>=1.26
31
+ Requires-Dist: pytest>=8
31
32
  Requires-Dist: PyYAML>=6.0
32
33
  Requires-Dist: scipy>=1.11
33
34
  Requires-Dist: typer>=0.12
@@ -39,7 +40,6 @@ Requires-Dist: jupyterlab; extra == "notebook"
39
40
  Requires-Dist: notebook; extra == "notebook"
40
41
  Provides-Extra: dev
41
42
  Requires-Dist: build; extra == "dev"
42
- Requires-Dist: pytest; extra == "dev"
43
43
  Requires-Dist: ruff; extra == "dev"
44
44
  Requires-Dist: twine; extra == "dev"
45
45
  Provides-Extra: docs
@@ -64,6 +64,10 @@ molecular force-field parameters from molecular-dynamics observables. It
64
64
  coordinates simulation campaigns, trajectory analysis, surrogate fitting, and
65
65
  Bayesian posterior learning.
66
66
 
67
+ Posterior learning supports explicit or data-derived effective observation
68
+ counts for scalar and curve-valued observables. It also writes QoI-attributed
69
+ marginals that show which observables support different posterior regions.
70
+
67
71
  Full documentation:
68
72
  [vojtechkostal.github.io/BayesicForceFields](https://vojtechkostal.github.io/BayesicForceFields/)
69
73
 
@@ -91,6 +95,12 @@ mamba activate bfflearn
91
95
  pip install bfflearn
92
96
  ```
93
97
 
98
+ For Jupyter notebooks, install the optional notebook tools:
99
+
100
+ ```bash
101
+ pip install "bfflearn[notebook]"
102
+ ```
103
+
94
104
  > **WARNING:** GPU-enabled PyTorch must be installed separately. Use the
95
105
  > [official PyTorch installation selector](https://pytorch.org/get-started/locally/)
96
106
  > to choose the command matching your CUDA version before running fitting,
@@ -14,6 +14,10 @@ molecular force-field parameters from molecular-dynamics observables. It
14
14
  coordinates simulation campaigns, trajectory analysis, surrogate fitting, and
15
15
  Bayesian posterior learning.
16
16
 
17
+ Posterior learning supports explicit or data-derived effective observation
18
+ counts for scalar and curve-valued observables. It also writes QoI-attributed
19
+ marginals that show which observables support different posterior regions.
20
+
17
21
  Full documentation:
18
22
  [vojtechkostal.github.io/BayesicForceFields](https://vojtechkostal.github.io/BayesicForceFields/)
19
23
 
@@ -41,6 +45,12 @@ mamba activate bfflearn
41
45
  pip install bfflearn
42
46
  ```
43
47
 
48
+ For Jupyter notebooks, install the optional notebook tools:
49
+
50
+ ```bash
51
+ pip install "bfflearn[notebook]"
52
+ ```
53
+
44
54
  > **WARNING:** GPU-enabled PyTorch must be installed separately. Use the
45
55
  > [official PyTorch installation selector](https://pytorch.org/get-started/locally/)
46
56
  > to choose the command matching your CUDA version before running fitting,
@@ -6,7 +6,7 @@ from importlib import import_module
6
6
  from pathlib import Path
7
7
  from typing import Any
8
8
 
9
- __version__ = "0.2.1"
9
+ __version__ = "0.3.0"
10
10
 
11
11
  _INTERNAL_MODULES = {
12
12
  "bayes",
@@ -0,0 +1,21 @@
1
+ from . import (
2
+ effective_observations,
3
+ gaussian_process,
4
+ kernels,
5
+ learning,
6
+ likelihoods,
7
+ priors,
8
+ results,
9
+ utils,
10
+ )
11
+
12
+ __all__ = [
13
+ 'effective_observations',
14
+ 'gaussian_process',
15
+ 'kernels',
16
+ 'learning',
17
+ 'likelihoods',
18
+ 'priors',
19
+ 'results',
20
+ 'utils',
21
+ ]
@@ -0,0 +1,31 @@
1
+ from __future__ import annotations
2
+
3
+ import numpy as np
4
+ from scipy.signal import find_peaks, savgol_filter
5
+
6
+
7
+ def estimate_curve_n_eff(y: np.ndarray, *, tolerance: float) -> float:
8
+ """Estimate an effective number of resolved features in one smooth curve."""
9
+ y = np.asarray(y, dtype=float)
10
+ tolerance = float(tolerance)
11
+
12
+ if y.ndim != 1 or y.size < 11:
13
+ raise ValueError("curve must be one-dimensional with at least 11 values.")
14
+ if not np.all(np.isfinite(y)):
15
+ raise ValueError("curve must contain only finite values.")
16
+ if not np.isfinite(tolerance) or tolerance <= 0.0:
17
+ raise ValueError("tolerance must be positive and finite.")
18
+
19
+ window_length = min(15, y.size if y.size % 2 else y.size - 1)
20
+ y_smooth = savgol_filter(y, window_length=window_length, polyorder=3)
21
+
22
+ noise = np.median(np.abs(y - y_smooth))
23
+ prominence_threshold = max(5.0 * noise, np.finfo(float).eps)
24
+
25
+ n_eff = 0.0
26
+ for sign in (1.0, -1.0):
27
+ _, properties = find_peaks(sign * y_smooth, prominence=prominence_threshold)
28
+ for prominence in properties["prominences"]:
29
+ n_eff += max(1.0, np.log(float(prominence) / tolerance))
30
+
31
+ return max(1.0, float(n_eff))
@@ -166,10 +166,10 @@ class LGPCommittee:
166
166
  ----------
167
167
  lgps : list of LocalGaussianProcess
168
168
  List of LGP models.
169
- n_observations : int
170
- Effective number of observations used in the likelihood term.
171
169
  reference_values : np.ndarray
172
170
  Reference observation vector matched by the surrogate outputs.
171
+ n_curves : int
172
+ Number of curves represented by the flattened reference vector.
173
173
  nuisance : float, optional
174
174
  Nuisance parameter for model selection (default is None).
175
175
  stochastic : bool, optional
@@ -195,15 +195,16 @@ class LGPCommittee:
195
195
  def __init__(
196
196
  self,
197
197
  lgps: list[LocalGaussianProcess],
198
- n_observations: int,
199
198
  reference_values: np.ndarray,
199
+ n_curves: int,
200
200
  nuisance: float | None = None,
201
201
  stochastic: bool = False
202
202
  ) -> None:
203
203
  self.lgps = lgps
204
204
  self.error: float | None = None
205
- self.n_observations = int(n_observations)
206
205
  self.reference_values = np.asarray(reference_values, dtype=float).reshape(-1)
206
+ self.n_curves = int(n_curves)
207
+ self.n_eff = float(self.reference_values.size)
207
208
  self.nuisance = nuisance
208
209
  self.stochastic = stochastic
209
210
 
@@ -211,6 +212,12 @@ class LGPCommittee:
211
212
  raise ValueError(
212
213
  "Reference observation size does not match surrogate output size."
213
214
  )
215
+ if self.n_curves <= 0:
216
+ raise ValueError("'n_curves' must be a positive integer.")
217
+ if self.reference_values.size % self.n_curves != 0:
218
+ raise ValueError(
219
+ "Reference observation size must be divisible by 'n_curves'."
220
+ )
214
221
 
215
222
  @property
216
223
  def size(self) -> int:
@@ -224,6 +231,10 @@ class LGPCommittee:
224
231
  def y_size(self) -> int:
225
232
  return self.lgps[0].y_size
226
233
 
234
+ @property
235
+ def curve_length(self) -> int:
236
+ return int(self.reference_values.size // self.n_curves)
237
+
227
238
  def predict(self, X: torch.Tensor) -> torch.Tensor:
228
239
  """
229
240
  Predict outputs by averaging predictions from all LGP models.
@@ -272,11 +283,17 @@ class LGPCommittee:
272
283
  @classmethod
273
284
  def load(cls: type[LGPCommitteeT], fn: PathLike) -> LGPCommitteeT:
274
285
  state = torch.load(fn, weights_only=False)
286
+ if "n_curves" not in state:
287
+ raise ValueError(
288
+ "This surrogate model predates BFF 0.3.0 and does not contain "
289
+ "reference-curve metadata. Refit the model before learning."
290
+ )
275
291
  lgps = [LocalGaussianProcess(**lgp_state) for lgp_state in state["lgps"]]
292
+ reference_values = np.asarray(state["reference_values"], dtype=float)
276
293
  committee = cls(
277
294
  lgps=lgps,
278
- n_observations=state["n_observations"],
279
- reference_values=np.asarray(state["reference_values"], dtype=float),
295
+ reference_values=reference_values,
296
+ n_curves=int(state["n_curves"]),
280
297
  nuisance=state["nuisance"],
281
298
  stochastic=state["stochastic"],
282
299
  )
@@ -286,8 +303,8 @@ class LGPCommittee:
286
303
  def write(self, fn_out: PathLike) -> None:
287
304
 
288
305
  state = {
289
- "n_observations": self.n_observations,
290
306
  "reference_values": self.reference_values.tolist(),
307
+ "n_curves": self.n_curves,
291
308
  "nuisance": self.nuisance,
292
309
  "stochastic": self.stochastic,
293
310
  "error": self.error,
@@ -78,20 +78,6 @@ class LearningProblem:
78
78
  + ", ".join(sorted(inconsistent_inputs))
79
79
  )
80
80
 
81
- model_input_shapes = {
82
- qoi: tuple(int(dim) for dim in model.lgps[0].X_train.shape)
83
- for qoi, model in self.models.items()
84
- }
85
- if len(set(model_input_shapes.values())) != 1:
86
- shape_summary = ", ".join(
87
- f"{qoi}={shape}"
88
- for qoi, shape in sorted(model_input_shapes.items())
89
- )
90
- raise ValueError(
91
- "All surrogate models must expect the same training input shape. "
92
- f"Found: {shape_summary}"
93
- )
94
-
95
81
  n_params = {model.n_params for model in self.models.values()}
96
82
  if len(n_params) != 1:
97
83
  raise ValueError("All surrogate models must have the same input dimension.")
@@ -114,6 +100,17 @@ class LearningProblem:
114
100
  + ", ".join(sorted(invalid))
115
101
  )
116
102
 
103
+ invalid_curve_schema = []
104
+ for qoi, model in self.models.items():
105
+ n_curves = int(getattr(model, "n_curves", 0))
106
+ if n_curves <= 0 or model.reference_values.size % n_curves != 0:
107
+ invalid_curve_schema.append(qoi)
108
+ if invalid_curve_schema:
109
+ raise ValueError(
110
+ "Models with incompatible n_curves metadata: "
111
+ + ", ".join(sorted(invalid_curve_schema))
112
+ )
113
+
117
114
  @property
118
115
  def qoi_names(self) -> list[str]:
119
116
  return list(self.models)
@@ -306,8 +303,8 @@ def fit_lgp_committee(
306
303
  test_fraction: float,
307
304
  n_hyper: int,
308
305
  committee: int,
309
- n_observations: int,
310
306
  reference_values: np.ndarray,
307
+ n_curves: int,
311
308
  nuisance: float | None,
312
309
  fn_out: PathLike | None,
313
310
  device: str,
@@ -395,10 +392,10 @@ def fit_lgp_committee(
395
392
  logger.status("Committee", f"{i}/{committee}", level=2, overwrite=True)
396
393
 
397
394
  lgp_committee = LGPCommittee(
398
- lgps,
399
- n_observations,
400
- reference_values,
401
- nuisance,
395
+ lgps=lgps,
396
+ reference_values=reference_values,
397
+ n_curves=n_curves,
398
+ nuisance=nuisance,
402
399
  )
403
400
  lgp_committee.validate(X_test, y_test)
404
401
  logger.done(
@@ -413,20 +410,11 @@ def fit_lgp_committee(
413
410
  return lgp_committee
414
411
 
415
412
 
416
- def _effective_observations(
417
- dataset: QoIDataset,
418
- observation_scale: float = 1.0,
419
- ) -> int:
420
- """Return the effective observation count used in the likelihood term."""
421
- return max(1, int(round(dataset.n_observations * float(observation_scale))))
422
-
423
-
424
413
  def fit_surrogates(
425
414
  datasets: Sequence[QoIDataset],
426
415
  *,
427
416
  y_means: Optional[Mapping[str, MeanFunction | str]] = None,
428
417
  hyperpriors: Optional[Mapping[str, Priors | Sequence]] = None,
429
- observation_scales: Optional[Mapping[str, float]] = None,
430
418
  model_paths: Optional[Mapping[str, PathLike | None]] = None,
431
419
  reuse_models: bool = True,
432
420
  n_hyper_max: int = 200,
@@ -441,7 +429,6 @@ def fit_surrogates(
441
429
  logger = logger or Logger("fit")
442
430
  y_means = dict(y_means or {})
443
431
  hyperpriors = dict(hyperpriors or {})
444
- observation_scales = dict(observation_scales or {})
445
432
  model_paths = dict(model_paths or {})
446
433
 
447
434
  if owns_logger:
@@ -456,10 +443,6 @@ def fit_surrogates(
456
443
  )
457
444
 
458
445
  qoi = dataset.name
459
- n_observations = _effective_observations(
460
- dataset,
461
- observation_scales.get(qoi, 1.0),
462
- )
463
446
  logger.info(f"QoI {qoi}", level=1)
464
447
 
465
448
  fn_model_raw = model_paths.get(qoi)
@@ -469,19 +452,20 @@ def fit_surrogates(
469
452
 
470
453
  if reuse_models and fn_model is not None and fn_model.exists():
471
454
  models[qoi] = LGPCommittee.load(fn_model)
472
- models[qoi].n_observations = n_observations
473
455
  models[qoi].reference_values = np.asarray(
474
456
  dataset.outputs_ref,
475
457
  dtype=float,
476
458
  ).reshape(-1)
459
+ models[qoi].n_eff = float(models[qoi].reference_values.size)
460
+ models[qoi].n_curves = dataset.n_curves
477
461
  if models[qoi].reference_values.size != models[qoi].y_size:
478
462
  raise ValueError(
479
463
  f"Cached surrogate for {qoi!r} is incompatible with the "
480
464
  "current reference observation size."
481
465
  )
466
+ models[qoi].write(fn_model)
482
467
  logger.info(
483
- f"Using cached surrogate model. | obs = {n_observations} "
484
- f"| MAPE = {models[qoi].error:.2f}",
468
+ f"Using cached surrogate model. | MAPE = {models[qoi].error:.2f}",
485
469
  level=2,
486
470
  )
487
471
  logger.blank()
@@ -494,8 +478,8 @@ def fit_surrogates(
494
478
  test_fraction=test_fraction,
495
479
  n_hyper=n_hyper_max,
496
480
  committee=committee_size,
497
- n_observations=n_observations,
498
481
  reference_values=dataset.outputs_ref,
482
+ n_curves=dataset.n_curves,
499
483
  nuisance=dataset.nuisance,
500
484
  fn_out=fn_model,
501
485
  device=device,
@@ -503,7 +487,6 @@ def fit_surrogates(
503
487
  opt_kwargs=opt_kwargs,
504
488
  hyperpriors=hyperpriors.get(qoi),
505
489
  )
506
- logger.kv("Effective observations", n_observations, level=2)
507
490
  logger.blank()
508
491
 
509
492
  return models
@@ -10,6 +10,42 @@ if TYPE_CHECKING:
10
10
  from .learning import LearningProblem
11
11
 
12
12
 
13
+ def _split_parameters_and_sigmas(
14
+ theta: torch.Tensor,
15
+ problem: LearningProblem,
16
+ ) -> tuple[torch.Tensor, torch.Tensor]:
17
+ """Return explicit parameters and one scalar nuisance per QoI."""
18
+ n_params = problem.n_params
19
+ params = theta[:, :n_params]
20
+ nuisance_free = theta[:, n_params:]
21
+
22
+ sigma_columns = []
23
+ j = 0
24
+ for model in problem.models.values():
25
+ if model.nuisance is None:
26
+ sigma_columns.append(nuisance_free[:, j].exp())
27
+ j += 1
28
+ else:
29
+ sigma_columns.append(
30
+ torch.full(
31
+ (len(theta),),
32
+ float(model.nuisance),
33
+ device=theta.device,
34
+ dtype=theta.dtype,
35
+ )
36
+ )
37
+ return params, torch.stack(sigma_columns, dim=1)
38
+
39
+
40
+ def _valid_parameter_mask(
41
+ params: torch.Tensor,
42
+ problem: LearningProblem,
43
+ ) -> torch.Tensor:
44
+ if problem.constraint is None:
45
+ return torch.ones(len(params), dtype=bool, device=params.device)
46
+ return problem.constraint(params)
47
+
48
+
13
49
  def loo_log_likelihood(
14
50
  theta: torch.Tensor, X: torch.Tensor, y: torch.Tensor
15
51
  ) -> torch.Tensor:
@@ -103,50 +139,37 @@ def gaussian_log_likelihood(
103
139
  Log likelihood for each sample in `theta`, shape (n_samples,).
104
140
  """
105
141
 
106
- # Assign device
107
- device = theta.device
108
-
109
- # Split the theta into parameters and nuisances
110
- n_params = problem.n_params
111
- params = theta[:, :n_params]
112
- nuisance_free = theta[:, n_params:]
142
+ contributions = gaussian_log_likelihood_by_qoi(theta, problem)
143
+ return torch.stack(tuple(contributions.values())).sum(dim=0)
113
144
 
114
- sigma_columns = []
115
- j = 0
116
- for model in problem.models.values():
117
- if model.nuisance is None:
118
- sigma_columns.append(nuisance_free[:, j].exp())
119
- j += 1
120
- else:
121
- sigma_columns.append(
122
- torch.full(
123
- (len(theta),),
124
- float(model.nuisance),
125
- device=device,
126
- dtype=theta.dtype,
127
- )
128
- )
129
- sigmas = torch.stack(sigma_columns, dim=1)
130
145
 
131
- # Check if the parameters are within the valid bounds
132
- if problem.constraint is not None:
133
- mask = problem.constraint(params)
134
- else:
135
- mask = torch.ones(len(theta), dtype=bool, device=device)
146
+ def gaussian_log_likelihood_by_qoi(
147
+ theta: torch.Tensor,
148
+ problem: LearningProblem,
149
+ ) -> dict[str, torch.Tensor]:
150
+ """Compute one Gaussian log-likelihood contribution per QoI."""
151
+ params, sigmas = _split_parameters_and_sigmas(theta, problem)
152
+ mask = _valid_parameter_mask(params, problem)
136
153
  params = params[mask]
137
154
  sigmas = sigmas[mask]
138
155
 
139
- # Compute the log-likelihod
140
- log_likelihood = torch.full((len(theta), ), -torch.inf, device=device)
141
- if mask.any():
142
- log_like_valid = torch.zeros(mask.sum(), device=device)
143
- for (qoi, model), sigma in zip(problem.models.items(), sigmas.T):
156
+ contributions = {}
157
+ for (qoi, model), sigma in zip(problem.models.items(), sigmas.T):
158
+ contribution = torch.full(
159
+ (len(theta),),
160
+ -torch.inf,
161
+ device=theta.device,
162
+ dtype=theta.dtype,
163
+ )
164
+ if mask.any():
144
165
  y_trial = model.predict(params)
145
- N = model.n_observations
146
166
  diff = problem.observations[qoi] - y_trial
147
- ssq = torch.sum(diff**2, dim=1)
148
- log_like_valid += -0.5 * ssq / sigma**2 - N * torch.log(sigma)
149
-
150
- log_likelihood[mask] = log_like_valid
167
+ mse = torch.mean(diff**2, dim=1)
168
+ n_eff = float(model.n_eff)
169
+ contribution[mask] = (
170
+ -0.5 * n_eff * mse / sigma**2
171
+ - n_eff * torch.log(sigma)
172
+ )
173
+ contributions[qoi] = contribution
151
174
 
152
- return log_likelihood
175
+ return contributions
@@ -1,6 +1,5 @@
1
1
  """Command-line entry points for Bayesic Force Fields."""
2
2
 
3
- import os
4
3
  from pathlib import Path
5
4
  from typing import Callable
6
5
 
@@ -9,121 +8,10 @@ import typer
9
8
  app = typer.Typer(
10
9
  help="BayesicForceFields: Bayesian optimization of molecular force fields.",
11
10
  no_args_is_help=True,
12
- add_completion=False,
11
+ add_completion=True,
13
12
  )
14
13
 
15
14
  WorkflowMain = Callable[[Path], object]
16
- WORKFLOW_COMMANDS = (
17
- "build",
18
- "prepare-assets",
19
- "evaluate-snapshots",
20
- "sample",
21
- "analyze",
22
- "fit",
23
- "learn",
24
- "validate",
25
- "examples",
26
- )
27
- WORKFLOW_FILE_COMMANDS = tuple(
28
- command for command in WORKFLOW_COMMANDS if command != "examples"
29
- )
30
- WORKFLOW_FILE_COMMAND_PATTERN = "|".join(WORKFLOW_FILE_COMMANDS)
31
-
32
- _COMPLETION_ACTIVATE = f"""# Managed automatically by Bayesic Force Fields.
33
- if [ -n "${{BASH_VERSION-}}" ]; then
34
- _bff_completion() {{
35
- local cur prev words cword
36
- if declare -F _init_completion >/dev/null 2>&1; then
37
- _init_completion || return
38
- else
39
- COMPREPLY=()
40
- cur="${{COMP_WORDS[COMP_CWORD]}}"
41
- prev="${{COMP_WORDS[COMP_CWORD-1]}}"
42
- words=("${{COMP_WORDS[@]}}")
43
- cword=$COMP_CWORD
44
- fi
45
-
46
- local commands="{" ".join(WORKFLOW_COMMANDS)}"
47
- if [[ $COMP_CWORD -eq 1 ]]; then
48
- COMPREPLY=($(compgen -W "$commands" -- "$cur"))
49
- return 0
50
- fi
51
-
52
- case "${{COMP_WORDS[1]}}" in
53
- {WORKFLOW_FILE_COMMAND_PATTERN})
54
- COMPREPLY=($(compgen -f -- "$cur"))
55
- ;;
56
- *)
57
- COMPREPLY=()
58
- ;;
59
- esac
60
- }}
61
-
62
- complete -o nosort -F _bff_completion bff
63
- elif [ -n "${{ZSH_VERSION-}}" ]; then
64
- autoload -Uz compinit 2>/dev/null || true
65
- compinit -i 2>/dev/null || true
66
-
67
- _bff_completion() {{
68
- local -a commands
69
- commands=({" ".join(WORKFLOW_COMMANDS)})
70
-
71
- if (( CURRENT == 2 )); then
72
- compadd -- "${{commands[@]}}"
73
- return 0
74
- fi
75
-
76
- case "${{words[2]}}" in
77
- {WORKFLOW_FILE_COMMAND_PATTERN})
78
- _files
79
- ;;
80
- esac
81
- }}
82
-
83
- compdef _bff_completion bff
84
- fi
85
- """
86
-
87
- _COMPLETION_DEACTIVATE = """# Managed automatically by Bayesic Force Fields.
88
- if [ -n "${BASH_VERSION-}" ]; then
89
- complete -r bff 2>/dev/null || true
90
- elif [ -n "${ZSH_VERSION-}" ]; then
91
- unfunction _bff_completion 2>/dev/null || true
92
- compdef -d bff 2>/dev/null || true
93
- fi
94
- """
95
-
96
-
97
- def _conda_completion_hook_paths() -> tuple[Path, Path] | None:
98
- conda_prefix = os.environ.get("CONDA_PREFIX")
99
- if not conda_prefix:
100
- return None
101
-
102
- conda_dir = Path(conda_prefix) / "etc" / "conda"
103
- return (
104
- conda_dir / "activate.d" / "bff-completion.sh",
105
- conda_dir / "deactivate.d" / "bff-completion.sh",
106
- )
107
-
108
-
109
- def _write_if_changed(path: Path, content: str) -> None:
110
- if path.exists() and path.read_text() == content:
111
- return
112
- path.parent.mkdir(parents=True, exist_ok=True)
113
- path.write_text(content)
114
-
115
-
116
- def _ensure_completion_hooks() -> None:
117
- hook_paths = _conda_completion_hook_paths()
118
- if hook_paths is None:
119
- return
120
-
121
- activate_hook, deactivate_hook = hook_paths
122
- try:
123
- _write_if_changed(activate_hook, _COMPLETION_ACTIVATE)
124
- _write_if_changed(deactivate_hook, _COMPLETION_DEACTIVATE)
125
- except OSError:
126
- return
127
15
 
128
16
 
129
17
  def run_workflow(
@@ -155,11 +43,6 @@ def config_argument() -> Path:
155
43
  )
156
44
 
157
45
 
158
- @app.callback()
159
- def init_cli() -> None:
160
- _ensure_completion_hooks()
161
-
162
-
163
46
  @app.command()
164
47
  def version() -> None:
165
48
  from importlib import metadata