penicillium 0.1.0a0__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.
- penicillium-0.1.0a0/.gitignore +23 -0
- penicillium-0.1.0a0/CHANGELOG.md +63 -0
- penicillium-0.1.0a0/CITATION.cff +86 -0
- penicillium-0.1.0a0/LICENSE +58 -0
- penicillium-0.1.0a0/NOTICE +30 -0
- penicillium-0.1.0a0/PKG-INFO +156 -0
- penicillium-0.1.0a0/README.md +74 -0
- penicillium-0.1.0a0/docs/api.md +27 -0
- penicillium-0.1.0a0/docs/background.md +29 -0
- penicillium-0.1.0a0/docs/benchmarking.md +35 -0
- penicillium-0.1.0a0/docs/benchmarks.md +27 -0
- penicillium-0.1.0a0/docs/citing.md +40 -0
- penicillium-0.1.0a0/docs/cli.md +78 -0
- penicillium-0.1.0a0/docs/configuration.md +156 -0
- penicillium-0.1.0a0/docs/control.md +187 -0
- penicillium-0.1.0a0/docs/cookbook.md +108 -0
- penicillium-0.1.0a0/docs/dcs.md +148 -0
- penicillium-0.1.0a0/docs/differentiable.md +53 -0
- penicillium-0.1.0a0/docs/faq.md +121 -0
- penicillium-0.1.0a0/docs/faults.md +149 -0
- penicillium-0.1.0a0/docs/glossary.md +102 -0
- penicillium-0.1.0a0/docs/gymnasium.md +63 -0
- penicillium-0.1.0a0/docs/index.md +91 -0
- penicillium-0.1.0a0/docs/installation.md +74 -0
- penicillium-0.1.0a0/docs/licensing.md +102 -0
- penicillium-0.1.0a0/docs/measurements.md +218 -0
- penicillium-0.1.0a0/docs/model.md +299 -0
- penicillium-0.1.0a0/docs/process.md +112 -0
- penicillium-0.1.0a0/docs/quickstart.md +60 -0
- penicillium-0.1.0a0/docs/quirks.md +63 -0
- penicillium-0.1.0a0/docs/related.md +30 -0
- penicillium-0.1.0a0/docs/scenarios.md +170 -0
- penicillium-0.1.0a0/docs/solver_sensitivity.md +16 -0
- penicillium-0.1.0a0/docs/validation.md +70 -0
- penicillium-0.1.0a0/docs/validation_100_batches.md +69 -0
- penicillium-0.1.0a0/docs/validation_coverage.md +48 -0
- penicillium-0.1.0a0/docs/validation_octave.md +283 -0
- penicillium-0.1.0a0/docs/variables.md +101 -0
- penicillium-0.1.0a0/pyproject.toml +94 -0
- penicillium-0.1.0a0/src/penicillium/__init__.py +32 -0
- penicillium-0.1.0a0/src/penicillium/__main__.py +3 -0
- penicillium-0.1.0a0/src/penicillium/_numba.py +30 -0
- penicillium-0.1.0a0/src/penicillium/_paa_pls.py +220 -0
- penicillium-0.1.0a0/src/penicillium/_raman_reference.py +11 -0
- penicillium-0.1.0a0/src/penicillium/_version.py +1 -0
- penicillium-0.1.0a0/src/penicillium/batches.py +91 -0
- penicillium-0.1.0a0/src/penicillium/cli.py +151 -0
- penicillium-0.1.0a0/src/penicillium/constants.py +7 -0
- penicillium-0.1.0a0/src/penicillium/dcs/__init__.py +10 -0
- penicillium-0.1.0a0/src/penicillium/dcs/server.py +211 -0
- penicillium-0.1.0a0/src/penicillium/dcs/session.py +583 -0
- penicillium-0.1.0a0/src/penicillium/dcs/static/assets/index-DuCJmz0W.js +10 -0
- penicillium-0.1.0a0/src/penicillium/dcs/static/assets/index-DypehqXk.css +1 -0
- penicillium-0.1.0a0/src/penicillium/dcs/static/index.html +13 -0
- penicillium-0.1.0a0/src/penicillium/envs.py +327 -0
- penicillium-0.1.0a0/src/penicillium/jax_model.py +176 -0
- penicillium-0.1.0a0/src/penicillium/model.py +403 -0
- penicillium-0.1.0a0/src/penicillium/noise.py +203 -0
- penicillium-0.1.0a0/src/penicillium/raman.py +111 -0
- penicillium-0.1.0a0/src/penicillium/recipe.py +119 -0
- penicillium-0.1.0a0/src/penicillium/result.py +119 -0
- penicillium-0.1.0a0/src/penicillium/rng.py +18 -0
- penicillium-0.1.0a0/src/penicillium/scenarios.py +291 -0
- penicillium-0.1.0a0/src/penicillium/simulator.py +756 -0
- penicillium-0.1.0a0/src/penicillium/solvers.py +89 -0
- penicillium-0.1.0a0/tests/data/determinism_golden.json +313 -0
- penicillium-0.1.0a0/tests/test_cli.py +32 -0
- penicillium-0.1.0a0/tests/test_dcs.py +210 -0
- penicillium-0.1.0a0/tests/test_determinism.py +57 -0
- penicillium-0.1.0a0/tests/test_envs.py +116 -0
- penicillium-0.1.0a0/tests/test_jax_model.py +91 -0
- penicillium-0.1.0a0/tests/test_model_and_raman.py +86 -0
- penicillium-0.1.0a0/tests/test_noise.py +138 -0
- penicillium-0.1.0a0/tests/test_octave_v201.py +41 -0
- penicillium-0.1.0a0/tests/test_recipe.py +62 -0
- penicillium-0.1.0a0/tests/test_scenarios.py +65 -0
- penicillium-0.1.0a0/tests/test_simulator.py +272 -0
- penicillium-0.1.0a0/validation/compare_100_batches.py +389 -0
- penicillium-0.1.0a0/validation/indpensim_100_batches.py +135 -0
- penicillium-0.1.0a0/validation/solver_sensitivity.py +85 -0
- penicillium-0.1.0a0/validation/train_paa_pls.py +70 -0
- penicillium-0.1.0a0/webapp/src/App.jsx +277 -0
- penicillium-0.1.0a0/webapp/src/components/Displays.jsx +406 -0
- penicillium-0.1.0a0/webapp/src/components/Faceplate.jsx +229 -0
- penicillium-0.1.0a0/webapp/src/components/ProcessGraphic.jsx +421 -0
- penicillium-0.1.0a0/webapp/src/components/Trend.jsx +262 -0
- penicillium-0.1.0a0/webapp/src/format.js +35 -0
- penicillium-0.1.0a0/webapp/src/main.jsx +10 -0
- penicillium-0.1.0a0/webapp/src/styles.css +1022 -0
- penicillium-0.1.0a0/webapp/src/useDcs.js +127 -0
|
@@ -0,0 +1,23 @@
|
|
|
1
|
+
.venv/
|
|
2
|
+
__pycache__/
|
|
3
|
+
*.egg-info/
|
|
4
|
+
.pytest_cache/
|
|
5
|
+
.ruff_cache/
|
|
6
|
+
.coverage
|
|
7
|
+
dist/
|
|
8
|
+
build/
|
|
9
|
+
notebooks/output/
|
|
10
|
+
.ipynb_checkpoints/
|
|
11
|
+
|
|
12
|
+
*.csv
|
|
13
|
+
*.zip
|
|
14
|
+
# Local-only material (IndPenSim MATLAB sources, not redistributed; private notes)
|
|
15
|
+
external/
|
|
16
|
+
private/
|
|
17
|
+
PLAN.md
|
|
18
|
+
IndPenSim_V2.01.zip
|
|
19
|
+
# Publisher PDFs of the IndPenSim papers: link them, do not redistribute
|
|
20
|
+
docs/*.pdf
|
|
21
|
+
docs/feedback/
|
|
22
|
+
site/
|
|
23
|
+
webapp/node_modules/
|
|
@@ -0,0 +1,63 @@
|
|
|
1
|
+
# Changelog
|
|
2
|
+
|
|
3
|
+
All notable changes are listed here. Versions follow [Semantic Versioning](https://semver.org/); a minor version
|
|
4
|
+
may change numerical output only with an entry here and an updated validation report.
|
|
5
|
+
|
|
6
|
+
## 0.1.0a0 (2026-10-30)
|
|
7
|
+
|
|
8
|
+
First public release (alpha).
|
|
9
|
+
|
|
10
|
+
- The project is called **Penicillium**: `pip install penicillium`, `import penicillium`, command `penicillium`
|
|
11
|
+
(working name during development: `pypensim-next`).
|
|
12
|
+
|
|
13
|
+
### Simulator
|
|
14
|
+
- Pure-Python IndPenSim simulator (NumPy, SciPy, pandas), optional Numba backend (`[fast]`), 55 channels, Raman
|
|
15
|
+
spectra, faults 0-8, disturbances, PRBS feeds, PAT control of PAA, offline lab measurements, batches up to 350 h.
|
|
16
|
+
- Follows MATLAB IndPenSim V2.01.
|
|
17
|
+
- `SimulationConfig(quirks="matlab" | "fixed")`: IndPenSim's controller code as written, or an alternative reading of two of its details.
|
|
18
|
+
- `SimulationConfig(inhibition=0 | 1 | 2)`: IndPenSim's three inhibition models.
|
|
19
|
+
- `BatchParameters`: the random inputs of a batch, replayable; `BatchParameters.from_matlab()` replays a batch
|
|
20
|
+
captured from MATLAB IndPenSim.
|
|
21
|
+
- Seeding: every random quantity has its own `SeedSequence` stream; the random inputs of a batch are bit-identical
|
|
22
|
+
on every platform (trajectories agree to ~1e-4) and independent of NumPy's global generator. The eight disturbances are independent, as in IndPenSim.
|
|
23
|
+
- `PAA_pred` channel: PAA predicted from the Raman spectrum with a PLS model fitted by this project.
|
|
24
|
+
- `metadata["growth_rate_resets"]`: samples where IndPenSim's growth-rate reset fired.
|
|
25
|
+
- `MeasurementNoise` and `Sensor`: sensor and assay errors (white noise, relative error, calibration offset,
|
|
26
|
+
drift, quantisation, bounds) added to a copy of a result; exact values kept in `BatchResult.truth`. `OUR`/`CER`
|
|
27
|
+
are recomputed from the measured off-gas. Preset `MeasurementNoise.typical()` (measurements only; recipe set
|
|
28
|
+
points stay exact); `measurement_noise=` on the
|
|
29
|
+
campaign builders; `penicillium generate --noise typical`. The simulator stays noise-free.
|
|
30
|
+
|
|
31
|
+
### Ecosystem
|
|
32
|
+
- `penicillium.envs`: Gymnasium environments `IndPenSim-v0`, `IndPenSimFault-v0`, `IndPenSimRaman-v0`
|
|
33
|
+
(pass `check_env`, seeded); `collect_minari_dataset` for offline RL.
|
|
34
|
+
- `penicillium.scenarios`: one-call fault, soft-sensor and Goldrick-2019 campaigns in tidy form, batch-wise
|
|
35
|
+
splits, and a control benchmark.
|
|
36
|
+
- `simulate_batches` for parallel campaigns; `penicillium generate` command-line dataset generator.
|
|
37
|
+
|
|
38
|
+
### Operator console
|
|
39
|
+
- `penicillium dcs` (`[dcs]` extra): one batch in real time in a DCS-style web app (FastAPI + React):
|
|
40
|
+
process graphic with PV and SP, faceplates with trends, CAS/AUTO modes and operator setpoints, priority alarms
|
|
41
|
+
with acknowledgement, event journal, offline lab results, Raman spectrum, hold/resume/abort, ×1 to ×36000 speed.
|
|
42
|
+
`penicillium.dcs.BatchSession` is the same logic without the web layer.
|
|
43
|
+
|
|
44
|
+
### Differentiable model
|
|
45
|
+
- `penicillium.jax_model` (`[jax]` extra): the model right-hand side as a jit-compiled JAX function generated from
|
|
46
|
+
the same source as the NumPy/Numba versions, with exact state and parameter Jacobians; gradients through stiff
|
|
47
|
+
integration (e.g. diffrax) match finite differences.
|
|
48
|
+
|
|
49
|
+
### Validation
|
|
50
|
+
- Line-by-line comparison with IndPenSim V2.01 (`docs/quirks.md`).
|
|
51
|
+
- Statistical comparison with the published 100-batch IndPenSim data set (`docs/validation.md`), with mean
|
|
52
|
+
differences, Welch intervals and KS power per strategy, a coverage appendix for every channel, and the method
|
|
53
|
+
and evidence behind the fault types inferred for published batches 91-100.
|
|
54
|
+
- Cross-platform determinism tests.
|
|
55
|
+
|
|
56
|
+
### Documentation
|
|
57
|
+
- Changes after a review of the pre-release manual: "Limitations at a glance" on
|
|
58
|
+
the first page; "no significant difference detected" with KS p-values and test power instead of
|
|
59
|
+
"indistinguishable"; fault labels (`fault_active`) defined for process and sensor faults; the limits of the
|
|
60
|
+
worked examples stated (batch-level counting and Clopper-Pearson intervals in tutorial 5, sample-wise
|
|
61
|
+
cross-validation in tutorial 6, baselines in tutorial 7, Student's t interval in tutorial 2); reward shaping
|
|
62
|
+
equivalence of the Gymnasium reward; replay of the 15 Octave batches in Penicillium and MATLAB; seeding overlap
|
|
63
|
+
warnings; new pages on reporting a benchmark and related simulators.
|
|
@@ -0,0 +1,86 @@
|
|
|
1
|
+
cff-version: 1.2.0
|
|
2
|
+
message: >-
|
|
3
|
+
If you use Penicillium, please cite the software and the two IndPenSim papers it implements
|
|
4
|
+
(Goldrick et al. 2015 and 2019, listed under references). Penicillium is MIT-licensed; the parts
|
|
5
|
+
derived from IndPenSim are also under IndPenSim's BSD 3-Clause License (Copyright (c) 2016, Stephen Goldrick).
|
|
6
|
+
title: "Penicillium: a validated, pip-installable Python implementation of the IndPenSim penicillin fermentation simulator"
|
|
7
|
+
type: software
|
|
8
|
+
authors:
|
|
9
|
+
- family-names: Sartori
|
|
10
|
+
given-names: Francesco
|
|
11
|
+
orcid: "https://orcid.org/0000-0002-7905-5690"
|
|
12
|
+
version: 0.1.0a0
|
|
13
|
+
date-released: 2026-10-30
|
|
14
|
+
repository-code: "https://github.com/francesco-sartori/Penicillium"
|
|
15
|
+
url: "https://penicillium.readthedocs.io"
|
|
16
|
+
license:
|
|
17
|
+
- MIT
|
|
18
|
+
- BSD-3-Clause
|
|
19
|
+
keywords:
|
|
20
|
+
- IndPenSim
|
|
21
|
+
- penicillin
|
|
22
|
+
- fed-batch fermentation
|
|
23
|
+
- bioprocess simulation
|
|
24
|
+
- process control
|
|
25
|
+
- fault detection
|
|
26
|
+
- soft sensors
|
|
27
|
+
- reinforcement learning
|
|
28
|
+
references:
|
|
29
|
+
- type: article
|
|
30
|
+
title: The development of an industrial-scale fed-batch fermentation simulation
|
|
31
|
+
authors:
|
|
32
|
+
- family-names: Goldrick
|
|
33
|
+
given-names: Stephen
|
|
34
|
+
- family-names: Ştefan
|
|
35
|
+
given-names: Andrei
|
|
36
|
+
- family-names: Lovett
|
|
37
|
+
given-names: David
|
|
38
|
+
- family-names: Montague
|
|
39
|
+
given-names: Gary
|
|
40
|
+
- family-names: Lennox
|
|
41
|
+
given-names: Barry
|
|
42
|
+
journal: Journal of Biotechnology
|
|
43
|
+
volume: 193
|
|
44
|
+
start: 70
|
|
45
|
+
end: 82
|
|
46
|
+
year: 2015
|
|
47
|
+
doi: 10.1016/j.jbiotec.2014.10.029
|
|
48
|
+
- type: article
|
|
49
|
+
title: Modern day monitoring and control challenges outlined on an industrial-scale benchmark fermentation process
|
|
50
|
+
authors:
|
|
51
|
+
- family-names: Goldrick
|
|
52
|
+
given-names: Stephen
|
|
53
|
+
- family-names: Duran-Villalobos
|
|
54
|
+
given-names: Carlos A.
|
|
55
|
+
- family-names: Jankauskas
|
|
56
|
+
given-names: Karolis
|
|
57
|
+
- family-names: Lovett
|
|
58
|
+
given-names: David
|
|
59
|
+
- family-names: Farid
|
|
60
|
+
given-names: Suzanne S.
|
|
61
|
+
- family-names: Lennox
|
|
62
|
+
given-names: Barry
|
|
63
|
+
journal: Computers and Chemical Engineering
|
|
64
|
+
volume: 130
|
|
65
|
+
start: 106471
|
|
66
|
+
year: 2019
|
|
67
|
+
doi: 10.1016/j.compchemeng.2019.05.037
|
|
68
|
+
- type: data
|
|
69
|
+
title: "Data for: Modern day monitoring and control challenges outlined on an industrial-scale benchmark fermentation process"
|
|
70
|
+
authors:
|
|
71
|
+
- family-names: Goldrick
|
|
72
|
+
given-names: Stephen
|
|
73
|
+
publisher:
|
|
74
|
+
name: Mendeley Data
|
|
75
|
+
year: 2019
|
|
76
|
+
doi: 10.17632/pdnjz7zz5x.1
|
|
77
|
+
license: CC-BY-4.0
|
|
78
|
+
- type: software
|
|
79
|
+
title: "IndPenSim: Industrial-scale Penicillin Simulation"
|
|
80
|
+
authors:
|
|
81
|
+
- family-names: Goldrick
|
|
82
|
+
given-names: Stephen
|
|
83
|
+
version: "2.01"
|
|
84
|
+
date-released: 2019-07-09
|
|
85
|
+
license: BSD-3-Clause
|
|
86
|
+
url: https://www.mathworks.com/matlabcentral/fileexchange/49041-industrial-scale-penicillin-simulationv2
|
|
@@ -0,0 +1,58 @@
|
|
|
1
|
+
Penicillium is distributed under the MIT License (part 1). The process model, controllers, recipe tables,
|
|
2
|
+
fault profiles and reference Raman spectrum derive from IndPenSim, which is distributed under the BSD 3-Clause
|
|
3
|
+
License (part 2). Both notices apply to the parts derived from IndPenSim. See NOTICE for details.
|
|
4
|
+
|
|
5
|
+
|
|
6
|
+
1. MIT License
|
|
7
|
+
|
|
8
|
+
Copyright (c) 2026 Francesco Sartori
|
|
9
|
+
|
|
10
|
+
Portions derived from PenSimPy 1.2.5:
|
|
11
|
+
Copyright (c) 2020 Quartic.ai Canada, Inc.
|
|
12
|
+
|
|
13
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
14
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
15
|
+
in the Software without restriction, including without limitation the rights
|
|
16
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
17
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
18
|
+
furnished to do so, subject to the following conditions:
|
|
19
|
+
|
|
20
|
+
The above copyright notice and this permission notice shall be included in all
|
|
21
|
+
copies or substantial portions of the Software.
|
|
22
|
+
|
|
23
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
24
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
25
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
26
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
27
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
28
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
29
|
+
SOFTWARE.
|
|
30
|
+
|
|
31
|
+
|
|
32
|
+
2. IndPenSim (BSD 3-Clause License)
|
|
33
|
+
|
|
34
|
+
Copyright (c) 2016, Stephen Goldrick
|
|
35
|
+
All rights reserved.
|
|
36
|
+
|
|
37
|
+
Redistribution and use in source and binary forms, with or without
|
|
38
|
+
modification, are permitted provided that the following conditions are met:
|
|
39
|
+
|
|
40
|
+
* Redistributions of source code must retain the above copyright notice, this
|
|
41
|
+
list of conditions and the following disclaimer.
|
|
42
|
+
|
|
43
|
+
* Redistributions in binary form must reproduce the above copyright notice,
|
|
44
|
+
this list of conditions and the following disclaimer in the documentation
|
|
45
|
+
and/or other materials provided with the distribution
|
|
46
|
+
* Neither the name of University College London nor the names of its
|
|
47
|
+
contributors may be used to endorse or promote products derived from this
|
|
48
|
+
software without specific prior written permission.
|
|
49
|
+
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
|
50
|
+
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
51
|
+
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
|
52
|
+
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
|
|
53
|
+
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
54
|
+
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
|
|
55
|
+
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
|
|
56
|
+
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
|
|
57
|
+
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
58
|
+
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
Penicillium
|
|
2
|
+
Copyright (c) 2026 Francesco Sartori. Licensed under the MIT License; the parts derived from IndPenSim are
|
|
3
|
+
also subject to IndPenSim's BSD 3-Clause License. Both licence texts are in LICENSE.
|
|
4
|
+
|
|
5
|
+
This product includes or derives from the following works.
|
|
6
|
+
|
|
7
|
+
1. IndPenSim (Industrial-scale Penicillin Simulation), MATLAB, version 2.01
|
|
8
|
+
Copyright (c) 2016, Stephen Goldrick. All rights reserved.
|
|
9
|
+
Distributed under the BSD 3-Clause License, reproduced in full in LICENSE (part 2).
|
|
10
|
+
https://www.mathworks.com/matlabcentral/fileexchange/49041-industrial-scale-penicillin-simulationv2
|
|
11
|
+
http://www.industrialpenicillinsimulation.com/
|
|
12
|
+
The process model, controllers, recipe tables, fault profiles and the reference Raman spectrum of this
|
|
13
|
+
package are derived from IndPenSim. Penicillium is not endorsed by Stephen Goldrick, by University
|
|
14
|
+
College London or by any other contributor to IndPenSim. IndPenSim is described in:
|
|
15
|
+
S. Goldrick, A. Stefan, D. Lovett, G. Montague, B. Lennox. The development of an industrial-scale
|
|
16
|
+
fed-batch fermentation simulation. Journal of Biotechnology 193 (2015) 70-82.
|
|
17
|
+
doi:10.1016/j.jbiotec.2014.10.029
|
|
18
|
+
S. Goldrick, C. A. Duran-Villalobos, K. Jankauskas, D. Lovett, S. S. Farid, B. Lennox. Modern day
|
|
19
|
+
monitoring and control challenges outlined on an industrial-scale benchmark fermentation process.
|
|
20
|
+
Computers and Chemical Engineering 130 (2019) 106471. doi:10.1016/j.compchemeng.2019.05.037
|
|
21
|
+
No MATLAB source file of IndPenSim is distributed with this package.
|
|
22
|
+
|
|
23
|
+
2. PenSimPy 1.2.5, Copyright (c) 2020 Quartic.ai Canada, Inc., MIT License.
|
|
24
|
+
src/penicillium/model.py, simulator.py and raman.py contain portions derived from PenSimPy. Its MIT
|
|
25
|
+
permission notice is reproduced in LICENSE (part 1).
|
|
26
|
+
|
|
27
|
+
3. Data. The PLS coefficients in src/penicillium/_paa_pls.py were fitted by this project on spectra it
|
|
28
|
+
simulated (validation/train_paa_pls.py); IndPenSim's PAA_PLS_model.mat is not used or distributed.
|
|
29
|
+
Comparisons with the published IndPenSim data set (Goldrick, 2019, Mendeley Data, doi:10.17632/pdnjz7zz5x.1,
|
|
30
|
+
CC BY 4.0) download it at run time; it is not redistributed.
|
|
@@ -0,0 +1,156 @@
|
|
|
1
|
+
Metadata-Version: 2.5
|
|
2
|
+
Name: penicillium
|
|
3
|
+
Version: 0.1.0a0
|
|
4
|
+
Summary: Penicillium: a Python implementation of the IndPenSim penicillin fermentation simulator
|
|
5
|
+
Project-URL: Documentation, https://penicillium.readthedocs.io
|
|
6
|
+
Project-URL: Source, https://github.com/francesco-sartori/Penicillium
|
|
7
|
+
Project-URL: Changelog, https://github.com/francesco-sartori/Penicillium/blob/main/CHANGELOG.md
|
|
8
|
+
Project-URL: Issues, https://github.com/francesco-sartori/Penicillium/issues
|
|
9
|
+
Author: Francesco Sartori
|
|
10
|
+
License-Expression: MIT AND BSD-3-Clause
|
|
11
|
+
License-File: LICENSE
|
|
12
|
+
License-File: NOTICE
|
|
13
|
+
Keywords: IndPenSim,Python implementation of IndPenSim,benchmark,bioprocess,fermentation,penicillin,penicillin fermentation simulator,process-control,simulation
|
|
14
|
+
Classifier: Development Status :: 3 - Alpha
|
|
15
|
+
Classifier: Intended Audience :: Science/Research
|
|
16
|
+
Classifier: Operating System :: OS Independent
|
|
17
|
+
Classifier: Programming Language :: Python :: 3
|
|
18
|
+
Classifier: Programming Language :: Python :: 3 :: Only
|
|
19
|
+
Classifier: Programming Language :: Python :: 3.10
|
|
20
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
21
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
22
|
+
Classifier: Programming Language :: Python :: 3.13
|
|
23
|
+
Classifier: Programming Language :: Python :: 3.14
|
|
24
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
25
|
+
Classifier: Topic :: Scientific/Engineering :: Chemistry
|
|
26
|
+
Requires-Python: >=3.10
|
|
27
|
+
Requires-Dist: numpy>=1.25
|
|
28
|
+
Requires-Dist: pandas>=2.0
|
|
29
|
+
Requires-Dist: scipy>=1.12
|
|
30
|
+
Provides-Extra: all
|
|
31
|
+
Requires-Dist: gymnasium>=1.0; extra == 'all'
|
|
32
|
+
Requires-Dist: numba>=0.59; extra == 'all'
|
|
33
|
+
Requires-Dist: pyarrow; extra == 'all'
|
|
34
|
+
Provides-Extra: dcs
|
|
35
|
+
Requires-Dist: fastapi>=0.110; extra == 'dcs'
|
|
36
|
+
Requires-Dist: numba>=0.59; extra == 'dcs'
|
|
37
|
+
Requires-Dist: uvicorn[standard]>=0.29; extra == 'dcs'
|
|
38
|
+
Provides-Extra: dev
|
|
39
|
+
Requires-Dist: diffrax; extra == 'dev'
|
|
40
|
+
Requires-Dist: fastapi>=0.110; extra == 'dev'
|
|
41
|
+
Requires-Dist: gymnasium>=1.0; extra == 'dev'
|
|
42
|
+
Requires-Dist: httpx; extra == 'dev'
|
|
43
|
+
Requires-Dist: jax>=0.4.30; extra == 'dev'
|
|
44
|
+
Requires-Dist: numba>=0.59; extra == 'dev'
|
|
45
|
+
Requires-Dist: pyarrow; extra == 'dev'
|
|
46
|
+
Requires-Dist: pytest-cov; extra == 'dev'
|
|
47
|
+
Requires-Dist: pytest>=8; extra == 'dev'
|
|
48
|
+
Requires-Dist: ruff>=0.6; extra == 'dev'
|
|
49
|
+
Requires-Dist: uvicorn[standard]>=0.29; extra == 'dev'
|
|
50
|
+
Provides-Extra: docs
|
|
51
|
+
Requires-Dist: gymnasium>=1.0; extra == 'docs'
|
|
52
|
+
Requires-Dist: mkdocs-jupyter<0.26,>=0.25; extra == 'docs'
|
|
53
|
+
Requires-Dist: mkdocs-material<10,>=9.5; extra == 'docs'
|
|
54
|
+
Requires-Dist: mkdocstrings[python]>=0.25; extra == 'docs'
|
|
55
|
+
Provides-Extra: fast
|
|
56
|
+
Requires-Dist: numba>=0.59; extra == 'fast'
|
|
57
|
+
Provides-Extra: gym
|
|
58
|
+
Requires-Dist: gymnasium>=1.0; extra == 'gym'
|
|
59
|
+
Provides-Extra: jax
|
|
60
|
+
Requires-Dist: jax>=0.4.30; extra == 'jax'
|
|
61
|
+
Provides-Extra: minari
|
|
62
|
+
Requires-Dist: gymnasium>=1.0; extra == 'minari'
|
|
63
|
+
Requires-Dist: minari[create,hdf5]>=0.5; extra == 'minari'
|
|
64
|
+
Requires-Dist: pillow; extra == 'minari'
|
|
65
|
+
Provides-Extra: notebooks
|
|
66
|
+
Requires-Dist: gymnasium>=1.0; extra == 'notebooks'
|
|
67
|
+
Requires-Dist: ipykernel; extra == 'notebooks'
|
|
68
|
+
Requires-Dist: jinja2; extra == 'notebooks'
|
|
69
|
+
Requires-Dist: matplotlib>=3.7; extra == 'notebooks'
|
|
70
|
+
Requires-Dist: minari[create,hdf5]>=0.5; extra == 'notebooks'
|
|
71
|
+
Requires-Dist: numba>=0.59; extra == 'notebooks'
|
|
72
|
+
Requires-Dist: pillow; extra == 'notebooks'
|
|
73
|
+
Requires-Dist: pyarrow; extra == 'notebooks'
|
|
74
|
+
Requires-Dist: scikit-learn>=1.3; extra == 'notebooks'
|
|
75
|
+
Provides-Extra: parquet
|
|
76
|
+
Requires-Dist: pyarrow; extra == 'parquet'
|
|
77
|
+
Provides-Extra: test
|
|
78
|
+
Requires-Dist: pyarrow; extra == 'test'
|
|
79
|
+
Requires-Dist: pytest-cov; extra == 'test'
|
|
80
|
+
Requires-Dist: pytest>=8; extra == 'test'
|
|
81
|
+
Description-Content-Type: text/markdown
|
|
82
|
+
|
|
83
|
+
# Penicillium: a Python implementation of the IndPenSim penicillin fermentation simulator
|
|
84
|
+
|
|
85
|
+
**IndPenSim for Python: validated, pip-installable, fast.** Penicillium is a maintained implementation of
|
|
86
|
+
IndPenSim, the industrial-scale (100 m³) fed-batch penicillin fermentation simulator of Goldrick et al. (2015,
|
|
87
|
+
2019), the benchmark process for bioprocess monitoring, soft sensors, fault detection and control.
|
|
88
|
+
|
|
89
|
+
- **Installs anywhere.** Python 3.10 or later; tested on Linux, macOS and Windows with Python 3.10 to 3.14; no
|
|
90
|
+
compiler, no MATLAB.
|
|
91
|
+
- **Follows IndPenSim.** Checked line by line against IndPenSim V2.01; 14 of 15 batches of V2.01 (run in GNU
|
|
92
|
+
Octave) replayed within 0.1 % on yield. Against the published 100-batch data set, no significant difference
|
|
93
|
+
detected in the yield distributions of recipe-driven, operator-controlled and faulty batches (KS p = 0.31-0.93;
|
|
94
|
+
the tests have low power) ([validation](https://penicillium.readthedocs.io/en/latest/validation/)).
|
|
95
|
+
- **Fast.** 0.6 s per 230 h batch; about 380 batches per minute in parallel on an Intel Core i7-7700HQ
|
|
96
|
+
([benchmarks](https://penicillium.readthedocs.io/en/latest/benchmarks/)).
|
|
97
|
+
- **Ready for ML and control.** One-call campaigns with labels and batch-wise splits, Gymnasium environments, Minari
|
|
98
|
+
datasets and a differentiable JAX model.
|
|
99
|
+
- **Reproducible.** Random inputs bit-identical and trajectories within about 1e-4 on Linux, macOS and Windows.
|
|
100
|
+
- **Honest about its limits.** Fixed fault windows, true values in the sensor-fault channels and the other known
|
|
101
|
+
limitations are listed on the first page of the documentation ([limitations](https://penicillium.readthedocs.io/en/latest/#limitations-at-a-glance)).
|
|
102
|
+
- **Open licence.** MIT, with the parts derived from IndPenSim also under IndPenSim's BSD 3-Clause licence;
|
|
103
|
+
documented provenance ([licensing](https://penicillium.readthedocs.io/en/latest/licensing/)).
|
|
104
|
+
|
|
105
|
+
> Status: alpha (0.1.0a0). The API may still change.
|
|
106
|
+
|
|
107
|
+
**Documentation: <https://penicillium.readthedocs.io>**
|
|
108
|
+
|
|
109
|
+
## Install
|
|
110
|
+
|
|
111
|
+
```bash
|
|
112
|
+
pip install "penicillium[fast]" # with the Numba backend (recommended)
|
|
113
|
+
```
|
|
114
|
+
|
|
115
|
+
Other extras: `gym`, `minari`, `jax`, `dcs`, `parquet`, `notebooks`, `docs`, `dev`
|
|
116
|
+
([installation](https://penicillium.readthedocs.io/en/latest/installation/)).
|
|
117
|
+
|
|
118
|
+
## Example
|
|
119
|
+
|
|
120
|
+
```python
|
|
121
|
+
from penicillium import Simulator
|
|
122
|
+
|
|
123
|
+
result = Simulator().run(seed=1) # one 230 h batch, 1150 samples of 12 min
|
|
124
|
+
print(result.total_yield) # harvested penicillin, kg
|
|
125
|
+
df = result.to_pandas() # 55 channels, indexed by time in hours
|
|
126
|
+
```
|
|
127
|
+
|
|
128
|
+
## Learn more
|
|
129
|
+
|
|
130
|
+
| | |
|
|
131
|
+
|---|---|
|
|
132
|
+
| [Tutorials](https://penicillium.readthedocs.io/en/latest/tutorials/) | Nine notebooks, one per feature: first batch, recipes and control, faults, lab data and Raman, campaigns and fault detection, soft sensors, reinforcement learning, the JAX model, the command line |
|
|
133
|
+
| [User guide](https://penicillium.readthedocs.io/en/latest/) | The process and the model, every option, variables and units, campaigns, the operator console, FAQ |
|
|
134
|
+
| [Validation](https://penicillium.readthedocs.io/en/latest/validation/) | What is checked against IndPenSim and the published data, and the results; [differences and quirks](https://penicillium.readthedocs.io/en/latest/quirks/) |
|
|
135
|
+
| [Case studies](https://github.com/francesco-sartori/Penicillium/tree/main/notebooks) | Replications of Goldrick et al. (2015, 2019), a comparison with the published 100-batch data, a statistical check of the noise model |
|
|
136
|
+
|
|
137
|
+
All of these are on the documentation site, <https://penicillium.readthedocs.io>.
|
|
138
|
+
|
|
139
|
+
From a clone of the repository, run the tests: `pytest -m "not slow"` (seconds) or `pytest` (adds full batches, ~4 min).
|
|
140
|
+
|
|
141
|
+
## Citation
|
|
142
|
+
|
|
143
|
+
If you use Penicillium, cite IndPenSim and this package ([CITATION.cff](https://github.com/francesco-sartori/Penicillium/blob/main/CITATION.cff); see [citing](https://penicillium.readthedocs.io/en/latest/citing/)):
|
|
144
|
+
|
|
145
|
+
- S. Goldrick, A. Ştefan, D. Lovett, G. Montague, B. Lennox. *The development of an industrial-scale fed-batch
|
|
146
|
+
fermentation simulation.* Journal of Biotechnology 193 (2015) 70-82. doi:10.1016/j.jbiotec.2014.10.029
|
|
147
|
+
- S. Goldrick, C. A. Duran-Villalobos, K. Jankauskas, D. Lovett, S. S. Farid, B. Lennox. *Modern day monitoring and
|
|
148
|
+
control challenges outlined on an industrial-scale benchmark fermentation process.* Computers and Chemical
|
|
149
|
+
Engineering 130 (2019) 106471. doi:10.1016/j.compchemeng.2019.05.037
|
|
150
|
+
|
|
151
|
+
## Licence
|
|
152
|
+
|
|
153
|
+
Penicillium is released under the MIT License. Its process model, controllers, recipes, fault profiles and
|
|
154
|
+
reference Raman spectrum derive from IndPenSim, Copyright (c) 2016, Stephen Goldrick, distributed under the BSD
|
|
155
|
+
3-Clause License; that notice applies to those parts too. It is not endorsed by IndPenSim's
|
|
156
|
+
authors or by University College London. Full texts are in [LICENSE](https://github.com/francesco-sartori/Penicillium/blob/main/LICENSE) and [NOTICE](https://github.com/francesco-sartori/Penicillium/blob/main/NOTICE).
|
|
@@ -0,0 +1,74 @@
|
|
|
1
|
+
# Penicillium: a Python implementation of the IndPenSim penicillin fermentation simulator
|
|
2
|
+
|
|
3
|
+
**IndPenSim for Python: validated, pip-installable, fast.** Penicillium is a maintained implementation of
|
|
4
|
+
IndPenSim, the industrial-scale (100 m³) fed-batch penicillin fermentation simulator of Goldrick et al. (2015,
|
|
5
|
+
2019), the benchmark process for bioprocess monitoring, soft sensors, fault detection and control.
|
|
6
|
+
|
|
7
|
+
- **Installs anywhere.** Python 3.10 or later; tested on Linux, macOS and Windows with Python 3.10 to 3.14; no
|
|
8
|
+
compiler, no MATLAB.
|
|
9
|
+
- **Follows IndPenSim.** Checked line by line against IndPenSim V2.01; 14 of 15 batches of V2.01 (run in GNU
|
|
10
|
+
Octave) replayed within 0.1 % on yield. Against the published 100-batch data set, no significant difference
|
|
11
|
+
detected in the yield distributions of recipe-driven, operator-controlled and faulty batches (KS p = 0.31-0.93;
|
|
12
|
+
the tests have low power) ([validation](https://penicillium.readthedocs.io/en/latest/validation/)).
|
|
13
|
+
- **Fast.** 0.6 s per 230 h batch; about 380 batches per minute in parallel on an Intel Core i7-7700HQ
|
|
14
|
+
([benchmarks](https://penicillium.readthedocs.io/en/latest/benchmarks/)).
|
|
15
|
+
- **Ready for ML and control.** One-call campaigns with labels and batch-wise splits, Gymnasium environments, Minari
|
|
16
|
+
datasets and a differentiable JAX model.
|
|
17
|
+
- **Reproducible.** Random inputs bit-identical and trajectories within about 1e-4 on Linux, macOS and Windows.
|
|
18
|
+
- **Honest about its limits.** Fixed fault windows, true values in the sensor-fault channels and the other known
|
|
19
|
+
limitations are listed on the first page of the documentation ([limitations](https://penicillium.readthedocs.io/en/latest/#limitations-at-a-glance)).
|
|
20
|
+
- **Open licence.** MIT, with the parts derived from IndPenSim also under IndPenSim's BSD 3-Clause licence;
|
|
21
|
+
documented provenance ([licensing](https://penicillium.readthedocs.io/en/latest/licensing/)).
|
|
22
|
+
|
|
23
|
+
> Status: alpha (0.1.0a0). The API may still change.
|
|
24
|
+
|
|
25
|
+
**Documentation: <https://penicillium.readthedocs.io>**
|
|
26
|
+
|
|
27
|
+
## Install
|
|
28
|
+
|
|
29
|
+
```bash
|
|
30
|
+
pip install "penicillium[fast]" # with the Numba backend (recommended)
|
|
31
|
+
```
|
|
32
|
+
|
|
33
|
+
Other extras: `gym`, `minari`, `jax`, `dcs`, `parquet`, `notebooks`, `docs`, `dev`
|
|
34
|
+
([installation](https://penicillium.readthedocs.io/en/latest/installation/)).
|
|
35
|
+
|
|
36
|
+
## Example
|
|
37
|
+
|
|
38
|
+
```python
|
|
39
|
+
from penicillium import Simulator
|
|
40
|
+
|
|
41
|
+
result = Simulator().run(seed=1) # one 230 h batch, 1150 samples of 12 min
|
|
42
|
+
print(result.total_yield) # harvested penicillin, kg
|
|
43
|
+
df = result.to_pandas() # 55 channels, indexed by time in hours
|
|
44
|
+
```
|
|
45
|
+
|
|
46
|
+
## Learn more
|
|
47
|
+
|
|
48
|
+
| | |
|
|
49
|
+
|---|---|
|
|
50
|
+
| [Tutorials](https://penicillium.readthedocs.io/en/latest/tutorials/) | Nine notebooks, one per feature: first batch, recipes and control, faults, lab data and Raman, campaigns and fault detection, soft sensors, reinforcement learning, the JAX model, the command line |
|
|
51
|
+
| [User guide](https://penicillium.readthedocs.io/en/latest/) | The process and the model, every option, variables and units, campaigns, the operator console, FAQ |
|
|
52
|
+
| [Validation](https://penicillium.readthedocs.io/en/latest/validation/) | What is checked against IndPenSim and the published data, and the results; [differences and quirks](https://penicillium.readthedocs.io/en/latest/quirks/) |
|
|
53
|
+
| [Case studies](https://github.com/francesco-sartori/Penicillium/tree/main/notebooks) | Replications of Goldrick et al. (2015, 2019), a comparison with the published 100-batch data, a statistical check of the noise model |
|
|
54
|
+
|
|
55
|
+
All of these are on the documentation site, <https://penicillium.readthedocs.io>.
|
|
56
|
+
|
|
57
|
+
From a clone of the repository, run the tests: `pytest -m "not slow"` (seconds) or `pytest` (adds full batches, ~4 min).
|
|
58
|
+
|
|
59
|
+
## Citation
|
|
60
|
+
|
|
61
|
+
If you use Penicillium, cite IndPenSim and this package ([CITATION.cff](https://github.com/francesco-sartori/Penicillium/blob/main/CITATION.cff); see [citing](https://penicillium.readthedocs.io/en/latest/citing/)):
|
|
62
|
+
|
|
63
|
+
- S. Goldrick, A. Ştefan, D. Lovett, G. Montague, B. Lennox. *The development of an industrial-scale fed-batch
|
|
64
|
+
fermentation simulation.* Journal of Biotechnology 193 (2015) 70-82. doi:10.1016/j.jbiotec.2014.10.029
|
|
65
|
+
- S. Goldrick, C. A. Duran-Villalobos, K. Jankauskas, D. Lovett, S. S. Farid, B. Lennox. *Modern day monitoring and
|
|
66
|
+
control challenges outlined on an industrial-scale benchmark fermentation process.* Computers and Chemical
|
|
67
|
+
Engineering 130 (2019) 106471. doi:10.1016/j.compchemeng.2019.05.037
|
|
68
|
+
|
|
69
|
+
## Licence
|
|
70
|
+
|
|
71
|
+
Penicillium is released under the MIT License. Its process model, controllers, recipes, fault profiles and
|
|
72
|
+
reference Raman spectrum derive from IndPenSim, Copyright (c) 2016, Stephen Goldrick, distributed under the BSD
|
|
73
|
+
3-Clause License; that notice applies to those parts too. It is not endorsed by IndPenSim's
|
|
74
|
+
authors or by University College London. Full texts are in [LICENSE](https://github.com/francesco-sartori/Penicillium/blob/main/LICENSE) and [NOTICE](https://github.com/francesco-sartori/Penicillium/blob/main/NOTICE).
|
|
@@ -0,0 +1,27 @@
|
|
|
1
|
+
# API reference
|
|
2
|
+
|
|
3
|
+
::: penicillium.simulator.Simulator
|
|
4
|
+
|
|
5
|
+
::: penicillium.simulator.SimulationConfig
|
|
6
|
+
|
|
7
|
+
::: penicillium.simulator.BatchParameters
|
|
8
|
+
|
|
9
|
+
::: penicillium.simulator.Fault
|
|
10
|
+
|
|
11
|
+
::: penicillium.simulator.RamanMode
|
|
12
|
+
|
|
13
|
+
::: penicillium.result.BatchResult
|
|
14
|
+
|
|
15
|
+
::: penicillium.recipe.Recipe
|
|
16
|
+
|
|
17
|
+
::: penicillium.recipe.Profile
|
|
18
|
+
|
|
19
|
+
::: penicillium.batches
|
|
20
|
+
|
|
21
|
+
::: penicillium.scenarios
|
|
22
|
+
|
|
23
|
+
::: penicillium.envs.IndPenSimEnv
|
|
24
|
+
|
|
25
|
+
::: penicillium.envs.collect_minari_dataset
|
|
26
|
+
|
|
27
|
+
::: penicillium.jax_model
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
# Background: IndPenSim and PenSimPy (checked 2026-10-17)
|
|
2
|
+
|
|
3
|
+
## IndPenSim (MATLAB)
|
|
4
|
+
|
|
5
|
+
- Reference: IndPenSim V2.01, the release on [MATLAB File Exchange](https://www.mathworks.com/matlabcentral/fileexchange/49041-industrial-scale-penicillin-simulationv2) (version 2.0.1,
|
|
6
|
+
9 July 2019). Default solver `ode15s` with MATLAB's default options
|
|
7
|
+
(`solv = 2` in `indpensim_run.m`).
|
|
8
|
+
- Published 100-batch data set (Goldrick et al. 2019): compared by distribution in
|
|
9
|
+
[validation_100_batches.md](validation_100_batches.md). MATLAB's random streams differ from NumPy's, so
|
|
10
|
+
individual batches cannot be matched.
|
|
11
|
+
|
|
12
|
+
## PenSimPy
|
|
13
|
+
|
|
14
|
+
[PenSimPy](https://pypi.org/project/pensimpy/) (Quartic.ai,
|
|
15
|
+
MIT) was the first Python port of IndPenSim, and Penicillium builds on it: its first version was ported from
|
|
16
|
+
PenSimPy 1.2.5 before being checked line by line against IndPenSim V2.01. PenSimPy's last release dates from June 2021 and its original repository (github.com/Quarticai/PenSimPy) is
|
|
17
|
+
no longer online. It pins NumPy 1.18, pandas 1.0 and SciPy 1.4, and its default solver path uses `fastodeint`, a
|
|
18
|
+
C++ extension built on Boost and OpenMP that is not published on PyPI, so `pip install pensimpy` does not complete
|
|
19
|
+
on current Python versions. It offers a step-by-step interface with a random-agent example, but no Gymnasium
|
|
20
|
+
environment.
|
|
21
|
+
|
|
22
|
+
Penicillium adds:
|
|
23
|
+
|
|
24
|
+
- an open licence (MIT, with IndPenSim's BSD 3-Clause for the parts derived from it) and documented provenance;
|
|
25
|
+
- pure Python on 3.10–3.14 with no compiler, tested on Linux, macOS and Windows, with an optional Numba backend;
|
|
26
|
+
- Gymnasium environments, Minari datasets and a differentiable JAX model;
|
|
27
|
+
- a browser-based DCS operator interface;
|
|
28
|
+
- all of IndPenSim's options, including its three inhibition models and eight faults, and statistical validation
|
|
29
|
+
against the published IndPenSim data set.
|
|
@@ -0,0 +1,35 @@
|
|
|
1
|
+
# Reporting a benchmark
|
|
2
|
+
|
|
3
|
+
A result obtained on simulated data can be repeated exactly by anyone, but only if the paper says how the data
|
|
4
|
+
were made and how the method was scored. When you publish a fault-detection, soft-sensor or control result
|
|
5
|
+
obtained with Penicillium, report at least the following.
|
|
6
|
+
|
|
7
|
+
| Item | Why | Where to find it |
|
|
8
|
+
|---|---|---|
|
|
9
|
+
| **Package version** | Numerical output may change between minor versions (with a changelog entry and an updated validation report) | `penicillium.__version__`, `campaign.json` |
|
|
10
|
+
| **Campaign seed** and campaign call | The seed fixes every batch of a campaign; the call fixes how many batches of each fault | `scenarios.fault_campaign(n_normal=..., n_per_fault=..., seed=...)` |
|
|
11
|
+
| **Non-default options** | Faults, Raman mode, batch length, measurement noise and so on change the data | `result.config` of any batch; `campaign.json` records only the generator, version and description |
|
|
12
|
+
| **Split seed** and split rule | Which batches went into training, validation and test; always split by batch | `camp.split(test_fraction=..., seed=...)`, or the list of batch ids |
|
|
13
|
+
| **Label used** | `fault_active` means "the cause acts" for process faults but "this batch is faulty" for sensor faults ([fault labels](scenarios.md#fault-labels-in-campaigns)); say which label each metric uses, and how you treated the sensor faults | your code |
|
|
14
|
+
| **Data the method saw** | Online measurements only (`camp.online()`), or also lab results, spectra or hidden states | your code |
|
|
15
|
+
| **Scoring rule** | Batch level or sample level; how an alarm before the fault starts is counted (a false alarm, not a detection); how delays are measured | your code |
|
|
16
|
+
| **Uncertainty** | With few test batches per class, give an interval for every rate (for example Clopper–Pearson for a detection rate) | your code |
|
|
17
|
+
|
|
18
|
+
Some properties of the IndPenSim faults affect what a benchmark measures:
|
|
19
|
+
|
|
20
|
+
- each fault type acts in the same fixed windows in every batch, so a classifier can learn the batch time instead
|
|
21
|
+
of the process, and detection delays reflect the windows ([Faults](faults.md#faults));
|
|
22
|
+
- under the sensor faults, `T` and `pH` hold the true broth values, which makes detection easier than on a plant
|
|
23
|
+
([Faults](faults.md#sensor-faults-the-subtle-ones));
|
|
24
|
+
- a campaign with `seed=0` shares its batches with `Simulator().run(seed=n)` and with the default seeds of the
|
|
25
|
+
command line ([Campaigns](scenarios.md#ready-made-campaigns)).
|
|
26
|
+
|
|
27
|
+
Fixed seed sets, a scoring script and the list of faults that can be detected from `camp.online()` are planned, as
|
|
28
|
+
a versioned benchmark protocol.
|
|
29
|
+
|
|
30
|
+
A suggested sentence for the methods section, next to the one in [Citing](citing.md):
|
|
31
|
+
|
|
32
|
+
> The data were generated with Penicillium x.y.z, `scenarios.fault_campaign(n_normal=N, n_per_fault=M, seed=S)`
|
|
33
|
+
> with default options, and split by batch with `camp.split(test_fraction=0.3, seed=R)`. Detection was scored per
|
|
34
|
+
> batch on `fault`, an alarm before the first fault window counting as a false alarm, and per sample on
|
|
35
|
+
> `fault_active`, with sensor faults scored from 40 h.
|
|
@@ -0,0 +1,27 @@
|
|
|
1
|
+
# Speed
|
|
2
|
+
|
|
3
|
+
Measured with `benchmarks/speed.py` on an Intel Core i7-7700HQ (4 cores / 8 threads) with 16 GB RAM,
|
|
4
|
+
Windows 10, Python 3.14, 2026-10-02. One default 230 h batch (recipe driven, disturbances on); best of three runs.
|
|
5
|
+
|
|
6
|
+
| implementation | Raman spectra | seconds per batch | solver tolerance |
|
|
7
|
+
|---|---|---|---|
|
|
8
|
+
| Penicillium, Numba backend (`fast` extra) | no | **0.58** | rtol 1e-5 |
|
|
9
|
+
| Penicillium, Numba backend | yes | **0.87** | rtol 1e-5 |
|
|
10
|
+
| Penicillium, pure Python | no | 11.0 | rtol 1e-5 |
|
|
11
|
+
| Penicillium, pure Python | yes | 11.4 | rtol 1e-5 |
|
|
12
|
+
|
|
13
|
+
These timings are at Penicillium's default tolerance (LSODA, rtol 1e-5), which is not traded for speed: in the
|
|
14
|
+
[batch-by-batch comparison with IndPenSim V2.01](validation_octave.md), batches run with the defaults stay within
|
|
15
|
+
about 1e-6 (mean over the batch, relative to each variable's peak) and 5e-6 in yield of the same batches solved at
|
|
16
|
+
tight tolerances. IndPenSim V2.01 integrates at `ode15s`'s default RelTol 1e-3. A reference-grade run at
|
|
17
|
+
rtol 1e-10 takes about 2 s per batch.
|
|
18
|
+
|
|
19
|
+
Campaigns run in parallel processes (`simulate_batches`, `scenarios.*`): **200 batches in 32 s** on the same machine
|
|
20
|
+
(380 batches per minute), so a 1,000-batch campaign takes under 3 minutes.
|
|
21
|
+
|
|
22
|
+
Reproduce:
|
|
23
|
+
|
|
24
|
+
```bash
|
|
25
|
+
pip install -e ".[fast]" # from a clone of the repository
|
|
26
|
+
python benchmarks/speed.py
|
|
27
|
+
```
|