Repository navigation
Releases
v1.0: IEEE paper data snapshot (2026-05)
First archived snapshot of the computational environment that
generated the data tables and figures in the IEEE submission
"Making Algebraic Parameter Estimation Practical for Noisy Data via
Gaussian Process Regression" (Bassik, Demin, Ovchinnikov, 2026).
This release captures the exact Julia + Python + MATLAB toolchain
used for the paper, vendored at paper-era commit SHAs, plus the 2200
per-experiment input/output cells (550 each across four variants:
odepe, odepe_polish, amigo2_0_10, amigo2_0_100).
What is in here
julia/: source for ODEParameterEstimation and four dependencies
(GaussianProcesses.jl, mpfi.jl, rs.jl, RationalUnivariateRepresentation.jl)
vendored at paper-era SHAs, with the small[compat]-removing
patches preserved inpatches/.julia/env/:Project.tomlandManifest.tomlresolving the
paper-era environment under Julia 1.11.5.pipeline/: Python wrapper, Julia/Mustache templates, paper-era
config.json+systems.json(NOISE_TYPE=MULTIPLICATIVE,
ODEPE_POLISH=false, SEARCH_BOUNDS=[0,1], NUM_TESTS=10).data/source_csvs/: October and September 2025 benchmark output
CSVs that were assembled into the paper's
combined_results_filtered.csv.data/per_experiment/: 2200 per-experiment directories. Each ODEPE
cell carriesscript.jl,data.csv,result.csv. Each AMIGO2 cell
carriesscript.m,data.csv,objf_nl2sol.m, plus either
result.csv(parameters parsed from AMIGO2's
Estimated global parameterssection) orno_result_in_paper.txt
(138 cells where AMIGO2 hit the knownmklJacpost-processing
crash; preserved as paper-era failures rather than retroactively
faked).bin/gcc: 3-line GCC wrapper that downgrades
-Wincompatible-pointer-typesfrom error to warning, required for
AMIGO2 MEX compilation on GCC 14+. Paper-era runs used an older
GCC and did not need this wrapper; the flag does not change
numerical output.setup.sh,reproduce_one.sh,reproduce_all.sh,
expand_with_new_system.sh: run scripts.SNAPSHOT.toml: machine-readable manifest of every vendored
component with version/commit/SHA.docs/: RECIPE, PAPER_LINEAGE, ADDING_A_SYSTEM, AMIGO2_SETUP.
Verification
Smoke-tested on the author's workstation:
| variant | cell | tolerance | observed | wall time |
|---|---|---|---|---|
odepe |
harmonic_0_0 |
1e-6 | 3.07e-14 | ~6.5 min |
amigo2_0_10 |
harmonic_0_0 |
1e-2 | 0.000 | ~14 min |
The ODEPE residual gap of ~1e-9 (above bit-exact) on noisy cells
comes from HomotopyContinuation.jl's unseeded RNG; this is the
level-3 reproducibility floor without further changes upstream.
AMIGO2 is a stochastic global optimizer (eSS); per-run variance is
much larger than ODEPE's, and AMIGO2's stdout output is rounded to
4 decimals, which is why the table shows 0.000.
External dependencies (not vendored)
- Julia 1.11.5.
setup.shwill install viajuliaupif missing. - Python 3.11+.
setup.shbuilds auv venvfrom
python/requirements.txt. - MATLAB. Proprietary; tested with R2024b. Required only for
amigo2_*variants. - AMIGO2_R2025. Free download from CSIC; user must install and
setAMIGO2_PATH. Seedocs/AMIGO2_SETUP.md.
How to cite
If you use this archive, please cite both the paper and this
snapshot's Zenodo DOI.
License
GPL-3.0, inherited from ODEParameterEstimation. See LICENSE and
NOTICES.md for full per-component attribution.