Issues

Output-grid classification: protocol-block actions with unread options are refused under auto with a wrong message, and print_on_stop=>0 without stop_if is needlessly sent to BNG2.pl

lanl/PyBNF#1000 · opened

on Oct 15, 2026

An edition-2 network-free experiment (method: nf) under bngl_backend = bionetgen fails every simulation, because BNG2.pl's NFsim refuses the sample_times PyBNF writes, and nothing refuses it at load

lanl/PyBNF#999 · opened

on Oct 15, 2026

find_t_length refuses valid BNGL simulate actions for every job type (n_output_steps, no step count), misreads n_steps+sample_times and skips commented lines, to compute a length only am reads

lanl/PyBNF#998 · opened

on Oct 15, 2026

A neg_bin column accepts any normalization and is then scored as non-integer 'counts' through the PMF's continuous extension, which is not a likelihood and moves the fit with the normalizing constant

lanl/PyBNF#997 · opened

on Oct 15, 2026

normalization = scale profiles the scale out, yet every Bayesian sampler accepts it, while noise_profiling and linear_profiling are refused for exactly that reason, so the draws are not posterior draws

lanl/PyBNF#996 · opened

on Oct 15, 2026

Scoring turns every exception into a failed evaluation (#388's catch-all), so a configuration error, a data error or a bug in a postprocess script ends a fit with exit 0 and a random draw as its best fit

lanl/PyBNF#995 · opened

on Oct 15, 2026

On the bngsim SBML backend, a free parameter that feeds a chain of parameter initialAssignments (k2 := 2*k1, k3 := k2 + k1) is refused on the gradient path, although bngsim's own column for it is exact

lanl/PyBNF#993 · opened

on Oct 15, 2026

On the bngsim .net backend, a scored Newton steady-state scan reports every initial-condition sensitivity as exactly zero, so a gradient column routed through a conserved species' initial amount comes back 0

lanl/PyBNF#992 · opened

on Oct 15, 2026

am trajectory outputs: a condition experiment gets no files, normalization crashes, and normalization = scale under neg_bin is not refused at load

lanl/PyBNF#990 · opened

on Oct 15, 2026

A setOption line ending in a stray backslash runs in BNG2.pl, but PyBNF refuses it (edition 2) or breaks its simulations (edition 1), and neither message mentions the backslash

lanl/PyBNF#983 · opened

on Oct 5, 2026

An invisible character beside a setOption line (a non-breaking space, for example) makes BNG2.pl skip the line while PyBNF applies it

lanl/PyBNF#982 · opened

on Oct 5, 2026

A setOption written after generate_network is applied by a PyBNF fit but ignored by BNG2.pl, so PyBNF fits a different model from the one in the file

lanl/PyBNF#981 · opened

on Oct 5, 2026

Add a cross-backend differential harness: evaluate one job at a fixed parameter set on every backend that can run it, and compare the simulated outputs and the objective value

lanl/PyBNF#980 · opened

on Oct 5, 2026

Feature request: semiparametric (monotone-spline) link for qualitative/ordinal data — learn the measurement-model shape instead of fixing probit/logit

lanl/PyBNF#979 · opened

on Oct 4, 2026

job_type = check crashes with KeyError when the noise scale is a free parameter (sigma = fit s)

lanl/PyBNF#978 · opened

on Oct 3, 2026

Gradient fits reject an experiment whose only data file is a .prop: 'no routing was supplied'

lanl/PyBNF#977 · opened

on Oct 3, 2026

Gradient fits of function-heavy models spend almost all their time re-running bngsim's SymPy analyses on every evaluation

lanl/PyBNF#976 · opened

on Oct 3, 2026

1

PEtab v2 cannot carry a negative-binomial noise model or the cumulative-to-incident transform: encode the transform with per-period conditions, and add negative-binomial noise as a versioned PyBNF extension

lanl/PyBNF#974 · opened

on Sep 29, 2026

Add an IBIS sampler (job_type = ibis) that updates the exact posterior as observations arrive, absorbs revised observations, and finishes within wall_time_fit

lanl/PyBNF#973 · opened

on Sep 29, 2026

An edition-2 job cannot set ODE solver tolerances for a BNGL model: the model: line refuses rtol:/atol: for .bngl, and the simulations built from experiment: lines never carry them

lanl/PyBNF#971 · opened

on Sep 26, 2026

Check whether the bngsim SBML backend's sensitivity warm-up (#543) serves a stale chain rule when a condition or fitted value writes a derived parameter (lanl/bngsim#708)

lanl/PyBNF#967 · opened

on Sep 25, 2026

A condition that sets the same target twice (perturbations: L = 2, L = 7) is accepted, and the last value silently wins

lanl/PyBNF#964 · opened

on Sep 25, 2026

Follow-up to the PEtab fixes: three valid jobs are refused, by checks added in PRs #940 and #943 and by numbered BNGL action lines that PyBNF does not recognise

lanl/PyBNF#963 · opened

on Sep 25, 2026

Nine loud failures that block a valid job or give the wrong cause: RuleMonkey experiments, one-time-point data, PEtab import and export cases, error messages that hide their reason, and a crash on deeply nested formulas

lanl/PyBNF#962 · opened

on Sep 25, 2026

A noise formula that uses a fixed model parameter (PEtab noiseFormula `0.5 * v3`) passes every load check, then fails each parameter set as it is scored, so the fit either stops with "Your simulations are failing to run" or finishes with an objective of inf

lanl/PyBNF#961 · opened

on Sep 25, 2026

Navigation Menu

Repository navigation

Issues

Output-grid classification: protocol-block actions with unread options are refused under auto with a wrong message, and print_on_stop=>0 without stop_if is needlessly sent to BNG2.pl

An edition-2 network-free experiment (method: nf) under bngl_backend = bionetgen fails every simulation, because BNG2.pl's NFsim refuses the sample_times PyBNF writes, and nothing refuses it at load

find_t_length refuses valid BNGL simulate actions for every job type (n_output_steps, no step count), misreads n_steps+sample_times and skips commented lines, to compute a length only am reads

A neg_bin column accepts any normalization and is then scored as non-integer 'counts' through the PMF's continuous extension, which is not a likelihood and moves the fit with the normalizing constant

normalization = scale profiles the scale out, yet every Bayesian sampler accepts it, while noise_profiling and linear_profiling are refused for exactly that reason, so the draws are not posterior draws

Scoring turns every exception into a failed evaluation (#388's catch-all), so a configuration error, a data error or a bug in a postprocess script ends a fit with exit 0 and a random draw as its best fit

On the bngsim SBML backend, a free parameter that feeds a chain of parameter initialAssignments (k2 := 2*k1, k3 := k2 + k1) is refused on the gradient path, although bngsim's own column for it is exact

On the bngsim .net backend, a scored Newton steady-state scan reports every initial-condition sensitivity as exactly zero, so a gradient column routed through a conserved species' initial amount comes back 0

am trajectory outputs: a condition experiment gets no files, normalization crashes, and normalization = scale under neg_bin is not refused at load

A setOption line ending in a stray backslash runs in BNG2.pl, but PyBNF refuses it (edition 2) or breaks its simulations (edition 1), and neither message mentions the backslash

An invisible character beside a setOption line (a non-breaking space, for example) makes BNG2.pl skip the line while PyBNF applies it

A setOption written after generate_network is applied by a PyBNF fit but ignored by BNG2.pl, so PyBNF fits a different model from the one in the file

Add a cross-backend differential harness: evaluate one job at a fixed parameter set on every backend that can run it, and compare the simulated outputs and the objective value

Feature request: semiparametric (monotone-spline) link for qualitative/ordinal data — learn the measurement-model shape instead of fixing probit/logit

job_type = check crashes with KeyError when the noise scale is a free parameter (sigma = fit s)

Gradient fits reject an experiment whose only data file is a .prop: 'no routing was supplied'

Gradient fits of function-heavy models spend almost all their time re-running bngsim's SymPy analyses on every evaluation

PEtab v2 cannot carry a negative-binomial noise model or the cumulative-to-incident transform: encode the transform with per-period conditions, and add negative-binomial noise as a versioned PyBNF extension

Add an IBIS sampler (job_type = ibis) that updates the exact posterior as observations arrive, absorbs revised observations, and finishes within wall_time_fit

An edition-2 job cannot set ODE solver tolerances for a BNGL model: the model: line refuses rtol:/atol: for .bngl, and the simulations built from experiment: lines never carry them

Check whether the bngsim SBML backend's sensitivity warm-up (#543) serves a stale chain rule when a condition or fitted value writes a derived parameter (lanl/bngsim#708)

A condition that sets the same target twice (perturbations: L = 2, L = 7) is accepted, and the last value silently wins

Follow-up to the PEtab fixes: three valid jobs are refused, by checks added in PRs #940 and #943 and by numbered BNGL action lines that PyBNF does not recognise

Nine loud failures that block a valid job or give the wrong cause: RuleMonkey experiments, one-time-point data, PEtab import and export cases, error messages that hide their reason, and a crash on deeply nested formulas

A noise formula that uses a fixed model parameter (PEtab noiseFormula `0.5 * v3`) passes every load check, then fails each parameter set as it is scored, so the fit either stops with "Your simulations are failing to run" or finishes with an objective of inf