Add `petab.v2.reparametrize_log` by dweindl · Pull Request #560 · PEtab-dev/libpetab-python

dweindl · 2026-10-19T14:43:46Z

PEtab v2 has no parameterScale. To estimate a parameter <id> on log scale, or to specify a prior for log10(<id>), the v2 spec recommends replacing <id> by a new parameter log10_<id> with <id> = 10^log10_<id>.

This adds petab.v2.reparametrize_log(problem, parameter_ids, ...), which does this for the given parameters on a copy of the problem and returns the copy and the mapping of old to new parameter IDs:

parameter table: log-transformed bounds (dropped for non-estimated parameters) and nominal value; priors from the priors argument, or derived from log-normal, log-laplace and log-uniform priors on <id> (other priors on <id> raise)
SBML model: new parameter log10_<id> and initial assignment <id> = 10^log10_<id> (replacing any existing one; <id> stays constant=true)
observable/noise formulas and condition target values: <id> → 10^log10_<id>
measurement overrides: replaced by the new IDs, numeric overrides are log-transformed, and the placeholder ph is replaced by 10^ph in the formulas

A lower bound of 0 becomes -inf, following PEtab-dev/PEtab#682. If other parameters have priors, such a parameter needs an explicit prior, since the default uniform(lb, ub) would be improper; otherwise, a ValueError is raised.

Base e (log_<id>) is supported as well. The procedure follows the one described in the proposed v1→v2 migration guide (PEtab-dev/PEtab#680).

Split off from #539, which uses this in petab1to2(reparametrize=...) to preserve v1 parameter-scale priors.

🤖 Generated with Claude Code

codecov-commenter · 2026-10-19T14:48:08Z

Codecov Report

❌ Patch coverage is 92.77108% with 12 lines in your changes missing coverage. Please review.
✅ Project coverage is 78.81%. Comparing base (e654711) to head (7fc047b).

Files with missing lines	Patch %	Lines
petab/v2/reparametrize.py	93.42%	5 Missing and 5 partials ⚠️
petab/v2/models/_sbml_utils.py	81.81%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #560      +/-   ##
==========================================
+ Coverage   78.44%   78.81%   +0.36%     
==========================================
  Files          68       69       +1     
  Lines        7876     8037     +161     
  Branches     1406     1445      +39     
==========================================
+ Hits         6178     6334     +156     
- Misses       1214     1215       +1     
- Partials      484      488       +4

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

dilpath

Thanks!

dilpath · 2026-10-20T14:41:59Z

+    problem["scaling"].nominal_value = None
+    problem["offset1"].nominal_value = 0
+    problem["offset2"].lb = -1
+    problem["offset2"].ub = 10
+    with pytest.warns(UserWarning, match="no positive nominal value"):
+        problem, _ = reparametrize_log(
+            problem, ["scaling", "offset1", "offset2"]
+        )


Is this warning for offset1? If so, then maybe it's fine if the spec allows -inf for log10_offset1 nominal value.

dilpath · 2026-10-20T14:45:24Z

+@pytest.mark.parametrize("prior", [C.UNIFORM, C.NORMAL, C.GAMMA])
+def test_reparametrize_log_inexpressible_prior(prior):
+    problem = _problem()
+    problem["scaling"].prior_distribution = prior
+    problem["scaling"].prior_parameters = [1, 2]
+    with pytest.raises(ValueError, match="cannot be expressed"):
+        reparametrize_log(problem, ["scaling"])


Docstring? Is the issue that we can't implement the same "uniform" distribution after log-scaling the parameter?

dilpath · 2026-10-20T15:07:08Z

+        ln_b = math.log(10)
+        # not `math.log(x) / ln_b`, which is inexact for powers of 10
+        log_b = math.log10


ln and log are the same for me -- maybe log_factor for ln_b/log_function for log_b?

dilpath · 2026-10-20T15:08:41Z

+    if problem.extensions.sciml is not None:
+        raise NotImplementedError(
+            "Reparametrization of PEtab SciML problems is not supported."
+        )


NotImplementedError if any kind of extension is required in the problem config?

dilpath · 2026-10-20T15:11:11Z

+            if model.sbml_model.getParameter(p) is None:
+                raise NotImplementedError(
+                    f"Model entity `{p}' is not an SBML parameter."
+                )


Fine as is. Although I guess this error should also occur elsewhere/earlier in the package in general, that a parameter in the parameter table must be a parameter in the model.

dilpath · 2026-10-20T15:15:54Z

+        elif p in priors or par.prior_distribution is None or not par.estimate:
+            # priors of non-estimated parameters are irrelevant
+            prior_distribution, prior_parameters = None, []


re: not par.estimate: is it necessary to drop the pre-existing prior instead of converting it? Seems more like a NotImplementedError

dilpath · 2026-10-20T15:19:39Z

+            )
+
+        if par.estimate:
+            lb = log_b(par.lb) if par.lb > 0 else -math.inf


I guess we could define somewhere, some log and log10 functions like

def log10(x): return math.log10(x) if x != 0 else -math.inf

dilpath · 2026-10-20T15:24:47Z

+        set_initial_assignment(
+            model,
+            parameter_id,
+            b_expr ** sp.Symbol(new_id),


petab_symbol? Probably irrelevant.

dilpath · 2026-10-20T15:26:21Z

+    b_expr: sp.Expr,
+    log_b,
+) -> None:
+    """Log-transform placeholders overridden by reparametrized parameters."""


Describe here any potential issues? e.g. that placeholders need to be overridden by either no reparametrized parameters, or only reparametrized parameters with the same base.

PEtab v2 has no `parameterScale`. To estimate a parameter `p` on log scale (and to specify priors for `log10(p)`), `p` has to be replaced by a new parameter `log10_p` with `p = 10^log10_p`. `reparametrize_log` does this for a given set of parameters on a copy of the problem and returns the copy and the mapping of old to new IDs: * parameter table: log-transformed bounds and nominal values (a lower bound of 0 becomes -inf); priors given explicitly, or derived from `log-normal`, `log-laplace` and `log-uniform` priors * SBML model: new parameter `log10_p` and initial assignment `p = 10^log10_p` * observable/noise formulas and condition table target values: `p` -> `10^log10_p` * measurement table overrides: `p` -> `log10_p`, numeric overrides are log-transformed, and the placeholder `ph` -> `10^ph` in the formulas Base e is supported as well. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

dweindl self-assigned this Oct 19, 2026

dweindl force-pushed the reparametrize-log branch from e2290ab to 408403c Compare October 19, 2026 14:44

dweindl mentioned this pull request Oct 19, 2026

Add petab1to2(reparametrize=...) #539

Draft

dweindl mentioned this pull request Oct 19, 2026

Add Problem.get_unused_id #561

Merged

dweindl force-pushed the reparametrize-log branch 2 times, most recently from 74b7276 to e74ee3b Compare October 19, 2026 15:26

This was referenced Oct 19, 2026

sbml_math_to_sympy fails on powers with a dimensionless numeric base and a symbolic exponent dweindl/sbmlmath#50

Open

Add petab_symbol #563

Merged

dweindl marked this pull request as ready for review October 20, 2026 06:22

dweindl requested a review from a team as a code owner October 20, 2026 06:22

dweindl force-pushed the reparametrize-log branch 4 times, most recently from e1d0f09 to 2c303c3 Compare October 20, 2026 14:56

dilpath approved these changes Oct 20, 2026

View reviewed changes

dweindl force-pushed the reparametrize-log branch from 2c303c3 to 7fc047b Compare October 20, 2026 20:30

Navigation Menu

Repository navigation

Add `petab.v2.reparametrize_log` - #560

Add `petab.v2.reparametrize_log`#560
dweindl wants to merge 1 commit into
PEtab-dev:mainfrom
dweindl:reparametrize-log

dweindl commented Oct 19, 2026 •

edited

Loading

Uh oh!

codecov-commenter commented Oct 19, 2026 •

edited

Loading

Uh oh!

dilpath left a comment

Uh oh!

dilpath Oct 20, 2026

Uh oh!

dilpath Oct 20, 2026

Uh oh!

dilpath Oct 20, 2026

Uh oh!

dilpath Oct 20, 2026

Uh oh!

dilpath Oct 20, 2026

Uh oh!

dilpath Oct 20, 2026

Uh oh!

dilpath Oct 20, 2026

Uh oh!

dilpath Oct 20, 2026

Uh oh!

dilpath Oct 20, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

dweindl commented Oct 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov-commenter commented Oct 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

dilpath left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

dweindl commented Oct 19, 2026 •

edited

Loading

codecov-commenter commented Oct 19, 2026 •

edited

Loading