[FFroehlich]

Stacked on #556 (merge that first).

assignments_to_experiments rejects an assignment whose value changes with time within a period, because a condition change is applied once at the start of its period, with time set to the period's start: on a constant target it can only step. Fujita_SciSignal2010's EGF input is such an assignment,

EGF := piecewise(EGF_rate * time + EGF_0, time <= EGF_end, 0)

steps in the training data and ramps (EGF_0 = 0, EGF_rate > 0) in its held-out data.

What. A value may now change linearly with time (its rate of change must not itself depend on time; time^2, exp(time) are still rejected). The only change to the model is to make the target a differential entity, as the v2 condition table allows targets to be constant or differential entities: d target/dt = <target>_slope, with a new parameter for the slope, a placeholder at 0. Everything else goes through the condition and experiment tables: every period that starts a piece, including the first at time 0, sets both the target's value at the period start and its slope:

Fujita condition	t	EGF	EGF_slope
pulse	0	EGF_0	0
pulse	EGF_end	0	0
ramp	0	0	EGF_rate
ramp	EGF_end	0	0

The slope parameter stands in for a setRate operation, which earlier v2 drafts had (PEtab-dev/PEtab#581) and which was dropped before release. If it returns, the rate rule becomes d target/dt = 0 and the slopes become setRate rows.

Changes at the start of an experiment may only refer to parameter table symbols, so constant model parameters in them take their model values.

During preequilibration the value is held at its value at time 0 with slope zero (consistent with the existing preequilibration semantics and warning), and the first period after preequilibration sets the slope again. An assignment that only switches between constant values converts exactly as before, as a constant parameter; the rate rule is only introduced where a piece actually ramps.

Checked

• Tests: a ramp ending at a switch, Fujita-style pulses and ramps set by the condition table, a ramp after preequilibration; time^2 and exp(time) values rejected.

• Fujita, PEtab v1 against the v2 upgrade converted by this branch, AMICI at the nominal parameters:

  problem	v1 llh	converted v2 llh (atol 1e-10 / 1e-14)
  training (6 step doses)	53.0829	53.0833 / 53.0838
  test data: 6 pulses	−4028.1483	−4028.1482 / −4028.1483
  test data: 4 ramps	−1085.2180	−1085.2180 / −1085.2180
  test data: all 10	−5060.2833	−5060.2829 / −5060.2825

  (The test-data problems were loaded without their visualization table, which they share with the training problem and which names datasets their measurement tables lack; the lint rejects them otherwise.)

• Brannmark_JBC2010, Isensee_JCB2018 and Raimundez_PCB2020 convert with log-likelihoods identical to assignments_to_experiments: convert assignments that switch more than once #556 (Raimundez with ExperimentsToSbmlConverter: write one flat initial assignment per target #546 merged in, which it needs to import at all).

🤖 Generated with Claude Code

https://claude.ai/code/session_01MNJePorYs9zmPYMam6g4EP

[codecov-commenter]

Codecov Report

❌ Patch coverage is 93.33333% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 78.12%. Comparing base (da74567) to head (422e254).

Files with missing lines	Patch %	Lines
petab/v2/converters.py	93.33%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #557      +/-   ##
==========================================
+ Coverage   78.07%   78.12%   +0.05%     
==========================================
  Files          68       68              
  Lines        7823     7846      +23     
  Branches     1399     1403       +4     
==========================================
+ Hits         6108     6130      +22     
- Misses       1227     1228       +1     
  Partials      488      488              

☔ 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.

[FFroehlich]

see also discussion in #556

[dweindl]

Looks good.

Since that function is getting longer and longer, how about converting that into a class and moving some parts to some private helpers (find available ID, add parameter, add RateRule, ...) and encapsulating the currently free non-reusable helpers like _changes? So far, this is unreleased and wouldn't really break much. (Independent of this PR to keep the diff more readable.)

[dweindl]

rename to i_period? i don't recall what i was 50 lines further up.

[dweindl]

On second thought, turning that into a class is probably not really helpful, because it's just one operation with one result. Slimming down that function would still be good.

[dweindl]

There is petab.v2.models._sbml_utils.add_sbml_parameter which already does some more error handling.