Repository navigation
petab_files_1to2 keeps a stale model file when converting into a non-empty output directory #566
Description
petab_files_1to2 (and petab1to2 with output_dir) doesn't overwrite an existing model file in the output directory. The other tables are overwritten, so the result mixes the new tables with the old model, without any warning.
The cause is _copy_file in petab/v2/petab1to2.py (since #328):
try:
if dest.samefile(src):
return
except FileNotFoundError:
shutil.copy(str(src), str(dest))If dest exists and is a different file, samefile returns False and nothing is copied.
Reproducer (two v1 problems that differ only in the model's value of k1, converted into the same directory):
import tempfile
from pathlib import Path
import pandas as pd
from petab import v1
from petab.v1.models.sbml_model import SbmlModel
from petab.v2 import Problem
from petab.v2.petab1to2 import petab1to2
def write_v1(path: Path, k1: float) -> Path:
return v1.Problem(
model=SbmlModel.from_antimony(f"A = 1; B = 0; k1 = {k1}; A -> B; k1 * A"),
condition_df=v1.get_condition_df(pd.DataFrame({"conditionId": ["c1"]})),
observable_df=v1.get_observable_df(
pd.DataFrame(
{"observableId": ["obs_B"], "observableFormula": ["B"], "noiseFormula": [0.1]}
)
),
measurement_df=v1.get_measurement_df(
pd.DataFrame(
{"observableId": ["obs_B"], "simulationConditionId": ["c1"], "time": [1.0], "measurement": [0.5]}
)
),
parameter_df=v1.get_parameter_df(
pd.DataFrame(
{"parameterId": ["k1"], "parameterScale": ["lin"], "lowerBound": [1e-3],
"upperBound": [1e3], "nominalValue": [0.5], "estimate": [1]}
)
),
).to_files_generic(path)
with tempfile.TemporaryDirectory() as tmp:
tmp = Path(tmp)
out = tmp / "v2"
out.mkdir()
petab1to2(write_v1(tmp / "v1_a", k1=0.5), out)
petab1to2(write_v1(tmp / "v1_b", k1=0.7), out)
problem = Problem.from_yaml(out / "problem.yaml")
print(problem.model.sbml_model.getParameter("k1").getValue())
# 0.5, expected 0.7With reparametrize="priors" or "all" (#539), converting the same problem twice into the same directory reparametrizes twice: the model left from the first run already contains log10_<id>, so the second run adds log10_<id>_1.
The behaviour for a non-empty output directory is inconsistent: if the problem has a non-empty condition table, the second conversion fails, because an existing experiments.tsv raises a ValueError. Without one, it silently succeeds as above.
Possible fixes:
- Require the output directory to be empty or non-existent (and create it), and raise a
ValueErrorotherwise. This makes the behaviour consistent, replaces the special case forexperiments.tsv, and avoids leftover files from earlier runs. It rules out converting in place (output directory = input directory), which thesamefilecheck seems meant to support, but converting in place overwrites the v1 files anyway. - Only fix
_copy_file, i.e., overwritedestunless it's the same file assrc. This is the minimal change, but leaves the inconsistency withexperiments.tsv.
I'd go with 1.
🤖 Generated with Claude Code
Activity