fix

Author: dweindl

pypesto/objective/amici/amici_calculator.py

@@ -265,14 +265,28 @@ def __call__(
                 # TODO: to amici -- set sllh even if llh is nan?
                 snllh = np.full(len(x_ids), np.nan)
             else:
-                # llh to nllh, dict to array
-                snllh = -np.array(
-                    [
-                        result["sllh"][x_id]  # if x_id in res["sllh"] else 0.0
-                        for x_id in x_ids
-                        if x_id in x_dct.keys()
-                    ]
-                )
+                try:
+                    # llh to nllh, dict to array
+                    snllh = -np.array(
+                        [
+                            result["sllh"][
+                                x_id
+                            ]  # if x_id in res["sllh"] else 0.0
+                            for x_id in x_ids
+                            if x_id in x_dct.keys()
+                        ]
+                    )
+                except KeyError as e:
+                    # A requested sensitivity is missing.
+                    # Probably the affected parameter is a fixed parameter
+                    # in amici instead of a sensitivity parameter
+                    # (non_estimated_parameters_as_constants=True ?).
+                    # In this case, only max(sensi_orders) == 0 is supported
+                    # unless this parameter is fixed in the pypesto problem.
+                    raise ValueError(
+                        f"Cannot compute gradient, missing entry for "
+                        f"{set(result['sllh']) - set(x_dct.keys())}."
+                    ) from e
         if 2 in sensi_orders:
             if result["s2llh"] is None and np.isnan(result["llh"]):
                 # TODO: to amici -- set s2llh even if llh is nan?

test/petab/test_petab_import.py

@@ -303,13 +303,15 @@ def test_petab_v2_boehm():
     petab_problem = v2.Problem.from_yaml(
         models.get_problem_yaml_path(problem_id)
     )
+    expected_fval_nominal = 138.22199693517703
 
     # create model
     importer = PetabImporter(petab_problem)
     problem = importer.create_problem()
     assert problem.x_names == petab_problem.x_ids
     assert problem.dim == petab_problem.n_estimated == 9
     assert problem.dim_full == len(petab_problem.parameters) == 11
+    # petab-non-estimated parameters are fixed in the pypesto problem
     assert problem.x_fixed_indices == [
         i for i, p in enumerate(petab_problem.parameters) if not p.estimate
     ]
@@ -319,28 +321,46 @@ def test_petab_v2_boehm():
 
     # evaluate objective at nominal parameters
     fval = problem.objective(np.asarray(petab_problem.x_nominal_free))
-    assert np.isclose(fval, 138.22199693517703)
+    assert np.isclose(fval, expected_fval_nominal)
 
     # deepcopy works?
     problem_deepcopy = copy.deepcopy(problem)
     fval = problem_deepcopy.objective(np.asarray(petab_problem.x_nominal_free))
-    assert np.isclose(fval, 138.22199693517703)
+    assert np.isclose(fval, expected_fval_nominal)
 
     # pickling works?
     problem_pickled = pickle.loads(pickle.dumps(problem))  # noqa: S301
     fval = problem_pickled.objective(np.asarray(petab_problem.x_nominal_free))
-    assert np.isclose(fval, 138.22199693517703)
+    assert np.isclose(fval, expected_fval_nominal)
 
     # gradient works?
     fval, grad = problem.objective(
         np.asarray(petab_problem.x_nominal_free), sensi_orders=(0, 1)
     )
-    assert np.isclose(fval, 138.22199693517703)
+    assert np.isclose(fval, expected_fval_nominal)
+    assert len(grad) == petab_problem.n_estimated
+
+    # fixing parameters, ...
+    problem.unfix_parameters(petab_problem.x_fixed_indices)
+    assert problem.dim == len(petab_problem.parameters)
+    with pytest.raises(ValueError, match="Cannot compute gradient"):
+        # cannot compute sensitivities for fixed parameters
+        problem.objective(
+            np.asarray(petab_problem.x_nominal), sensi_orders=(0, 1)
+        )
+    fval = problem.objective(np.asarray(petab_problem.x_nominal))
+    assert np.isclose(fval, expected_fval_nominal)
+    # re-fixing parameters
+    problem.fix_parameters(
+        petab_problem.x_fixed_indices, petab_problem.x_nominal_fixed
+    )
+    assert problem.dim == petab_problem.n_estimated
+    fval, grad = problem.objective(
+        np.asarray(petab_problem.x_nominal_free), sensi_orders=(0, 1)
+    )
+    assert np.isclose(fval, expected_fval_nominal)
     assert len(grad) == petab_problem.n_estimated
 
-    # TODO: fixing parameters, ...
-    # problem.fix_parameters(...)
-    # problem.unfix_parameters(...)
     # TODO mode=res/fun,
     # TODO hess/fim...
 

tox.ini

@@ -53,7 +53,7 @@ extras = test,test_petab,amici,petab,emcee,dynesty,mltools,pymc,jax,fides,roadru
 deps =
     git+https://github.com/Benchmarking-Initiative/Benchmark-Models-PEtab.git@master\#subdirectory=src/python
 #    git+https://github.com/PEtab-dev/petab_test_suite@main
-    git+https://github.com/PEtab-dev/petab_test_suite@739b262e9e5f890d00e3bab227d248bb99ea4b66
+    git+https://github.com/PEtab-dev/petab_test_suite@c12b9dc4e4c5585b1b83a1d6e89fd22447c46d03
 commands_pre =
     python3 -m pip uninstall -y amici
     #python3 -m pip install git+https://github.com/dweindl/amici.git@main\#egg=amici&subdirectory=python/sdist