Make StereoCombiner configurable in ApplyModels

docs/changes/2724.bugfix.rst

@@ -0,0 +1,3 @@
+Make ``StereoCombiner`` configurable in ``ApplyModels``.
+Loaded joblib-pickled ``Reconstructor``s are already instantiated, which
+leads to a non-configurable ``StereoCombiner``.

src/ctapipe/tools/apply_models.py

@@ -149,7 +149,25 @@ def setup(self):
 
         self._reconstructors = []
         for path in self.reconstructor_paths:
-            r = Reconstructor.read(path, parent=self, subarray=self.loader.subarray)
+            r = Reconstructor.read(
+                path,
+                parent=self,
+                subarray=self.loader.subarray,
+            )
+
+            # Init new Reconstructor with config parameters and overwrite the StereoCombiner
+            model_keys = ["model_cls", "norm_cls", "sign_cls"]
+            model_kwargs = {
+                key: getattr(r, key) for key in model_keys if hasattr(r, key)
+            }
+            r.stereo_combiner = Reconstructor.from_name(
+                r.__class__.__name__,
+                subarray=self.loader.subarray,
+                parent=self,
+                prefix=r.prefix,
+                **model_kwargs,
+            ).stereo_combiner
+
             if self.n_jobs:
                 r.n_jobs = self.n_jobs
             self._reconstructors.append(r)

src/ctapipe/tools/tests/test_apply_models.py

@@ -1,5 +1,6 @@
 import numpy as np
 import pytest
+from numpy.testing import assert_allclose
 
 from ctapipe.containers import (
     EventIndexContainer,
@@ -31,7 +32,7 @@ def test_apply_energy_regressor(
             f"--input={input_path}",
             f"--output={output_path}",
             f"--reconstructor={energy_regressor_path}",
-            "--StereoMeanCombiner.weights=konrad",
+            "--StereoMeanCombiner.weights=intensity",
             "--chunk-size=5",  # small chunksize so we test multiple chunks for the test file
         ],
         raises=True,
@@ -73,6 +74,24 @@ def test_apply_energy_regressor(
         assert f"{prefix}_tel_is_valid" in tel_events.colnames
         assert "hillas_intensity" in tel_events.colnames
 
+        event_id = 301
+        valid_mask = tel_events[tel_events["event_id"] == event_id][
+            f"{prefix}_tel_is_valid"
+        ]
+        event_energy = np.average(
+            tel_events[tel_events["event_id"] == event_id][f"{prefix}_tel_energy"][
+                valid_mask
+            ],
+            weights=tel_events[tel_events["event_id"] == event_id]["hillas_intensity"][
+                valid_mask
+            ],
+        )
+        assert_allclose(
+            table[table["event_id"] == event_id][f"{prefix}_energy"].value,
+            event_energy,
+            atol=1e-7,
+        )
+
     trigger = read_table(output_path, "/dl1/event/subarray/trigger")
     energy = read_table(output_path, "/dl2/event/subarray/energy/ExtraTreesRegressor")