Don't apply sigma correction to scaled data in cosym (dials#2818)

Changes apply to running dials.symmetry, dials.cosym or dials.correlation_matrix on scaled data.

newsfragments/2818.misc

@@ -0,0 +1 @@
+``dials.correlation_matrix``: Don't apply further sigma correction on scaled data

src/dials/algorithms/correlation/analysis.py

@@ -97,6 +97,11 @@ def __init__(
         if params is None:
             params = phil_scope.extract()
         self.params = params
+
+        # if all datasets have been through scaling, a decision about error models has
+        # been made, so don't apply any further sigma correction
+        apply_sigma_correction = not all(s for s in experiments.scaling_models())
+
         self._reflections = []
         self.ids_to_identifiers_map = ids_to_identifiers_map
 
@@ -147,7 +152,9 @@ def __init__(
         self.params.__inject__("lattice_symmetry_max_delta", 0.0)
         self.params.__inject__("best_monoclinic_beta", True)
 
-        self.cosym_analysis = CosymAnalysis(self.datasets, self.params)
+        self.cosym_analysis = CosymAnalysis(
+            self.datasets, self.params, apply_sigma_correction=apply_sigma_correction
+        )
 
     def _merge_intensities(self, datasets: list) -> list:
         """

src/dials/algorithms/symmetry/__init__.py

@@ -43,6 +43,7 @@ def __init__(
         relative_length_tolerance=None,
         absolute_angle_tolerance=None,
         best_monoclinic_beta=True,
+        apply_sigma_correction=True,
     ):
         """Initialise a symmetry_base object.
 
@@ -69,6 +70,7 @@ def __init__(
           best_monoclinic_beta (bool): If True, then for monoclinic centered cells, I2
             will be preferred over C2 if it gives a less oblique cell (i.e. smaller
             beta angle).
+          apply_sigma_correction (bool): If True, correct SDs by "typical" SD factors.
         """
         self.input_intensities = intensities
 
@@ -127,7 +129,8 @@ def __init__(
         self.dataset_ids = self.dataset_ids.select(sel)
 
         # Correct SDs by "typical" SD factors
-        self._correct_sigmas(sd_fac=2.0, sd_b=0.0, sd_add=0.03)
+        if apply_sigma_correction:
+            self._correct_sigmas(sd_fac=2.0, sd_b=0.0, sd_add=0.03)
         self._normalise(normalisation)
         self._resolution_filter(d_min, min_i_mean_over_sigma_mean, min_cc_half)
 

src/dials/algorithms/symmetry/cosym/__init__.py

@@ -145,7 +145,13 @@ class CosymAnalysis(symmetry_base, Subject):
     the presence of an indexing ambiguity.
     """
 
-    def __init__(self, intensities, params, seed_dataset: Optional[int] = None):
+    def __init__(
+        self,
+        intensities,
+        params,
+        seed_dataset: Optional[int] = None,
+        apply_sigma_correction=True,
+    ):
         """Initialise a CosymAnalysis object.
 
         Args:
@@ -175,6 +181,7 @@ def __init__(self, intensities, params, seed_dataset: Optional[int] = None):
             relative_length_tolerance=None,
             absolute_angle_tolerance=None,
             best_monoclinic_beta=params.best_monoclinic_beta,
+            apply_sigma_correction=apply_sigma_correction,
         )
         Subject.__init__(
             self, events=["optimised", "analysed_symmetry", "analysed_clusters"]

src/dials/algorithms/symmetry/laue_group.py

@@ -40,6 +40,7 @@ def __init__(
         relative_length_tolerance=None,
         absolute_angle_tolerance=None,
         best_monoclinic_beta=True,
+        apply_sigma_correction=True,
     ):
         """Initialise a LaueGroupAnalysis object.
 
@@ -77,6 +78,7 @@ def __init__(
             relative_length_tolerance=relative_length_tolerance,
             absolute_angle_tolerance=absolute_angle_tolerance,
             best_monoclinic_beta=best_monoclinic_beta,
+            apply_sigma_correction=apply_sigma_correction,
         )
 
         self._estimate_cc_sig_fac()

src/dials/command_line/cosym.py

@@ -97,6 +97,10 @@ def __init__(self, experiments, reflections, params=None):
             params = phil_scope.extract()
         self.params = params
 
+        # if all datasets have been through scaling, a decision about error models has
+        # been made, so don't apply any further sigma correction
+        apply_sigma_correction = not all(s for s in experiments.scaling_models())
+
         reference_intensities = None
         if self.params.reference:
             wl = np.mean([expt.beam.get_wavelength() for expt in experiments])
@@ -191,10 +195,15 @@ def __init__(self, experiments, reflections, params=None):
             # The purpose of the reference is to help the clustering, not guarantee the indexing solution.
             datasets.append(reference_intensities)
             self.cosym_analysis = CosymAnalysis(
-                datasets, self.params, seed_dataset=len(datasets) - 1
+                datasets,
+                self.params,
+                seed_dataset=len(datasets) - 1,
+                apply_sigma_correction=apply_sigma_correction,
             )
         else:
-            self.cosym_analysis = CosymAnalysis(datasets, self.params)
+            self.cosym_analysis = CosymAnalysis(
+                datasets, self.params, apply_sigma_correction=apply_sigma_correction
+            )
 
     @property
     def experiments(self):

src/dials/command_line/symmetry.py

@@ -405,6 +405,10 @@ def symmetry(experiments, reflection_tables, params=None):
     input data and filtering settings e.g partiality_threshold"""
             )
 
+        # if all datasets have been through scaling, a decision about error models has
+        # been made, so don't apply any further sigma correction
+        apply_sigma_correction = not all(s for s in experiments.scaling_models())
+
         datasets = [
             ma.as_anomalous_array().merge_equivalents().array() for ma in datasets
         ]
@@ -417,6 +421,7 @@ def symmetry(experiments, reflection_tables, params=None):
             relative_length_tolerance=params.relative_length_tolerance,
             absolute_angle_tolerance=params.absolute_angle_tolerance,
             best_monoclinic_beta=params.best_monoclinic_beta,
+            apply_sigma_correction=apply_sigma_correction,
         )
         logger.info("")
         logger.info(result)