Feature spline regularization

pypesto/C.py

@@ -194,7 +194,14 @@ class InnerParameterType(str, Enum):
 
 SPLINE_RATIO = 'spline_ratio'
 MIN_DIFF_FACTOR = 'min_diff_factor'
-SPLINE_APPROXIMATION_OPTIONS = [SPLINE_RATIO, MIN_DIFF_FACTOR]
+REGULARIZE_SPLINE = 'regularize_spline'
+REGULARIZATION_FACTOR = 'regularization_factor'
+SPLINE_APPROXIMATION_OPTIONS = [
+    SPLINE_RATIO,
+    MIN_DIFF_FACTOR,
+    REGULARIZE_SPLINE,
+    REGULARIZATION_FACTOR,
+]
 
 MIN_SIM_RANGE = 1e-16
 

pypesto/ensemble/ensemble.py

@@ -853,12 +853,13 @@ def _map_parameters_by_objective(
         """
         # create short hands
         parameter_ids_objective = predictor.amici_objective.x_names
-        parameter_ids_ensemble = self.x_names
+        parameter_ids_ensemble = list(self.x_names)
         # map, and fill with `default_value` if not found and `default_value`
         # is specified.
         mapping = []
         for parameter_id_objective in parameter_ids_objective:
             if parameter_id_objective in parameter_ids_ensemble:
+                # Append index of parameter in ensemble.
                 mapping.append(
                     parameter_ids_ensemble.index(parameter_id_objective)
                 )

pypesto/hierarchical/inner_calculator_collector.py

@@ -496,18 +496,33 @@ def calculate_quantitative_result(
                 ]
             )
             # calculate the gradient for the condition
-            gradient_for_condition = ssigma_i @ (
-                (
-                    np.full(len(data_i), 1)
-                    - (data_i - sim_i) ** 2 / sigma_i**2
-                )
-                / sigma_i
-            ) - sensitivities_i @ ((data_i - sim_i) / sigma_i**2)
+            gradient_for_condition = np.nansum(
+                np.multiply(
+                    ssigma_i,
+                    (
+                        (
+                            np.full(len(data_i), 1)
+                            - (data_i - sim_i) ** 2 / sigma_i**2
+                        )
+                        / sigma_i
+                    ),
+                ),
+                axis=1,
+            ) + np.nansum(
+                np.multiply(
+                    sensitivities_i, ((sim_i - data_i) / sigma_i**2)
+                ),
+                axis=1,
+            )
 
             # add gradient to correct index of snllh
             for par_sim, par_opt in condition_map_sim_var.items():
                 if not isinstance(par_opt, str):
                     continue
+
+                if par_opt not in par_opt_ids:
+                    continue
+
                 par_opt_idx = par_opt_ids.index(par_opt)
                 par_sim_idx = par_sim_ids.index(par_sim)
                 par_edata_idx = (

pypesto/hierarchical/spline_approximation/calculator.py

@@ -228,6 +228,7 @@ def __call__(
                 parameter_mapping=parameter_mapping,
                 par_opt_ids=x_ids,
                 par_sim_ids=amici_model.getParameterIds(),
+                par_edatas_indices=[edata.plist for edata in edatas],
                 snllh=snllh,
             )