DemandModel: Improve save_model robustness.

Use `convert_marginal_params` before saving instead of manually taking
care of lognormal parameters.
Extends `convert_marginal_params` to support inverse conversions (from
internal units to user-defined).

pelicun/model/demand_model.py

@@ -691,26 +691,17 @@ def save_model(self, file_prefix):
                 log=self._asmnt.log,
             )
 
-        # the log standard deviations in the marginal parameters need to be
-        # scaled up before feeding to the saving method where they will be
-        # scaled back down and end up being saved unscaled to the target file
-
-        marginal_params = self.marginal_params.copy()
-
-        log_rows = marginal_params['Family'] == 'lognormal'
-        log_demands = marginal_params.loc[log_rows, :]
-
-        for label in log_demands.index:
-            if label in self.user_units.index:
-                unit_factor = self._asmnt.calc_unit_scale_factor(self.user_units[label])
-
-                marginal_params.loc[label, 'Theta_1'] *= unit_factor
+        # Converting the marginal parameters requires special
+        # treatment, so we can't rely on file_io's universal unit
+        # conversion functionality. We do it manually here instead.
+        marginal_params_user_units = self._convert_marginal_params(
+            self.marginal_params.copy(), self.user_units, inverse_conversion=True
+        )
+        marginal_params_user_units['Units'] = self.user_units
 
         file_io.save_to_csv(
-            marginal_params,
+            marginal_params_user_units,
             file_prefix + '_marginals.csv',
-            units=self.user_units,
-            unit_conversion_factors=self._asmnt.unit_conversion_factors,
             orientation=1,
             log=self._asmnt.log,
         )
@@ -767,10 +758,6 @@ def load_model(self, data_source):
         else:
             self.correlation = None
 
-        # the log standard deviations in the marginal parameters need to be
-        # adjusted after getting the data from the loading method where they
-        # were scaled according to the units of the corresponding variable
-
         # Note that a data source without marginal information is not valid
         marginal_params, units = file_io.load_data(
             marginal_data_source,

pelicun/model/pelicun_model.py

@@ -79,7 +79,12 @@ def __init__(self, assessment: Assessment):
         self.log = self._asmnt.log
 
     def _convert_marginal_params(
-        self, marginal_params, units, arg_units=None, divide_units=True
+        self,
+        marginal_params,
+        units,
+        arg_units=None,
+        divide_units=True,
+        inverse_conversion=False,
     ):
         """
         Converts the parameters of marginal distributions in a model to SI units.
@@ -110,6 +115,11 @@ def _convert_marginal_params(
             should be True when the arg units represent the quantity
             corresponding to the primary parameters, and False
             otherwise.
+        inverse_conversion: bool
+            If True, converts from user-defined units to internal. If
+            False, converts from internal units to
+            user-defined. Defaults to False, since the method is
+            mostly applied on user-defined data.
 
         Returns
         -------
@@ -211,11 +221,13 @@ def _convert_marginal_params(
                             if isinstance(arg, np.ndarray):
                                 args[a_i] = arg * arg_unit_factor
 
-                # convert the distribution parameters to SI
+                # convert units
                 if divide_units:
                     conversion_factor = unit_factor / arg_unit_factor
                 else:
                     conversion_factor = unit_factor
+                if inverse_conversion:
+                    conversion_factor = 1.00 / conversion_factor
                 theta, tr_limits = uq.scale_distribution(
                     conversion_factor, family, theta, tr_limits
                 )
@@ -325,9 +337,7 @@ def _get_locations(self, loc_str):
             if loc_str == "roof":
                 return np.array([stories + 1]).astype(str)
 
-            raise ValueError(
-                f"Cannot parse location string: " f"{loc_str}"
-            ) from exc
+            raise ValueError(f"Cannot parse location string: " f"{loc_str}") from exc
 
     def _get_directions(self, dir_str):
         """