Allowing out-of-search-space samples through cli, resume and default

This change just gives the user more freedom over which data to input. For real world cases this can be useful. In the future more safety features will be added in to make sure that this is what the user wants as well as go allow the same freedom through the non-cli interface.

hypermapper/param/parameters.py

@@ -357,10 +357,10 @@ def convert(
             - the converted value
         """
         if from_type in ["string", "internal"]:
-            intermediate_value = int(input_value)
+            intermediate_value = np.float64(input_value)
         elif from_type == "01":
             intermediate_value = (
-                int(
+                np.float64(
                     np.floor(input_value * (self.get_max() + 0.999999 - self.get_min()))
                 )
                 + self.get_min()
@@ -369,6 +369,8 @@ def convert(
             intermediate_value = input_value
 
         if to_type == "string":
+            if abs(np.round(intermediate_value) - intermediate_value) < 1e-8:
+                return f"{int(np.round(intermediate_value))}"
             return f"{intermediate_value}"
         elif to_type == "01":
             return (intermediate_value - self.get_min()) / (
@@ -521,21 +523,10 @@ def convert(
             ]
         else:
             intermediate_value = input_value
-            # this is a fix to safeguard against numerical imprecision
-            if intermediate_value not in self.values:
-                closest_value = min(
-                    self.values, key=lambda x: abs(x - intermediate_value)
-                )
-                if abs(closest_value - intermediate_value) > 1e-6:
-                    raise Exception(
-                        "The input value in OrdinalParameter.convert() is not in the list of values."
-                        f" Value: {intermediate_value} closest value: {closest_value}"
-                    )
-                intermediate_value = closest_value
-
+
         if to_type == "string":
-            if self.int_ordinal:
-                return f"{int(intermediate_value)}"
+            if self.int_ordinal and abs(intermediate_value - np.round(intermediate_value)) < 1e-8:
+                return f"{int(np.round(intermediate_value))}"
             else:
                 return f"{intermediate_value}"
         elif to_type == "01":

hypermapper/param/space.py

@@ -396,8 +396,8 @@ def get_default_configuration(self):
             return None
         default_configuration = torch.tensor([default_configuration])
         if self.conditional_space and not self.evaluate(default_configuration):
-            sys.stdout.write_to_logfile("Warning: infeasible  default configuration.")
-            return None
+            sys.stdout.write_to_logfile("Warning: the default configuration is infeasible. Are you sure you want this?.")
+            return default_configuration
         return default_configuration
 
     def convert(
@@ -625,9 +625,14 @@ def run_configurations_client_server(
             metric_indices = [
                 parameters_header.index(name) for name in self.metric_names
             ]
+            param_indices = [
+                parameters_header.index(name) for name in self.parameter_names
+            ]
+            new_configurations = []
             metrics = []
             feasible = []
             for row in parameters_data:
+                new_configurations.append([row[p] for p in param_indices])
                 metrics.append([float(row[m]) for m in metric_indices])
                 if self.enable_feasible_predictor:
                     feasible.append(
@@ -644,8 +649,12 @@ def run_configurations_client_server(
             print("Failed to parse received message:")
             print(ve)
             raise SystemError
+
+        new_configurations = self.convert(
+            new_configurations, "string", "internal"
+        )
         timestamps = torch.tensor([self.current_milli_time()] * len(configurations))
-        new_data_array = DataArray(configurations, metrics, timestamps, feasible)
+        new_data_array = DataArray(new_configurations, metrics, timestamps, feasible)
         write_data_array(self, new_data_array, output_data_file)
         return new_data_array