approximator builds: add guards against building networks twice

Author: vpratz

bayesflow/approximators/continuous_approximator.py

@@ -71,12 +71,14 @@ def build(self, data_shapes: dict[str, tuple[int] | dict[str, dict]]) -> None:
         summary_outputs_shape = None
         inference_conditions_shape = data_shapes.get("inference_conditions", None)
         if self.summary_network is not None:
-            self.summary_network.build(data_shapes["summary_variables"])
+            if not self.summary_network.built:
+                self.summary_network.build(data_shapes["summary_variables"])
             summary_outputs_shape = self.summary_network.compute_output_shape(data_shapes["summary_variables"])
         inference_conditions_shape = concatenate_valid_shapes(
             [inference_conditions_shape, summary_outputs_shape], axis=-1
         )
-        self.inference_network.build(data_shapes["inference_variables"], inference_conditions_shape)
+        if not self.inference_network.built:
+            self.inference_network.build(data_shapes["inference_variables"], inference_conditions_shape)
         if self.standardize == "all":
             self.standardize = [
                 var

bayesflow/approximators/model_comparison_approximator.py

@@ -70,13 +70,16 @@ def build(self, data_shapes: dict[str, tuple[int] | dict[str, dict]]) -> None:
         summary_outputs_shape = None
         classifier_conditions_shape = data_shapes.get("classifier_conditions", None)
         if self.summary_network is not None:
-            self.summary_network.build(data_shapes["summary_variables"])
+            if not self.summary_network.built:
+                self.summary_network.build(data_shapes["summary_variables"])
             summary_outputs_shape = self.summary_network.compute_output_shape(data_shapes["summary_variables"])
         classifier_conditions_shape = concatenate_valid_shapes(
             [classifier_conditions_shape, summary_outputs_shape], axis=-1
         )
-        self.classifier_network.build(classifier_conditions_shape)
-        self.logits_projector.build(self.classifier_network.compute_output_shape(classifier_conditions_shape))
+        if not self.classifier_network.built:
+            self.classifier_network.build(classifier_conditions_shape)
+        if not self.logits_projector.built:
+            self.logits_projector.build(self.classifier_network.compute_output_shape(classifier_conditions_shape))
         if self.standardize == "all":
             self.standardize = [var for var in ["summary_variables", "classifier_conditions"] if var in data_shapes]
 

tests/test_two_moons/test_two_moons.py

@@ -60,6 +60,9 @@ def test_serialize_deserialize(tmp_path, approximator, train_dataset):
     mock_data_shapes = keras.tree.map_structure(keras.ops.shape, mock_data)
     approximator.build(mock_data_shapes)
 
+    # run a single batch through the approximator
+    approximator.compute_metrics(**mock_data)
+
     keras.saving.save_model(approximator, tmp_path / "model.keras")
     loaded_approximator = keras.saving.load_model(tmp_path / "model.keras")