Implement comprehensive save/load tests for variants

Added a test to validate saving and loading for all standardize variants of the ContinuousApproximator.

Author: SermetPekin

tests/test_approximators/test_approximator_standardization/test_approximator_standardization.py

@@ -1,4 +1,5 @@
 import keras
+import pytest
 from tests.utils import assert_models_equal
 
 
@@ -16,3 +17,54 @@ def test_save_and_load(tmp_path, approximator, train_dataset, validation_dataset
     loaded = keras.saving.load_model(tmp_path / "model.keras")
 
     assert_models_equal(approximator, loaded)
+
+
+def test_save_and_load_all_variants(
+    tmp_path, adapter, inference_network, summary_network, train_dataset, validation_dataset
+):
+    """Run the same save/load assertions for all `standardize` options in one test node.
+
+    This avoids relying on pytest's bracketed node-id selection; it constructs
+    an approximator for each `standardize` value and runs the same checks.
+    Any failures across variants are aggregated and reported at the end.
+    """
+    from bayesflow import ContinuousApproximator
+
+    standardize_values = [
+        "all",
+        None,
+        "inference_variables",
+        "summary_variables",
+        ("inference_variables", "summary_variables", "inference_conditions"),
+    ]
+
+    failures = []
+
+    for standardize in standardize_values:
+        approximator = ContinuousApproximator(
+            adapter=adapter,
+            inference_network=inference_network,
+            summary_network=summary_network,
+            standardize=standardize,
+        )
+
+        try:
+            data_shapes = keras.tree.map_structure(keras.ops.shape, train_dataset[0])
+            approximator.build(data_shapes)
+            for layer in approximator.standardize_layers.values():
+                assert layer.built
+                for count in layer.count:
+                    assert count == 0.0
+            approximator.compute_metrics(**train_dataset[0])
+
+            model_path = tmp_path / f"model_{str(standardize)}.keras"
+            keras.saving.save_model(approximator, model_path)
+            loaded = keras.saving.load_model(model_path)
+
+            assert_models_equal(approximator, loaded)
+        except Exception as exc:  # collect failures and continue
+            failures.append((standardize, repr(exc)))
+
+    if failures:
+        msgs = ", ".join([f"{s!r}: {m}" for s, m in failures])
+        pytest.fail(f"One or more standardize variants failed: {msgs}")