Cover cases with no conditions

bayesflow/approximators/continuous_approximator.py

@@ -121,10 +121,12 @@ def get_config(self):
 
         return base_config | config
 
-    def sample(self, data: Mapping[str, Tensor], num_samples: int = 1, numpy: bool = True) -> dict[str, Tensor]:
+    def sample(
+        self, data: Mapping[str, Tensor], num_samples: int = 1, numpy: bool = True, batch_shape: Shape = None
+    ) -> dict[str, Tensor]:
         data = self.data_adapter.configure(data)
         data = keras.tree.map_structure(keras.ops.convert_to_tensor, data)
-        data = {"inference_variables": self._sample(num_samples, **data)}
+        data = {"inference_variables": self._sample(num_samples, batch_shape=batch_shape, **data)}
         data = self.data_adapter.deconfigure(data)
 
         if numpy:
@@ -133,7 +135,11 @@ def sample(self, data: Mapping[str, Tensor], num_samples: int = 1, numpy: bool =
         return data
 
     def _sample(
-        self, num_samples: Shape, inference_conditions: Tensor = None, summary_variables: Tensor = None
+        self,
+        num_samples: int,
+        batch_shape: Shape = None,
+        inference_conditions: Tensor = None,
+        summary_variables: Tensor = None,
     ) -> Tensor:
         if self.summary_network is not None:
             summary_outputs = self.summary_network(summary_variables)
@@ -143,8 +149,12 @@ def _sample(
             else:
                 inference_conditions = keras.ops.concatenate([inference_conditions, summary_outputs], axis=-1)
 
-        inference_conditions = expand_tile(inference_conditions, axis=1, n=num_samples)
-        batch_shape = (keras.ops.shape(inference_conditions)[0], num_samples)
+        if batch_shape is None:
+            if inference_conditions is not None:
+                inference_conditions = expand_tile(inference_conditions, axis=1, n=num_samples)
+                batch_shape = (keras.ops.shape(inference_conditions)[0], num_samples)
+            else:
+                batch_shape = (num_samples,)
 
         return self.inference_network.sample(batch_shape, conditions=inference_conditions)