Use SeedGenerator backend when creating variable (#439)

keras_core/layers/preprocessing/tf_data_layer.py

@@ -1,6 +1,6 @@
 from tensorflow import nest
 
-from keras_core import backend
+import keras_core.backend
 from keras_core.layers.layer import Layer
 from keras_core.random.seed_generator import SeedGenerator
 from keras_core.utils import backend_utils
@@ -22,7 +22,7 @@ def __init__(self, **kwargs):
 
     def __call__(self, inputs, **kwargs):
         if backend_utils.in_tf_graph() and not isinstance(
-            inputs, backend.KerasTensor
+            inputs, keras_core.backend.KerasTensor
         ):
             # We're in a TF graph, e.g. a tf.data pipeline.
             self.backend.set_backend("tensorflow")
@@ -47,7 +47,7 @@ def __call__(self, inputs, **kwargs):
 
     @tracking.no_automatic_dependency_tracking
     def _get_seed_generator(self, backend=None):
-        if backend is None or backend == self.backend._backend:
+        if backend is None or backend == keras_core.backend.backend():
             return self.generator
         if not hasattr(self, "_backend_generators"):
             self._backend_generators = {}

keras_core/random/seed_generator.py

@@ -2,6 +2,7 @@
 
 import numpy as np
 
+import keras_core.backend
 from keras_core.api_export import keras_core_export
 
 
@@ -30,9 +31,9 @@ def __init__(self, seed, **kwargs):
         if kwargs:
             raise ValueError(f"Unrecognized keyword arguments: {kwargs}")
         if custom_backend is not None:
-            backend = custom_backend
+            self.backend = custom_backend
         else:
-            from keras_core import backend
+            self.backend = keras_core.backend
 
         if seed is None:
             seed = make_default_seed()
@@ -43,9 +44,9 @@ def __init__(self, seed, **kwargs):
 
         def seed_initializer(*args, **kwargs):
             dtype = kwargs.get("dtype", None)
-            return backend.convert_to_tensor([seed, 0], dtype=dtype)
+            return self.backend.convert_to_tensor([seed, 0], dtype=dtype)
 
-        self.state = backend.Variable(
+        self.state = self.backend.Variable(
             seed_initializer,
             shape=(2,),
             dtype="uint32",
@@ -65,7 +66,9 @@ def draw_seed(seed):
         seed_state = seed.state
         # Use * 1 to create a copy
         new_seed_value = seed_state.value * 1
-        increment = convert_to_tensor(np.array([0, 1]), dtype="uint32")
+        increment = seed.backend.convert_to_tensor(
+            np.array([0, 1]), dtype="uint32"
+        )
         seed.state.assign(seed_state + increment)
         return new_seed_value
     elif isinstance(seed, int):