Question: Best practice for direct variable initialization about keras HOT 6 OPEN

@keras.saving.register_keras_serializable("my_package")
class RollInitializer(keras.Initializer):
    def __call__(self, shape, dtype=None):
        return keras.ops.roll(keras.ops.arange(shape[-1]), shift=shape[-1] // 2)


@keras.saving.register_keras_serializable("my_package")
class SomeLayer(keras.Layer):
    def build(self, input_shape):
        self.indices = self.add_variable(shape=(input_shape[-1],), initializer=RollInitializer(), trainable=False)

class SomeLayer(keras.Layer):
    def build(self, input_shape):
        indices = keras.ops.roll(
            keras.ops.arange(input_shape[-1]), shift=input_shape[-1] // 2
        )

        self.indices = self.add_variable(
            shape=input_shape[-1],
            initializer=lambda shape, dtype: indices,  # <-----
            trainable=False,
        )

self.constants = self.add_weight(shape=..., initializer=7)
indices = keras.ops.roll(....)
self.indices = self.add_weight(shape=..., initializer=indices)

self.zeros = self.add_weight(shape=..., initializer=keras.initializers.Constant(7))
indices = keras.ops.roll(...)
self.indices = self.add_weight(shaep=..., initializer=lambda shape, dtype: indices)