String seed = "test_seed";
        String[] tags = null;
        String[] roles = null;
        String[] fields = null;
        String[] excludes = null;

        String result = popularWordHelper.getCacheKey(seed, tags, roles, fields, excludes);

        assertNotNull(result);
        assertTrue(result.contains("test_seed"));

        `(batch_size, 1)`. This enables `Model` subclasses to process scalar
        data in their `train_step()`/`test_step()`/`predict_step()` methods. \
        Note that this change may break certain subclassed models. You can
        revert back to the previous behavior by adding upranking yourself in the
        `train_step()`/`test_step()`/`predict_step()` methods, e.g. `if
        x.shape.rank == 1: x = tf.expand_dims(x, axis=-1)`. Functional models as