Check stateless

Small change where background model samples are not loaded if the main pipeline is stateless. The world set typically is much larger than the dev and eval set, so it is a bit of a waste to load.

NB. The make_pipeline utility from scikit-learn generates pipelines that are never stateless, even if all transformers in the sequence are stateless, which makes this change useful.

bob/bio/base/pipelines/vanilla_biometrics/vanilla_biometrics.py

@@ -10,8 +10,7 @@ from bob.bio.base.pipelines.vanilla_biometrics import ZTNormPipeline
 from bob.bio.base.pipelines.vanilla_biometrics import checkpoint_vanilla_biometrics
 from bob.bio.base.pipelines.vanilla_biometrics import dask_vanilla_biometrics
 from bob.bio.base.pipelines.vanilla_biometrics import is_checkpointed
-from bob.pipelines.distributed import dask_get_partition_size
-from bob.pipelines.utils import isinstance_nested
+from bob.pipelines.utils import isinstance_nested, is_estimator_stateless
 from dask.delayed import Delayed
 
 logger = logging.getLogger(__name__)
@@ -93,7 +92,11 @@ def execute_vanilla_biometrics(
         hash_fn = database.hash_fn if hasattr(database, "hash_fn") else None
         pipeline = checkpoint_vanilla_biometrics(pipeline, output, hash_fn=hash_fn)
 
-    background_model_samples = database.background_model_samples()
+    # Load the background model samples only if the transformer requires fitting
+    if all([is_estimator_stateless(step) for step in pipeline.transformer]):
+        background_model_samples = []
+    else:
+        background_model_samples = database.background_model_samples()
 
     for group in groups: