Merge pull request #59 from outbrain/rank_joined_featured_from_modelspec

compute ranking for joined features in modelspec

outrank/__main__.py

@@ -165,7 +165,7 @@ def main():
     parser.add_argument(
         '--reference_model_JSON',
         type=str,
-        default='./ranking_outputs/reference_model.json',
+        default='',
         help='Reference model JSON',
     )
 

outrank/core_ranking.py

@@ -187,11 +187,15 @@ def compute_combined_features(
     join_string = ' AND_REL ' if is_3mr else ' AND '
     interaction_order = 2 if is_3mr else args.interaction_order
 
-    full_combination_space = list(
-        itertools.combinations(all_columns, interaction_order),
-    )
+    if args.reference_model_JSON != '':
+        combined_features = extract_features_from_reference_JSON(args.reference_model_JSON, combined_features_only = True)
+        full_combination_space = [combination.split(',') for combination in combined_features]
+    else:
+        full_combination_space = list(
+            itertools.combinations(all_columns, interaction_order),
+        )
 
-    if args.combination_number_upper_bound:
+    if args.combination_number_upper_bound and args.reference_model_JSON != '':
         random.shuffle(full_combination_space)
         full_combination_space = full_combination_space[
             : args.combination_number_upper_bound
@@ -517,7 +521,7 @@ def compute_batch_ranking(
             input_dataframe, logger, args, pbar,
         )
 
-    if args.interaction_order > 1:
+    if args.interaction_order > 1 or args.reference_model_JSON:
         pbar.set_description('Constructing new features')
         input_dataframe = compute_combined_features(
             input_dataframe, logger, args, pbar,

outrank/core_utils.py

@@ -393,7 +393,7 @@ def parse_csv_raw(data_path) -> DatasetInformationStorage:
     )
 
 
-def extract_features_from_reference_JSON(json_path: str) -> set[Any]:
+def extract_features_from_reference_JSON(json_path: str, combined_features_only = False) -> set[Any]:
     """Given a model's JSON, extract unique features"""
 
     with open(json_path) as jp:
@@ -404,6 +404,9 @@ def extract_features_from_reference_JSON(json_path: str) -> set[Any]:
     fields_space = content['desc'].get('fields', [])
     joint_space = feature_space + fields_space
 
+    if combined_features_only:
+        return {feature for feature in feature_space if len(feature.split(','))>1}
+
     for feature_tuple in joint_space:
         for individual_feature in feature_tuple.split(','):
             unique_features.add(individual_feature)

setup.py

@@ -23,7 +23,7 @@ def _read_description():
 packages = [x for x in setuptools.find_packages() if x != 'test']
 setuptools.setup(
     name='outrank',
-    version='0.95.5',
+    version='0.95.6',
     description='OutRank: Feature ranking for massive sparse data sets.',
     long_description=_read_description(),
     long_description_content_type='text/markdown',

tests/ranking_module_test.py

@@ -35,6 +35,7 @@ class args:
     combination_number_upper_bound: int = 1024
     disable_tqdm: bool = False
     mi_stratified_sampling_ratio: float = 1.0
+    reference_model_JSON: str = ''
 
 
 class CompareStrategiesTest(unittest.TestCase):