Add EqualizedOddsImprovement

Author: fealho

sdmetrics/single_table/__init__.py

@@ -77,6 +77,7 @@
 from sdmetrics.single_table.privacy.numerical_sklearn import NumericalLR, NumericalMLP, NumericalSVR
 from sdmetrics.single_table.privacy.radius_nearest_neighbor import NumericalRadiusNearestNeighbor
 from sdmetrics.single_table.table_structure import TableStructure
+from sdmetrics.single_table.equalized_odds import EqualizedOddsImprovement
 
 __all__ = [
     'bayesian_network',
@@ -140,4 +141,5 @@
     'TableStructure',
     'DCRBaselineProtection',
     'DCROverfittingProtection',
+    'EqualizedOddsImprovement',
 ]

sdmetrics/single_table/data_augmentation/utils.py

@@ -1,33 +1,12 @@
 """Utils method for data augmentation metrics."""
 
-import pandas as pd
-
 from sdmetrics._utils_metadata import _process_data_with_metadata, _validate_single_table_metadata
-
-
-def _validate_tables(real_training_data, synthetic_data, real_validation_data):
-    """Validate the tables of the Data Augmentation metrics."""
-    tables = [real_training_data, synthetic_data, real_validation_data]
-    if any(not isinstance(table, pd.DataFrame) for table in tables):
-        raise ValueError(
-            '`real_training_data`, `synthetic_data` and `real_validation_data` must be '
-            'pandas DataFrames.'
-        )
-
-
-def _validate_prediction_column_name(prediction_column_name):
-    """Validate the prediction column name of the Data Augmentation metrics."""
-    if not isinstance(prediction_column_name, str):
-        raise TypeError('`prediction_column_name` must be a string.')
-
-
-def _validate_classifier(classifier):
-    """Validate the classifier of the Data Augmentation metrics."""
-    if classifier is not None and not isinstance(classifier, str):
-        raise TypeError('`classifier` must be a string or None.')
-
-    if classifier != 'XGBoost':
-        raise ValueError('Currently only `XGBoost` is supported as classifier.')
+from sdmetrics.single_table.utils import (
+    _validate_classifier,
+    _validate_data_and_metadata,
+    _validate_prediction_column_name,
+    _validate_tables,
+)
 
 
 def _validate_fixed_recall_value(fixed_recall_value):
@@ -53,51 +32,6 @@ def _validate_parameters(
     _validate_fixed_recall_value(fixed_recall_value)
 
 
-def _validate_data_and_metadata(
-    real_training_data,
-    synthetic_data,
-    real_validation_data,
-    metadata,
-    prediction_column_name,
-    minority_class_label,
-):
-    """Validate the data and metadata of the Data Augmentation metrics."""
-    if prediction_column_name not in metadata['columns']:
-        raise ValueError(
-            f'The column `{prediction_column_name}` is not described in the metadata.'
-            ' Please update your metadata.'
-        )
-
-    if metadata['columns'][prediction_column_name]['sdtype'] not in ('categorical', 'boolean'):
-        raise ValueError(
-            f'The column `{prediction_column_name}` must be either categorical or boolean.'
-            ' Please update your metadata.'
-        )
-
-    if minority_class_label not in real_training_data[prediction_column_name].unique():
-        raise ValueError(
-            f'The value `{minority_class_label}` is not present in the column '
-            f'`{prediction_column_name}` for the real training data.'
-        )
-
-    if minority_class_label not in real_validation_data[prediction_column_name].unique():
-        raise ValueError(
-            f"The metric can't be computed because the value `{minority_class_label}` "
-            f'is not present in the column `{prediction_column_name}` for the real validation data.'
-            ' The `precision` and `recall` are undefined for this case.'
-        )
-
-    synthetic_labels = set(synthetic_data[prediction_column_name].unique())
-    real_labels = set(real_training_data[prediction_column_name].unique())
-    if not synthetic_labels.issubset(real_labels):
-        to_print = "', '".join(sorted(synthetic_labels - real_labels))
-        raise ValueError(
-            f'The ``{prediction_column_name}`` column must have the same values in the real '
-            'and synthetic data. The following values are present in the synthetic data and'
-            f" not the real data: '{to_print}'"
-        )
-
-
 def _validate_inputs(
     real_training_data,
     synthetic_data,
@@ -127,6 +61,16 @@ def _validate_inputs(
         minority_class_label,
     )
 
+    synthetic_labels = set(synthetic_data[prediction_column_name].unique())
+    real_labels = set(real_training_data[prediction_column_name].unique())
+    if not synthetic_labels.issubset(real_labels):
+        to_print = "', '".join(sorted(synthetic_labels - real_labels))
+        raise ValueError(
+            f'The `{prediction_column_name}` column must have the same values in the real '
+            'and synthetic data. The following values are present in the synthetic data and'
+            f" not the real data: '{to_print}'"
+        )
+
 
 def _process_data_with_metadata_ml_efficacy_metrics(
     real_training_data, synthetic_data, real_validation_data, metadata