gini as metric added

.gitignore

@@ -25,6 +25,7 @@ var/
 *.egg-info/
 .installed.cfg
 *.egg
+.idea/
 
 # PyInstaller
 #  Usually these files are written by a python script from a template

fairml/orthogonal_projection.py

@@ -11,6 +11,7 @@
 # import a few utility functions
 from .utils import mse
 from .utils import accuracy
+from .utils import gini
 from .utils import replace_column_of_matrix
 from .utils import detect_feature_sign
 
@@ -126,7 +127,7 @@ def audit_model(predict_function, input_dataframe, distance_metric="mse",
 
     input_dataframe -> dataframe with shape (n_samples, n_features)
 
-    distance_metric -> one of ["mse", "accuracy"], this
+    distance_metric -> one of ["mse", "accuracy", "gini"], this
                 variable defaults to regression.
 
     direct_input_pertubation_strategy -> This is referring to how to zero out a
@@ -155,8 +156,8 @@ def audit_model(predict_function, input_dataframe, distance_metric="mse",
     """
     assert isinstance(input_dataframe, pd.DataFrame), ("Data must be a pandas "
                                                        "dataframe")
-    assert distance_metric in ["mse", "accuracy"], ("Distance metric must be "
-                                                    "'mse' or 'accuracy'")
+    assert distance_metric in ["mse", "accuracy", "gini"], ("Distance metric must be "
+                                                    "'mse', 'accuracy' or 'gini'")
     assert direct_input_pertubation_strategy in ["constant-zero",
                                                  "constant-median",
                                                  "random-sample"
@@ -219,6 +220,9 @@ def audit_model(predict_function, input_dataframe, distance_metric="mse",
             if distance_metric == "accuracy":
                 output_difference_col = accuracy(
                     output_constant_col, normal_black_box_output)
+            elif distance_metric == "gini":
+                output_difference_col = gini(
+                    output_constant_col, normal_black_box_output)
             else:
                 output_difference_col = mse(
                     output_constant_col, normal_black_box_output)
@@ -241,6 +245,9 @@ def audit_model(predict_function, input_dataframe, distance_metric="mse",
             if distance_metric == "accuracy":
                 total_difference = accuracy(
                     total_transformed_output, normal_black_box_output)
+            elif distance_metric == "gini":
+                total_difference = gini(
+                    total_transformed_output, normal_black_box_output)
             else:
                 total_difference = mse(
                     total_transformed_output, normal_black_box_output)

fairml/tests/test_orthogonal_projection.py

@@ -10,6 +10,7 @@
 
 from fairml.utils import mse
 from fairml.utils import accuracy
+from fairml.utils import gini
 from fairml.utils import detect_feature_sign
 
 from fairml.perturbation_strategies import constant_zero
@@ -60,3 +61,11 @@ def test_accuracy():
     test_acc = accuracy(y_pred, y_true)
     print(test_acc)
     assert test_acc == 0.5
+
+def test_gini():
+    y_pred = [0, 0, 0, 1]
+    y_true = [0, 0, 1, 1]
+
+    test_gini= gini(y_true, y_pred)
+    print(test_gini)
+    assert test_gini == 0.5

fairml/utils.py

@@ -3,6 +3,7 @@
 from __future__ import print_function
 
 import numpy as np
+from sklearn.metrics import roc_auc_score
 
 # import dictionary with perturbation strategies.
 from .perturbation_strategies import perturbation_strategy_dictionary
@@ -40,6 +41,22 @@ def accuracy(y, y_hat):
 
     return accuracy
 
+def gini(y, y_hat):
+    """ function to calculate gini of y_hat given y"""
+    y = np.array(y)
+    y_hat = np.array(y_hat)
+
+    y = y.astype(int)
+    y_hat = y_hat.astype(float)
+
+    y_hat = np.reshape(y_hat, (y_hat.shape[0],))
+    y = np.reshape(y, (y.shape[0],))
+
+    ras = roc_auc_score(y, y_hat)
+    g = ras * 2 - 1
+
+    return g
+
 
 def replace_column_of_matrix(X, col_num, random_sample,
                              ptb_strategy):