Features Invoked Based on AI and Code Quality Improvements

lfr_with-ai-features-and-data-visualization.py

@@ -0,0 +1,112 @@
+import os
+import sys
+
+import numpy as np
+import pytest
+import matplotlib.pyplot as plt  # For visualization
+import seaborn as sns  # For advanced data visualization
+
+from aif360.algorithms.preprocessing import LFR
+from aif360.algorithms.preprocessing.optim_preproc_helpers.data_preproc_functions import load_preproc_data_adult
+
+sys.path.append("../../../")
+sys.path.append(os.path.dirname(__file__))
+
+# AI-Driven Enhancements: Visualization function for analyzing distribution before and after LFR transformation
+def visualize_data_distribution(dataset, title="Data Distribution"):
+    """Visualize the distribution of features in the dataset."""
+    plt.figure(figsize=(10, 6))
+    sns.histplot(data=dataset.features, kde=True)
+    plt.title(title)
+    plt.xlabel("Feature Values")
+    plt.ylabel("Frequency")
+    plt.show()
+
+@pytest.fixture(scope="module")
+def lfr_algorithm_instance():
+    """Fixture to create an instance of the LFR algorithm."""
+    privileged_groups = [{'sex': 1.0}]
+    unprivileged_groups = [{'sex': 0.0}]
+
+    lfr_instance = LFR(unprivileged_groups=unprivileged_groups, privileged_groups=privileged_groups)
+
+    return lfr_instance
+
+@pytest.fixture(scope="module")
+def adult_data():
+    """Fixture to load and split the adult dataset."""
+    return load_preproc_data_adult().split([0.7], shuffle=True)[0]
+
+@pytest.fixture(scope="module")
+def lfr_fit_model():
+    """Fixture to fit the LFR model on the adult dataset."""
+    privileged_groups = [{'sex': 1.0}]
+    unprivileged_groups = [{'sex': 0.0}]
+
+    lfr_instance = LFR(unprivileged_groups=unprivileged_groups, privileged_groups=privileged_groups)
+    ad = load_preproc_data_adult().split([0.7], shuffle=True)[0]
+
+    # AI Enhancement: Visualize the original data distribution before fitting
+    visualize_data_distribution(ad, title="Original Data Distribution")
+
+    transformed_data = lfr_instance.fit(ad)
+
+    # AI Enhancement: Visualize the transformed data distribution after fitting
+    visualize_data_distribution(transformed_data, title="Transformed Data Distribution (After LFR Fit)")
+
+    return transformed_data
+
+def test_fit_is_numpy(lfr_fit_model):
+    """Test that the fit function returns a numpy array."""
+    expected = True
+    result = isinstance(lfr_fit_model.learned_model, np.ndarray)
+    assert result == expected
+
+def test_fit_not_null(lfr_fit_model):
+    """Test that the fit function's result is not null."""
+    expected = False
+    result = lfr_fit_model.learned_model is None
+    assert result == expected
+
+def test_fit_not_all_zeros(lfr_fit_model):
+    """Test that the fit function's result is not all zeros."""
+    expected = False
+    all_zeros = not np.any(lfr_fit_model.learned_model)
+    assert all_zeros == expected
+
+def test_fit_not_nan(lfr_fit_model):
+    """Test that the fit function's result does not contain NaNs."""
+    expected = False
+    result = np.isnan(lfr_fit_model.learned_model).any()
+    assert result == expected
+
+# --------------------------------------------------------------------#
+# Transform function testing methods
+# --------------------------------------------------------------------#
+
+def test_transform_protected_dataset(lfr_fit_model, adult_data):
+    """Test that the protected attributes remain unchanged after transformation."""
+    transformed_data = lfr_fit_model.transform(adult_data, threshold=0.3)
+    result = np.array_equal(transformed_data.protected_attributes, adult_data.protected_attributes)
+    assert result == True
+
+    # AI Enhancement: Visualize the protected attribute distribution after transformation
+    visualize_data_distribution(transformed_data, title="Protected Attributes After LFR Transformation")
+
+def test_transform_not_nan(lfr_fit_model, adult_data):
+    """Test that the transformed data does not contain NaNs and does not have columns or rows summing up to zero."""
+    transformed_data = lfr_fit_model.transform(adult_data, threshold=0.3)
+    row_sums = np.sum(transformed_data.features, axis=1).tolist()
+    col_sums = np.sum(transformed_data.features, axis=0).tolist()
+    all_row_zeros = not np.any(row_sums)
+    all_col_zeros = not np.any(col_sums)
+    assert (all_row_zeros and all_col_zeros) == False
+
+def test_transform_not_nan2(lfr_fit_model, adult_data):
+    """Test that the transformed data does not contain NaNs with a threshold value of 0.3."""
+    transformed_data = lfr_fit_model.transform(adult_data, threshold=0.3)
+    result = np.isnan(transformed_data.features).any()
+    assert result == False
+
+    # AI Enhancement: Visualize the transformed data distribution for NaN checking
+    visualize_data_distribution(transformed_data, title="Transformed Data Distribution (No NaNs)")