ML (Sklearn Bridge)

from_sklearn_linear() (Helper)¶

Demonstrates the from_sklearn_linear function. This is a client-side (cleartext) helper function. It is not an FHE circuit and cannot be compiled with fhe.Compiler.

from_sklearn_linear converts a fitted sklearn linear model (LogisticRegression or LinearRegression) into an FHE-compatible model by quantizing the float weights into integers.

In [ ]:

from concrete_fhe_toolkit.ml.sklearn_bridge import from_sklearn_linear
from concrete_fhe_toolkit.ml.classes import FHELinearRegression, FHELogisticRegression
import numpy as np

# Create a mock sklearn-like LinearRegression model
class MockLinearRegression:
    pass

mock_model = MockLinearRegression()
mock_model.coef_ = np.array([1.5, -2.0])
mock_model.intercept_ = 3.0

# Convert to FHE model with scale=10
fhe_model = from_sklearn_linear(mock_model, scale=10)

assert isinstance(fhe_model, FHELinearRegression), f'Expected FHELinearRegression, got {type(fhe_model).__name__}'
assert fhe_model.weights == [15, -20], f'Weights mismatch: {fhe_model.weights}'
assert fhe_model.bias == 30, f'Bias mismatch: {fhe_model.bias}'
print(f'Converted linear model: weights={fhe_model.weights}, bias={fhe_model.bias}')

# Test with a mock LogisticRegression (has classes_ attribute)
class MockLogisticRegression:
    pass

mock_clf = MockLogisticRegression()
mock_clf.coef_ = np.array([[0.5, -1.0]])
mock_clf.intercept_ = np.array([2.0])
mock_clf.classes_ = np.array([0, 1])

fhe_clf = from_sklearn_linear(mock_clf, scale=100)
assert isinstance(fhe_clf, FHELogisticRegression), f'Expected FHELogisticRegression, got {type(fhe_clf).__name__}'
assert fhe_clf.weights == [50, -100], f'Weights mismatch: {fhe_clf.weights}'
print(f'Converted classifier: weights={fhe_clf.weights}, bias={fhe_clf.bias}')

print('from_sklearn_linear tests passed!')

from concrete_fhe_toolkit.ml.sklearn_bridge import from_sklearn_linear from concrete_fhe_toolkit.ml.classes import FHELinearRegression, FHELogisticRegression import numpy as np # Create a mock sklearn-like LinearRegression model class MockLinearRegression: pass mock_model = MockLinearRegression() mock_model.coef_ = np.array([1.5, -2.0]) mock_model.intercept_ = 3.0 # Convert to FHE model with scale=10 fhe_model = from_sklearn_linear(mock_model, scale=10) assert isinstance(fhe_model, FHELinearRegression), f'Expected FHELinearRegression, got {type(fhe_model).__name__}' assert fhe_model.weights == [15, -20], f'Weights mismatch: {fhe_model.weights}' assert fhe_model.bias == 30, f'Bias mismatch: {fhe_model.bias}' print(f'Converted linear model: weights={fhe_model.weights}, bias={fhe_model.bias}') # Test with a mock LogisticRegression (has classes_ attribute) class MockLogisticRegression: pass mock_clf = MockLogisticRegression() mock_clf.coef_ = np.array([[0.5, -1.0]]) mock_clf.intercept_ = np.array([2.0]) mock_clf.classes_ = np.array([0, 1]) fhe_clf = from_sklearn_linear(mock_clf, scale=100) assert isinstance(fhe_clf, FHELogisticRegression), f'Expected FHELogisticRegression, got {type(fhe_clf).__name__}' assert fhe_clf.weights == [50, -100], f'Weights mismatch: {fhe_clf.weights}' print(f'Converted classifier: weights={fhe_clf.weights}, bias={fhe_clf.bias}') print('from_sklearn_linear tests passed!')

from_sklearn_tree() (Helper)¶

Demonstrates the from_sklearn_tree function. This is a client-side (cleartext) helper function. It is not an FHE circuit and cannot be compiled with fhe.Compiler.

from_sklearn_tree converts a fitted sklearn DecisionTreeClassifier or DecisionTreeRegressor into an FHEDecisionTree. This requires a fitted model (with a tree_ attribute). Below we show the expected usage pattern.

In [ ]:

from concrete_fhe_toolkit.ml.sklearn_bridge import from_sklearn_tree
from concrete_fhe_toolkit.ml.classes import FHEDecisionTree

# from_sklearn_tree requires a fitted sklearn tree model with a tree_ attribute.
# We demonstrate expected error handling for unfitted models.
try:
    class UnfittedModel:
        pass
    from_sklearn_tree(UnfittedModel())
    assert False, 'Should have raised ValueError'
except ValueError as e:
    assert 'fitted sklearn decision tree' in str(e)
    print(f'Correctly caught unfitted model: {e}')

# If sklearn is available, perform a real conversion
try:
    from sklearn.tree import DecisionTreeClassifier
    import numpy as np
    X = np.array([[1, 2], [3, 4], [5, 6], [7, 8]])
    y = np.array([0, 1, 0, 1])
    sk_tree = DecisionTreeClassifier(max_depth=2).fit(X, y)
    fhe_tree = from_sklearn_tree(sk_tree, scale=1)
    assert isinstance(fhe_tree, FHEDecisionTree)
    assert isinstance(fhe_tree.tree, dict)
    print(f'Real conversion successful: tree={fhe_tree.tree}')
except ImportError:
    print('scikit-learn not installed, skipping real conversion test')

print('from_sklearn_tree tests passed!')

from concrete_fhe_toolkit.ml.sklearn_bridge import from_sklearn_tree from concrete_fhe_toolkit.ml.classes import FHEDecisionTree # from_sklearn_tree requires a fitted sklearn tree model with a tree_ attribute. # We demonstrate expected error handling for unfitted models. try: class UnfittedModel: pass from_sklearn_tree(UnfittedModel()) assert False, 'Should have raised ValueError' except ValueError as e: assert 'fitted sklearn decision tree' in str(e) print(f'Correctly caught unfitted model: {e}') # If sklearn is available, perform a real conversion try: from sklearn.tree import DecisionTreeClassifier import numpy as np X = np.array([[1, 2], [3, 4], [5, 6], [7, 8]]) y = np.array([0, 1, 0, 1]) sk_tree = DecisionTreeClassifier(max_depth=2).fit(X, y) fhe_tree = from_sklearn_tree(sk_tree, scale=1) assert isinstance(fhe_tree, FHEDecisionTree) assert isinstance(fhe_tree.tree, dict) print(f'Real conversion successful: tree={fhe_tree.tree}') except ImportError: print('scikit-learn not installed, skipping real conversion test') print('from_sklearn_tree tests passed!')

from_sklearn_forest() (Helper)¶

Demonstrates the from_sklearn_forest function. This is a client-side (cleartext) helper function. It is not an FHE circuit and cannot be compiled with fhe.Compiler.

from_sklearn_forest converts a fitted sklearn RandomForestClassifier into an FHERandomForest by converting each estimator tree individually.

In [ ]:

from concrete_fhe_toolkit.ml.sklearn_bridge import from_sklearn_forest
from concrete_fhe_toolkit.ml.classes import FHERandomForest

# Test error handling for unfitted model
try:
    class UnfittedForest:
        pass
    from_sklearn_forest(UnfittedForest())
    assert False, 'Should have raised ValueError'
except ValueError as e:
    assert 'fitted sklearn forest' in str(e)
    print(f'Correctly caught unfitted model: {e}')

# If sklearn is available, perform a real conversion
try:
    from sklearn.ensemble import RandomForestClassifier
    import numpy as np
    X = np.array([[1, 2], [3, 4], [5, 6], [7, 8]])
    y = np.array([0, 1, 0, 1])
    sk_rf = RandomForestClassifier(n_estimators=3, max_depth=2, random_state=42).fit(X, y)
    fhe_rf = from_sklearn_forest(sk_rf, scale=1)
    assert isinstance(fhe_rf, FHERandomForest)
    assert len(fhe_rf.trees) == 3
    print(f'Real conversion successful: {len(fhe_rf.trees)} trees converted')
except ImportError:
    print('scikit-learn not installed, skipping real conversion test')

print('from_sklearn_forest tests passed!')

from concrete_fhe_toolkit.ml.sklearn_bridge import from_sklearn_forest from concrete_fhe_toolkit.ml.classes import FHERandomForest # Test error handling for unfitted model try: class UnfittedForest: pass from_sklearn_forest(UnfittedForest()) assert False, 'Should have raised ValueError' except ValueError as e: assert 'fitted sklearn forest' in str(e) print(f'Correctly caught unfitted model: {e}') # If sklearn is available, perform a real conversion try: from sklearn.ensemble import RandomForestClassifier import numpy as np X = np.array([[1, 2], [3, 4], [5, 6], [7, 8]]) y = np.array([0, 1, 0, 1]) sk_rf = RandomForestClassifier(n_estimators=3, max_depth=2, random_state=42).fit(X, y) fhe_rf = from_sklearn_forest(sk_rf, scale=1) assert isinstance(fhe_rf, FHERandomForest) assert len(fhe_rf.trees) == 3 print(f'Real conversion successful: {len(fhe_rf.trees)} trees converted') except ImportError: print('scikit-learn not installed, skipping real conversion test') print('from_sklearn_forest tests passed!')