Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

chore: avoid checking accuracy when early stop in FHE training #642

Merged
merged 1 commit into from
Apr 24, 2024
+14 −12
Merged

chore: avoid checking accuracy when early stop in FHE training #642

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
26 changes: 14 additions & 12 deletions tests/sklearn/test_fhe_training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -309,7 +309,7 @@ def test_clear_fit(
model.predict_proba(x, fhe="simulate")


# pylint: disable=too-many-arguments
# pylint: disable=too-many-arguments, too-many-branches
def check_encrypted_fit(
x,
y,
Expand All @@ -318,7 +318,7 @@ def check_encrypted_fit(
parameters_range,
max_iter,
fit_intercept,
check_accuracy,
check_accuracy=None,
fhe=None,
partial_fit=False,
warm_fit=False,
Expand Down Expand Up @@ -414,8 +414,9 @@ def check_encrypted_fit(
(1, 1)
), "When the model is fitted without bias, the bias term should only be made of zeros."

# Check that we overfit properly a linearly separable dataset
check_accuracy(y, y_pred_class)
# If relevant, check that we overfit properly a linearly separable dataset
if check_accuracy is not None:
check_accuracy(y, y_pred_class)

return weights, bias, y_pred_proba, y_pred_class, model.random_number_generator

Expand Down Expand Up @@ -446,7 +447,7 @@ def test_encrypted_fit_coherence(
parameters_range,
max_iter,
fit_intercept,
check_accuracy,
check_accuracy=check_accuracy,
fhe="disable",
)
)
Expand All @@ -460,7 +461,7 @@ def test_encrypted_fit_coherence(
parameters_range,
max_iter,
fit_intercept,
check_accuracy,
check_accuracy=check_accuracy,
fhe="simulate",
)
)
Expand All @@ -475,7 +476,8 @@ def test_encrypted_fit_coherence(
# Define early break parameters, with a very high tolerance
early_break_kwargs = {"early_stopping": True, "tol": 1e100}

# We don't have any way to detect early break
# We don't have any way to properly test early break, we therefore disable the accuracy check
# in order to avoid flaky issues
check_encrypted_fit(
x,
y,
Expand All @@ -484,7 +486,7 @@ def test_encrypted_fit_coherence(
parameters_range,
max_iter,
fit_intercept,
check_accuracy,
check_accuracy=None,
fhe="simulate",
init_kwargs=early_break_kwargs,
)
Expand All @@ -498,7 +500,7 @@ def test_encrypted_fit_coherence(
parameters_range,
max_iter,
fit_intercept,
check_accuracy,
check_accuracy=check_accuracy,
partial_fit=True,
)
)
Expand All @@ -520,7 +522,7 @@ def test_encrypted_fit_coherence(
parameters_range,
max_iter,
fit_intercept,
check_accuracy,
check_accuracy=check_accuracy,
warm_fit=True,
init_kwargs=warm_fit_init_kwargs,
)
Expand All @@ -543,7 +545,7 @@ def test_encrypted_fit_coherence(
parameters_range,
first_iterations,
fit_intercept,
check_accuracy,
check_accuracy=check_accuracy,
fhe="simulate",
)

Expand All @@ -566,7 +568,7 @@ def test_encrypted_fit_coherence(
parameters_range,
last_iterations,
fit_intercept,
check_accuracy,
check_accuracy=check_accuracy,
fhe="simulate",
random_number_generator=rng_coef_init,
fit_kwargs=coef_init_fit_kwargs,
Expand Down
Loading