def test_calibrationbelt_algorithm_federated(test_input, expected):
result = get_algorithm_result(CalibrationBelt, test_input, num_workers=5)
expected = expected[0]
assert int(result["n_obs"]) == int(expected["n_obs"])
assert int(result["Model degree"]) == int(expected["Model degree"])
assert np.isclose(result["p value"], expected["p value"], atol=1e-3)
def test_pca_algorithm_federated(test_input, expected):
result = get_algorithm_result(PCA, test_input, num_workers=10)
assert int(result["n_obs"]) == int(expected["n_obs"])
assert np.isclose(result["eigenvalues"], expected["eigen_vals"],
atol=1e-3).all()
for u, v in zip(result["eigenvectors"], expected["eigen_vecs"]):
assert are_collinear(u, v)
def test_pearson_algorithm_federated(test_input, expected):
result = get_algorithm_result(Pearson, test_input, num_workers=10)
assert np.isclose(
result["Pearson correlation coefficient"],
expected["Pearson correlation coefficient"],
atol=1e-3,
).all()
assert np.isclose(result["p-value"], expected["p-value"], atol=1e-3).all()
assert int(result["n_obs"]) == int(expected["n_obs"])
def test_logistic_regression_algorithm_local(test_input, expected):
result = get_algorithm_result(LogisticRegression,
test_input,
num_workers=1)
# There is no way to choose which level will be the positive and which the
# negative level in sklearn's LogisticRegression. Sometimes the choice it
# makes agrees with our choice and sometimes it doesn't. In the latter case
# the coefficients lie on the same axis but have opposite orientation,
# hence we only check if the two results are collinear.
assert are_collinear(result["Coefficients"], expected["coeff"])
def test_anova_algorithm_federated(test_input, expected):
result = get_algorithm_result(Anova, test_input, num_workers=10)
aov = result["anova_table"]
tukey = result["tukey_table"]
e_aov = {k: v for k, v in expected.items() if k != "tukey_test"}
e_tukey = expected["tukey_test"]
assert set(e_aov) == set(aov.keys())
for key, e_val in e_aov.items():
r_val = aov[key]
assert np.isclose(e_val, r_val)
for et, rt in zip(e_tukey, tukey):
for key, e_val in et.items():
r_val = rt[key]
assert e_val == r_val or np.isclose(e_val, r_val)
def test_logistic_regression_algorithm_federated(test_input, expected):
result = get_algorithm_result(LogisticRegression, test_input, 10)
assert are_collinear(result["Coefficients"], expected["coeff"])