Nouveau test

Author: Gsaes

tests/benchmark/test_cross_validation.py

@@ -0,0 +1,77 @@
+import numpy as np
+import pandas as pd
+import pytest
+import skopt
+
+from qolmat.benchmark import cross_validation
+from qolmat.imputations.imputers import ImputerRPCA
+from qolmat.benchmark.missing_patterns import EmpiricalHoleGenerator
+from qolmat.benchmark.utils import get_search_space
+
+df_origin = pd.DataFrame({"col1": [0, np.nan, 2, 4, np.nan], "col2": [-1, np.nan, 0.5, 1, 1.5]})
+df_imputed = pd.DataFrame({"col1": [0, 1, 2, 3.5, 4], "col2": [-1.5, 0, 1.5, 2, 1.5]})
+df_mask = pd.DataFrame(
+    {"col1": [False, False, True, True, False], "col2": [True, False, True, True, False]}
+)
+
+df_corrupted = df_origin.copy()
+df_corrupted[df_mask] = np.nan
+
+imputer_rpca = ImputerRPCA(max_iter=100, tau=2)
+generator_holes = EmpiricalHoleGenerator(n_splits=1, ratio_masked=0.5)
+search_params = {"rpca": {"lam": {"min": 0.1, "max": 1, "type": "Real"}}}
+list_spaces = get_search_space(search_params.get("rpca", {}))
+cv = cross_validation.CrossValidation(
+    imputer=imputer_rpca, list_spaces=list_spaces, hole_generator=generator_holes
+)
+
+
+@pytest.mark.parametrize("df1", [df_origin])
+@pytest.mark.parametrize("df2", [df_imputed])
+@pytest.mark.parametrize("df_mask", [df_mask])
+def test_benchmark_cross_validation_loss_function(
+    df1: pd.DataFrame, df2: pd.DataFrame, df_mask: pd.DataFrame
+) -> None:
+
+    cv.loss_norm = 3
+    np.testing.assert_raises(ValueError, cv.loss_function, df1, df2, df_mask)
+    cv.loss_norm = 2
+    result_cv2 = cv.loss_function(df_origin=df1, df_imputed=df2, df_mask=df_mask)
+    np.testing.assert_allclose(result_cv2, 1.58113, atol=1e-5)
+    cv.loss_norm = 1
+    result_cv1 = cv.loss_function(df_origin=df1, df_imputed=df2, df_mask=df_mask)
+    np.testing.assert_allclose(result_cv1, 3, atol=1e-5)
+
+
+@pytest.mark.parametrize("df", [df_corrupted])
+def test_benchmark_cross_validation_deflat_hyperparams(df: pd.DataFrame) -> None:
+    res = skopt.gp_minimize(
+        cv.objective(df),
+        dimensions=cv.list_spaces,
+        n_calls=cv.n_calls,
+        n_initial_points=max(5, cv.n_calls // 5),
+        random_state=42,
+        n_jobs=cv.n_jobs,
+    )
+    hyperparams_flat = {space.name: val for space, val in zip(cv.list_spaces, res["x"])}
+    result_hyperparams = cv.deflat_hyperparams(hyperparams_flat)
+    result = result_hyperparams["lam"]
+    np.testing.assert_allclose(result, 0.816888, atol=1e-5)
+
+
+@pytest.mark.parametrize("df", [df_corrupted])
+@pytest.mark.parametrize("return_hyper_params", [True, False])
+def test_benchmark_cross_validation_fit_transform(
+    df: pd.DataFrame, return_hyper_params: bool
+) -> None:
+
+    if return_hyper_params:
+        result_cv, result_hyp = cv.fit_transform(
+            df_corrupted, return_hyper_params=return_hyper_params
+        )
+        np.testing.assert_allclose(result_hyp["lam"], 0.816888, atol=1e-5)
+    else:
+        result_cv = cv.fit_transform(df_corrupted, return_hyper_params=return_hyper_params)
+    result = np.array(result_cv)
+    result_expected = np.array([[0, 0], [0, 0], [0, 0], [0, 0], [0, 1.5]])
+    np.testing.assert_allclose(result, result_expected, atol=1e-5)