Maint: test coverage

Author: MarieSacksick

codecov.yml

@@ -25,3 +25,4 @@ coverage:
 
 ignore:
   - "**/setup.py"
+  - "skrub/skrub/_sklearn_compat.py" # already tested in https://github.com/sklearn-compat/sklearn-compat

skrub/_gap_encoder.py

@@ -338,7 +338,10 @@ def _init_w(self, V, X):
                 W2 = W2 + 0.1
                 W = np.concatenate((W, W2), axis=0)
         else:
-            raise ValueError(f"Initialization method {self.init!r} does not exist. ")
+            raise ValueError(
+                f"Initialization method {self.init!r} does not exist. It should be one"
+                " of ['k-means++', 'random', 'k-means']."
+            )
         W /= W.sum(axis=1, keepdims=True)
         A = np.ones((self.n_components, self.n_vocab)) * 1e-10
         B = A.copy()

skrub/conftest.py

@@ -236,3 +236,8 @@ def use_fit_transform(request):
     manually.
     """
     return request.param
+
+
+xfail_with_download_error = pytest.mark.xfail(
+    raises=OSError, match="Can't download the file '.*' from urls.*"
+)

skrub/datasets/tests/test_fetching.py

@@ -6,13 +6,15 @@
 from pandas.testing import assert_frame_equal, assert_series_equal
 
 import skrub.datasets
+from skrub.conftest import xfail_with_download_error
 from skrub.datasets import _fetching, _utils
 
 
 def _get_table_names_from_bunch(bunch):
     return [k for k in bunch if isinstance(bunch[k], pd.DataFrame)]
 
 
+@xfail_with_download_error
 @pytest.mark.parametrize("dataset_name", ["employee_salaries", "drug_directory"])
 def test_fetching(monkeypatch, dataset_name):
     with TemporaryDirectory() as temp_dir:
@@ -39,6 +41,7 @@ def _error_on_get(*args, **kwargs):
     assert bunch["metadata"] == local_bunch["metadata"]
 
 
+@xfail_with_download_error
 def test_fetch_credit_fraud():
     data = skrub.datasets.fetch_credit_fraud()
     assert data.baskets.shape == (61241, 2)
@@ -52,6 +55,7 @@ def test_fetch_credit_fraud():
         skrub.datasets.fetch_credit_fraud(split=None)
 
 
+@xfail_with_download_error
 def test_fetch_employee_salaries():
     data = skrub.datasets.fetch_employee_salaries()
     assert data.employee_salaries.shape == (9228, 9)
@@ -65,6 +69,48 @@ def test_fetch_employee_salaries():
         skrub.datasets.fetch_employee_salaries(split=None)
 
 
+@xfail_with_download_error
+@pytest.mark.parametrize(
+    "dataset_name, shape",
+    [
+        ("medical_charge", (163065, 12)),
+        ("midwest_survey", (2494, 29)),
+        ("open_payments", (73558, 6)),
+        ("traffic_violations", (1578154, 43)),
+        ("toxicity", (1000, 2)),
+        ("videogame_sales", (16572, 11)),
+        ("bike_sharing", (17379, 11)),
+    ],
+)
+def test_datasets_without_splitting(dataset_name, shape):
+    "Test datasets that do not have a split argument in their fetching function."
+    data = getattr(_fetching, f"fetch_{dataset_name}")()
+    assert data[dataset_name].shape == shape
+
+
+@xfail_with_download_error
+@pytest.mark.parametrize(
+    "dataset_name, keys",
+    [
+        ("flight_delays", ["flights", "airports", "weather", "stations", "metadata"]),
+        (
+            "country_happiness",
+            [
+                "happiness_report",
+                "happiness_report",
+                "life_expectancy",
+                "legal_rights_index",
+            ],
+        ),
+        ("movielens", ["movies", "ratings", "metadata"]),
+    ],
+)
+def test_fetching_several_tables(dataset_name, keys):
+    "Test fetching functions that return several tables."
+    data = getattr(_fetching, f"fetch_{dataset_name}")()
+    assert all(key in data.keys() for key in keys)
+
+
 def test_fetching_wrong_checksum(monkeypatch):
     dataset_info = _utils.DATASET_INFO["employee_salaries"]
     monkeypatch.setitem(dataset_info, "sha256", "bad_checksum")

skrub/tests/test_gap_encoder.py

@@ -306,3 +306,27 @@ def test_empty_column_name(df_module):
     s = df_module.make_column("", ["one", "two"] * 10)
     out = GapEncoder(n_components=3, random_state=0).fit_transform(s)
     assert sbd.column_names(out) == ["one, two", "two, one", "one, two (2)"]
+
+
+def test_non_supported_analyzer(generate_data):
+    n_samples = 70
+    X = generate_data(n_samples, random_state=0)
+    gap_encoder = GapEncoder(analyzer="unsupported")
+    with pytest.raises(
+        ValueError, match=r"analyzer should be one of \['word', 'char', 'char_wb']\."
+    ):
+        gap_encoder.fit(X)
+
+
+def test_non_supported_init(generate_data):
+    n_samples = 70
+    X = generate_data(n_samples, random_state=0)
+    gap_encoder = GapEncoder(init="unsupported")
+    with pytest.raises(
+        ValueError,
+        match=(
+            r"Initialization method 'unsupported' does not exist\. It should be one of"
+            r" \['k-means\++', 'random', 'k-means']\."
+        ),
+    ):
+        gap_encoder.fit(X)

skrub/tests/test_minhash_encoder.py

@@ -287,3 +287,14 @@ def test_zero_padding_in_feature_names_out(df_module, n_components, expected_col
     feature_names = encoder.get_feature_names_out()
 
     assert feature_names[: len(expected_columns)] == expected_columns
+
+
+def test_col_not_categorical(df_module):
+    X1 = df_module.make_column("some col", [1, 2, 3, 4, 5, 6, 7, 8])
+    X2 = df_module.make_column("col", ["a", "b", "c", "d", "e", "f", "g", "h"])
+    encoder = MinHashEncoder()
+    with pytest.raises(ValueError, match="does not contain strings"):
+        encoder.fit_transform(X1)
+
+    with pytest.raises(ValueError, match="does not contain strings"):
+        encoder.fit(X2).transform(X1)

skrub/tests/test_to_datetime.py

@@ -234,3 +234,16 @@ def test_error_dispatch(func):
     # Make codecov happy
     with pytest.raises(TypeError, match="Expecting a Pandas or Polars Series"):
         func(np.array([1]))
+
+
+def test_specific_time_encoding():
+    """Test another specific time zone encoding case.
+    Not using df_module fixture because it's not intended to test polars.
+    """
+    from zoneinfo import ZoneInfo
+
+    col = [
+        pd.Timestamp(1584226800, unit="s", tz=ZoneInfo("Europe/Paris")),
+        pd.Timestamp(1584226801, unit="s", tz=ZoneInfo("Europe/Paris")),
+    ]
+    assert _get_time_zone(pd.Series(name="dt", data=col)) == "Europe/Paris"