Merge pull request #341 from lvgig/feature/update_weight_mixin

added separate tests for WeightColumnMixin

tests/base_tests.py

@@ -633,20 +633,6 @@ def test_bad_values_in_weights_error(
         with pytest.raises(ValueError, match=expected_message):
             transformer.fit(df, df["a"])
 
-    def get_df_error_combos():
-        return [
-            (
-                pd.DataFrame({"a": [1, 2], "b": [3, 4]}),
-                r"weight col \(c\) is not present in columns of data",
-                "c",
-            ),
-            (
-                pd.DataFrame({"a": [1, 2], "b": ["a", "b"]}),
-                r"weight column must be numeric.",
-                "b",
-            ),
-        ]
-
     @pytest.mark.parametrize(
         "minimal_dataframe_lookup",
         ["pandas", "polars"],

tests/mixins/test_WeightColumnMixin.py

@@ -0,0 +1,150 @@
+import narwhals as nw
+import numpy as np
+import pytest
+
+from tests.test_data import create_df_2
+from tubular.mixins import WeightColumnMixin
+
+
+class TestCheckAndSetWeight:
+    @pytest.mark.parametrize("weights_column", (0, ["a"], {"a": 10}))
+    def test_weight_arg_errors(
+        self,
+        weights_column,
+    ):
+        """Test that appropriate errors are throw for bad weight arg."""
+
+        obj = WeightColumnMixin()
+
+        with pytest.raises(
+            TypeError,
+            match="weights_column should be str or None",
+        ):
+            obj.check_and_set_weight(weights_column)
+
+
+class TestCheckWeightsColumn:
+    @pytest.mark.parametrize(
+        "library",
+        [
+            "pandas",
+            "polars",
+        ],
+    )
+    @pytest.mark.parametrize(
+        "bad_weight_value, expected_message",
+        [
+            (None, "weight column must be non-null"),
+            (np.inf, "weight column must not contain infinite values."),
+            (-np.inf, "weight column must be positive"),
+            (-1, "weight column must be positive"),
+        ],
+    )
+    def test_bad_values_in_weights_error(
+        self,
+        bad_weight_value,
+        expected_message,
+        library,
+    ):
+        """Test that an exception is raised if there are negative/nan/inf values in sample_weight."""
+
+        df = create_df_2(library=library)
+
+        obj = WeightColumnMixin()
+
+        df = nw.from_native(df)
+        native_namespace = nw.get_native_namespace(df)
+
+        weight_column = "weight_column"
+
+        df = df.with_columns(
+            nw.new_series(
+                weight_column,
+                [*[bad_weight_value], *np.arange(2, len(df) + 1)],
+                native_namespace=native_namespace,
+            ),
+        )
+
+        df = nw.to_native(df)
+
+        with pytest.raises(ValueError, match=expected_message):
+            obj.check_weights_column(df, weight_column)
+
+    @pytest.mark.parametrize(
+        "library",
+        ["pandas", "polars"],
+    )
+    def test_weight_col_non_numeric(
+        self,
+        library,
+    ):
+        """Test an error is raised if weight is not numeric."""
+
+        obj = WeightColumnMixin()
+
+        df = create_df_2(library=library)
+        df = nw.from_native(df)
+
+        weight_column = "weight_column"
+        error = r"weight column must be numeric."
+        df = df.with_columns(nw.lit("a").alias(weight_column))
+        df = nw.to_native(df)
+
+        with pytest.raises(
+            ValueError,
+            match=error,
+        ):
+            # using check_weights_column method to test correct error is raised for transformers that use weights
+
+            obj.check_weights_column(df, weight_column)
+
+    @pytest.mark.parametrize(
+        "library",
+        ["pandas", "polars"],
+    )
+    def test_weight_not_in_X_error(
+        self,
+        library,
+    ):
+        """Test an error is raised if weight is not in X"""
+
+        obj = WeightColumnMixin()
+
+        df = create_df_2(library=library)
+
+        weight_column = "weight_column"
+        error = rf"weight col \({weight_column}\) is not present in columns of data"
+
+        with pytest.raises(
+            ValueError,
+            match=error,
+        ):
+            # using check_weights_column method to test correct error is raised for transformers that use weights
+
+            obj.check_weights_column(df, weight_column)
+
+    @pytest.mark.parametrize(
+        "library",
+        ["pandas", "polars"],
+    )
+    def test_zero_total_weight_error(
+        self,
+        library,
+    ):
+        """Test that an exception is raised if the total sample weights are 0."""
+
+        obj = WeightColumnMixin()
+
+        weight_column = "weight_column"
+
+        df = create_df_2(library=library)
+
+        df = nw.from_native(df)
+        df = df.with_columns(nw.lit(0).alias(weight_column))
+        df = nw.to_native(df)
+
+        with pytest.raises(
+            ValueError,
+            match="total sample weights are not greater than 0",
+        ):
+            obj.check_weights_column(df, weight_column)

tubular/mixins.py

@@ -138,14 +138,11 @@ class WeightColumnMixin:
     """
     Mixin class with weights functionality
 
-    Attributes
-    ----------
-
-    polars_compatible : bool
-        class attribute, indicates whether transformer has been converted to polars/pandas agnostic narwhals framework
     """
 
-    polars_compatible = True
+    def classname(self) -> str:
+        """Method that returns the name of the current class when called."""
+        return type(self).__name__
 
     @nw.narwhalify
     def check_weights_column(self, X: FrameT, weights_column: str) -> None:
@@ -177,7 +174,7 @@ def check_weights_column(self, X: FrameT, weights_column: str) -> None:
             msg = f"{self.classname()}: weight column must be non-null"
             raise ValueError(msg)
 
-        # check weight not inf, not currently a narwhals efficient way to do this
+        # check weight not inf, currently no polars-y way to do this in narwhals
         if np.isinf(X[weights_column].to_numpy()).any():
             msg = f"{self.classname()}: weight column must not contain infinite values."
             raise ValueError(msg)