Merge pull request #178 from lvgig/feature/log_issue

Change log(x+1) to log1p(x) in LogTransformer

Author: ChaitanMohr

CHANGELOG.rst

@@ -16,6 +16,17 @@ Subsections for each version can be one of the following;
 
 Each individual change should have a link to the pull request after the description of the change.
 
+1.2.2 (2024-02-20)
+------------------
+Added
+^^^^^
+- Created unit test for checking if log1p is working and well conditioned for small x `#178 <https://github.com/lvgig/tubular/pull/178>`_
+
+Changed
+^^^^^^^
+- Changed LogTransformer to use log1p(x) instead of log(x+1) `#178 <https://github.com/lvgig/tubular/pull/178>`_
+- Changed unit tests using log(x+1) to log1p(x) `#178 <https://github.com/lvgig/tubular/pull/178>`_
+
 1.2.1 (2024-02-08)
 ------------------
 Added

tests/numeric/test_LogTransformer.py

@@ -96,8 +96,8 @@ def expected_df_2():
         """Expected output of test_expected_output_2."""
         df = d.create_df_3()
 
-        df["a_new_col"] = np.log(df["a"] + 1)
-        df["b_new_col"] = np.log(df["b"] + 1)
+        df["a_new_col"] = np.log1p(df["a"])
+        df["b_new_col"] = np.log1p(df["b"])
 
         return df.drop(columns=["a", "b"])
 
@@ -114,8 +114,8 @@ def expected_df_4():
         """Expected output of test_expected_output_4."""
         df = d.create_df_3()
 
-        df["a_new_col"] = np.log(df["a"] + 1)
-        df["b_new_col"] = np.log(df["b"] + 1)
+        df["a_new_col"] = np.log1p(df["a"])
+        df["b_new_col"] = np.log1p(df["b"])
 
         return df
 
@@ -135,6 +135,28 @@ def expected_df_6():
 
         return df.drop("a", axis=1)
 
+    def test_log1p(self):
+        """Test that log1p is working as intended."""
+        df = pd.DataFrame(
+            {
+                "a": [0.00001, 0.00002, 0.00003],
+                "b": [0.00004, 0.00005, 0.00006],
+            },
+        )
+        # Values created using np.log1p() of original df
+        expected = pd.DataFrame(
+            {
+                "a_log": [9.999950e-06, 1.999980e-05, 2.999955e-05],
+                "b_log": [3.99992000e-05, 4.99987500e-05, 5.99982001e-05],
+            },
+        )
+        log_transformer = LogTransformer(
+            columns=["a", "b"],
+            add_1=True,
+        )
+        actual = log_transformer.transform(df)
+        pd.testing.assert_frame_equal(actual, expected)
+
     def test_super_transform_called(self, mocker):
         """Test that BaseTransformer.transform called."""
         df = d.create_df_3()

tubular/_version.py

@@ -1 +1 @@
-__version__ = "1.2.1"
+__version__ = "1.2.2"

tubular/numeric.py

@@ -122,10 +122,10 @@ def transform(self, X: pd.DataFrame) -> pd.DataFrame:
                 raise ValueError(msg)
 
             if self.base is None:
-                X[new_column_names] = np.log(X[self.columns] + 1)
+                X[new_column_names] = np.log1p(X[self.columns])
 
             else:
-                X[new_column_names] = np.log(X[self.columns] + 1) / np.log(self.base)
+                X[new_column_names] = np.log1p(X[self.columns]) / np.log(self.base)
 
         else:
             if (X[self.columns] <= 0).sum().sum() > 0: