fix: pandas 3.x compat and CI updates (#159)

* fix: pandas compat in impute.py, update CI runners and Python versions

- Fix chained assignment in impute.py _fit_encode/_encode to use .loc
  indexing instead of X[col].loc[] which breaks pandas Copy-on-Write
- Convert string columns to object dtype before processing to avoid
  StringDtype incompatibility with newer pandas
- Replace deprecated macos-13 runner with macos-14
- Promote Python 3.13 from experimental to stable in CI matrix
- Add Python 3.14 as experimental across all platforms

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: use iloc for positional indexing in process_test.py

Replace `F(len)(A)[0]` with `F(len)(A).iloc[0]` to fix KeyError on
pandas 3.x where string-indexed Series no longer accept integer keys.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: use tolist() instead of .values for StringArray compat

In pandas 3.x, Index.values returns StringArray instead of numpy array.
Using .tolist() ensures str() gives the expected list format `['A']`
instead of `<StringArray>` in the duplicate keys error message.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: replace deprecated applymap with map for pandas 3.x

DataFrame.applymap was removed in pandas 3.0. Use DataFrame.map instead.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: Secbone

.github/workflows/linux.yml

@@ -11,10 +11,10 @@ jobs:
   test:
     strategy:
       matrix:
-        python-version: ['3.9', '3.10', '3.11', '3.12']
+        python-version: ['3.9', '3.10', '3.11', '3.12', '3.13']
         experimental: [false]
         include:
-          - python-version: '3.13'
+          - python-version: '3.14'
             experimental: true
       fail-fast: false
     runs-on: ubuntu-latest

.github/workflows/macos.yml

@@ -12,7 +12,7 @@ jobs:
     strategy:
       matrix:
         python-version: ['3.9', '3.10', '3.11', '3.12', '3.13']
-        macos-version: ['macos-13', 'macos-latest']  # macos-13: Intel, macos-latest: Apple Silicon
+        macos-version: ['macos-14', 'macos-latest']  # macos-14: Intel(x86_64), macos-latest: Apple Silicon
         include:
           - experimental: false
           # macos-latest (Apple Silicon) is experimental
@@ -21,8 +21,12 @@ jobs:
           # Python 3.9 is experimental
           - python-version: '3.9'
             experimental: true
-          # Python 3.13 is experimental
-          - python-version: '3.13'
+          # Python 3.14 is experimental
+          - python-version: '3.14'
+            macos-version: 'macos-14'
+            experimental: true
+          - python-version: '3.14'
+            macos-version: 'macos-latest'
             experimental: true
       fail-fast: false
     runs-on: ${{ matrix.macos-version }}

.github/workflows/windows.yml

@@ -11,10 +11,10 @@ jobs:
   test:
     strategy:
       matrix:
-        python-version: ['3.9', '3.10', '3.11', '3.12']
+        python-version: ['3.9', '3.10', '3.11', '3.12', '3.13']
         experimental: [false]
         include:
-          - python-version: '3.13'
+          - python-version: '3.14'
             experimental: true
       fail-fast: false
     runs-on: windows-latest

toad/impute.py

@@ -51,6 +51,10 @@ def transform(self, X, **kwargs):
 
 
     def _replace_empty(self, X):
+        X = X.copy()
+        # Convert string columns to object dtype to avoid StringDtype issues in newer pandas
+        for col in X.select_dtypes(include=['string']).columns:
+            X[col] = X[col].astype(object)
         mask = X.isin(self.missing_values_list)
         X = X.where(~mask, np.nan)
         return X, mask
@@ -63,12 +67,14 @@ def _fit_encode(self, X, mask):
             mask (Mask): empty mask for X
         """
         category_data = X.select_dtypes(exclude = np.number).columns
-        
+
         for col in category_data:
-            unique, X[col].loc[~mask[col]] = np.unique(X[col][~mask[col]], return_inverse = True)
+            valid = ~mask[col]
+            unique, inverse = np.unique(X.loc[valid, col], return_inverse = True)
+            X.loc[valid, col] = inverse.astype(float)
 
             self.encoder_dict[col] = unique
-        
+
         return X
 
     def _encode(self, X, mask):
@@ -79,9 +85,10 @@ def _encode(self, X, mask):
             mask (Mask): empty mask for X
         """
         for col, unique in self.encoder_dict.items():
+            valid = ~mask[col]
             table = dict(zip(unique, np.arange(len(unique))))
-            X[col].loc[~mask[col]] = np.array([table[v] for v in X[col][~mask[col]]])
-        
+            X.loc[valid, col] = np.array([table[v] for v in X.loc[valid, col]]).astype(float)
+
         return X
 
     def _decode(self, X):

toad/preprocessing/process_test.py

@@ -41,7 +41,7 @@ def test_mask_isna():
     assert m.replay(df).sum() == df['A'].isna().sum()
 
 def test_f():
-    assert F(len)(A)[0] == 500
+    assert F(len)(A).iloc[0] == 500
 
 def test_processing():
     res = (

toad/transform.py

@@ -100,7 +100,7 @@ def _raise_unfitted(self):
 
     def _check_duplicated_keys(self, X):
         if isinstance(X, pd.DataFrame) and X.columns.has_duplicates:
-            keys = X.columns[X.columns.duplicated()].values
+            keys = X.columns[X.columns.duplicated()].tolist()
             raise Exception("X has duplicate keys `{keys}`".format(keys = str(keys)))
 
         return True

toad/utils/func_test.py

@@ -103,7 +103,7 @@ def test_bin_to_number_for_frame():
         },
     ])
 
-    res = df.applymap(bin_to_number())
+    res = df.map(bin_to_number())
     assert res.loc[1, 'area_2'] == 225
 
 def test_generate_target():