MAINT: Remove FutureWarnings

Remove future warnings from pandas, NumPy and internal

setup.cfg

@@ -51,6 +51,8 @@ filterwarnings =
     error:Only PeriodIndexes, DatetimeIndexes:UserWarning
     error:the 'unbiased'' keyword is deprecated:FutureWarning
     error:unbiased is deprecated in factor of adjusted:FutureWarning
+    error:categorical is deprecated:FutureWarning
+    error:After 0.13 initialization:FutureWarning
 markers =
     example: mark a test that runs example code
     matplotlib: mark a test that requires matplotlib

statsmodels/discrete/discrete_model.py

@@ -110,15 +110,16 @@
 
 # helper for MNLogit (will be generally useful later)
 def _numpy_to_dummies(endog):
-    if endog.dtype.kind in ['S', 'O']:
-        endog_dummies, ynames = tools.categorical(endog, drop=True,
-                                                  dictnames=True)
-    elif endog.ndim == 2:
+    if endog.ndim == 2 and endog.dtype.kind not in ["S", "O"]:
         endog_dummies = endog
         ynames = range(endog.shape[1])
     else:
-        endog_dummies, ynames = tools.categorical(endog, drop=True,
-                                                  dictnames=True)
+        dummies = get_dummies(endog, drop_first=False)
+        ynames = {i: dummies.columns[i] for i in range(dummies.shape[1])}
+        endog_dummies = np.asarray(dummies, dtype=float)
+
+        return endog_dummies, ynames
+
     return endog_dummies, ynames
 
 

statsmodels/imputation/tests/test_ros.py

@@ -3,6 +3,7 @@
 from io import StringIO
 from textwrap import dedent
 
+import numpy as np
 import numpy.testing as npt
 
 import numpy
@@ -379,10 +380,10 @@ def test_cohn(self):
             'ncen_equal', 'prob_exceedance'
         ]
         cohn = ros.cohn_numbers(self.df, self.rescol, self.cencol)
+        # Use round in place of the deprecated check_less_precise arg
         assert_frame_equal(
-            cohn[cols],
-            self.expected_cohn[cols],
-            check_less_precise=True,
+            np.round(cohn[cols], 3),
+            np.round(self.expected_cohn[cols], 3),
         )
 
 

statsmodels/regression/_tools.py

@@ -55,7 +55,7 @@ def __init__(self, endog, exog, weights=1.0, check_endog=False,
         if np.isscalar(weights):
             self.wexog = w_half * exog
         else:
-            self.wexog = w_half[:, None] * exog
+            self.wexog = np.asarray(w_half)[:, None] * exog
 
     def fit(self, method='pinv'):
         """

statsmodels/tools/tests/test_tools.py

@@ -17,6 +17,8 @@
 from statsmodels.tools import tools
 from statsmodels.tools.tools import pinv_extended
 
+# Ignore future warnings from test code
+pytestmark = pytest.mark.filterwarnings("ignore:categorical is deprecated:FutureWarning")
 
 @pytest.fixture(scope='module')
 def string_var():

statsmodels/tsa/ar_model.py

@@ -516,7 +516,7 @@ def _wrap_prediction(self, prediction, start, end):
         n_values = end - start
         if not isinstance(self.data.orig_endog, (pd.Series, pd.DataFrame)):
             return prediction[-n_values:]
-        index = self.data.orig_endog.index
+        index = self._index
         if end > self.endog.shape[0]:
             freq = getattr(index, 'freq', None)
             if freq:

statsmodels/tsa/deterministic.py

@@ -570,7 +570,7 @@ def _compute_ratio(
     ) -> np.ndarray:
         if isinstance(index, pd.PeriodIndex):
             index = index.to_timestamp()
-        delta = index.to_perioddelta(self._freq)
+        delta = index - index.to_period(self._freq).to_timestamp()
         pi = index.to_period(self._freq)
         gap = (pi + 1).to_timestamp() - pi.to_timestamp()
         return to_numpy(delta) / to_numpy(gap)

statsmodels/tsa/holtwinters/model.py

@@ -403,17 +403,17 @@ def fix_params(self, values):
 
     def _initialize(self):
         if self._initialization_method is None:
-            warnings.warn(
-                "After 0.13 initialization must be handled at model creation",
-                FutureWarning,
-            )
             msg = "initial_{0} given but no initialization method specified."
             if self._initial_level is not None:
                 raise ValueError(msg.format("level"))
             if self._initial_trend is not None:
                 raise ValueError(msg.format("trend"))
             if self._initial_seasonal is not None:
                 raise ValueError(msg.format("seasonal"))
+            warnings.warn(
+                "After 0.13 initialization must be handled at model creation",
+                FutureWarning,
+            )
             return
         if self._initialization_method == "known":
             return self._initialize_known()