BUG: Resample across multiple days

Fixes an issue where resampling over multiple days causes a ValueError
when a number of days between the normalized first and normalized last
days is not a multiple of the frequency.

Added test TestResample.test_resample_anchored_multiday

Closes #8683

Author: hkleynhans

doc/source/whatsnew/v0.15.2.txt

@@ -145,7 +145,8 @@ Bug Fixes
 
 
 
-
+- Bug in resample that causes a ValueError when resampling across multiple days
+  and the last offset is not calculated from the start of the range (:issue:`8683`)
 
 
 - Bug in `pd.infer_freq`/`DataFrame.inferred_freq` that prevented proper sub-daily frequency inference

pandas/tseries/resample.py

@@ -411,15 +411,19 @@ def _get_range_edges(first, last, offset, closed='left', base=0):
 def _adjust_dates_anchored(first, last, offset, closed='right', base=0):
     from pandas.tseries.tools import normalize_date
 
+    # First and last offsets should be calculated from the start day to fix an
+    # error cause by resampling across multiple days when a one day period is
+    # not a multiple of the frequency.
+    #
+    # See https://github.com/pydata/pandas/issues/8683
+
     start_day_nanos = Timestamp(normalize_date(first)).value
-    last_day_nanos = Timestamp(normalize_date(last)).value
 
     base_nanos = (base % offset.n) * offset.nanos // offset.n
     start_day_nanos += base_nanos
-    last_day_nanos += base_nanos
 
     foffset = (first.value - start_day_nanos) % offset.nanos
-    loffset = (last.value - last_day_nanos) % offset.nanos
+    loffset = (last.value - start_day_nanos) % offset.nanos
 
     if closed == 'right':
         if foffset > 0:

pandas/tseries/tests/test_resample.py

@@ -705,6 +705,29 @@ def test_resample_anchored_monthstart(self):
         for freq in freqs:
             result = ts.resample(freq, how='mean')
 
+    def test_resample_anchored_multiday(self):
+        # When resampling a range spanning multiple days, ensure that the
+        # start date gets used to determine the offset.  Fixes issue where
+        # a one day period is not a multiple of the frequency.
+        #
+        # See: https://github.com/pydata/pandas/issues/8683
+
+        s = pd.Series(np.random.randn(5),
+                      index=pd.date_range('2014-10-14 23:06:23.206',
+                                          periods=3, freq='400L')
+                      | pd.date_range('2014-10-15 23:00:00',
+                                      periods=2, freq='2200L'))
+
+        # Ensure left closing works
+        result = s.resample('2200L', 'mean')
+        self.assertEqual(result.index[-1],
+                         pd.Timestamp('2014-10-15 23:00:02.000'))
+
+        # Ensure right closing works
+        result = s.resample('2200L', 'mean', label='right')
+        self.assertEqual(result.index[-1],
+                         pd.Timestamp('2014-10-15 23:00:04.200'))
+
     def test_corner_cases(self):
         # miscellaneous test coverage