WIP: fix regression about datetime_to_numeric

Author: fujiisoup

doc/whats-new.rst

@@ -51,6 +51,8 @@ Bug fixes
 
 - Silenced warnings that appear when using pandas 0.24.
   By `Stephan Hoyer <https://github.com/shoyer>`_
+- Bug fix for interpolation with an datetime array. (:issue:`2668`)
+  By `Keisuke Fujii <https://github.com/fujiisoup>`_.
 - Interpolating via resample now internally specifies ``bounds_error=False``
   as an argument to ``scipy.interpolate.interp1d``, allowing for interpolation
   from higher frequencies to lower frequencies.  Datapoints outside the bounds

xarray/core/dataset.py

@@ -30,9 +30,9 @@
 from .options import OPTIONS, _get_keep_attrs
 from .pycompat import dask_array_type
 from .utils import (
-    Frozen, SortedKeysDict, _check_inplace, datetime_to_numeric,
-    decode_numpy_dict_values, either_dict_or_kwargs, ensure_us_time_resolution,
-    hashable, maybe_wrap_array)
+    Frozen, SortedKeysDict, _check_inplace, decode_numpy_dict_values,
+    either_dict_or_kwargs, ensure_us_time_resolution, hashable,
+    maybe_wrap_array)
 from .variable import IndexVariable, Variable, as_variable, broadcast_variables
 from ..plot.dataset_plot import _Dataset_PlotMethods
 
@@ -3853,15 +3853,14 @@ def differentiate(self, coord, edge_order=1, datetime_unit=None):
                 datetime_unit, _ = np.datetime_data(coord_var.dtype)
             elif datetime_unit is None:
                 datetime_unit = 's'  # Default to seconds for cftime objects
-            coord_var = datetime_to_numeric(
-                coord_var, datetime_unit=datetime_unit)
+            coord_var = coord_var._to_numeric(datetime_unit=datetime_unit)
 
         variables = OrderedDict()
         for k, v in self.variables.items():
             if (k in self.data_vars and dim in v.dims and
                     k not in self.coords):
                 if _contains_datetime_like_objects(v):
-                    v = datetime_to_numeric(v, datetime_unit=datetime_unit)
+                    v = v._to_numeric(datetime_unit=datetime_unit)
                 grad = duck_array_ops.gradient(
                     v.data, coord_var, edge_order=edge_order,
                     axis=v.get_axis_num(dim))

xarray/core/duck_array_ops.py

@@ -11,7 +11,7 @@
 import numpy as np
 import pandas as pd
 
-from . import dask_array_ops, dtypes, npcompat, nputils, utils
+from . import dask_array_ops, dtypes, npcompat, nputils
 from .nputils import nanfirst, nanlast
 from .pycompat import dask_array_type
 
@@ -277,14 +277,51 @@ def f(values, axis=None, skipna=None, **kwargs):
 _mean = _create_nan_agg_method('mean')
 
 
+def datetime_to_numeric(array, offset=None, datetime_unit=None, dtype=float):
+    """Convert an array containing datetime-like data to an array of floats.
+
+    Parameters
+    ----------
+    da : array
+        Input data
+    offset: Scalar with the same type of array or None
+        If None, subtract minimum values to reduce round off error
+    datetime_unit: None or any of {'Y', 'M', 'W', 'D', 'h', 'm', 's', 'ms',
+        'us', 'ns', 'ps', 'fs', 'as'}
+    dtype: target dtype
+
+    Returns
+    -------
+    array
+    """
+    if offset is None:
+        offset = array.min()
+    array = array - offset
+
+    if not hasattr(array, 'dtype'):  # scalar is converted to 0d-array
+        array = np.array(array)
+
+    if array.dtype.kind in 'O':
+        # possibly convert object array containing datetime.timedelta
+        array = np.asarray(pd.Series(array.ravel())).reshape(array.shape)
+
+    if datetime_unit:
+        array = array / np.timedelta64(1, datetime_unit)
+
+    # convert np.NaT to np.nan
+    if array.dtype.kind in 'mM':
+        return np.where(isnull(array), np.nan, array.astype(dtype))
+    return array.astype(dtype)
+
+
 def mean(array, axis=None, skipna=None, **kwargs):
     """ inhouse mean that can handle datatime dtype """
     array = asarray(array)
-    if array.dtype.kind == 'M':
+    if array.dtype.kind in 'Mm':
         offset = min(array)
         # xarray always uses datetime[ns] for datetime
         dtype = 'timedelta64[ns]'
-        return _mean(utils.datetime_to_numeric(array, offset), axis=axis,
+        return _mean(datetime_to_numeric(array, offset), axis=axis,
                      skipna=skipna, **kwargs).astype(dtype) + offset
     else:
         return _mean(array, axis=axis, skipna=skipna, **kwargs)

xarray/core/missing.py

@@ -10,7 +10,7 @@
 from .common import _contains_datetime_like_objects
 from .computation import apply_ufunc
 from .duck_array_ops import dask_array_type
-from .utils import OrderedSet, datetime_to_numeric, is_scalar
+from .utils import OrderedSet, is_scalar
 from .variable import Variable, broadcast_variables
 
 
@@ -411,10 +411,9 @@ def _floatize_x(x, new_x):
             # We assume that the most of the bits are used to represent the
             # offset (min(x)) and the variation (x - min(x)) can be
             # represented by float.
-            xmin = x[i].min()
-            x[i] = datetime_to_numeric(x[i], offset=xmin, dtype=np.float64)
-            new_x[i] = datetime_to_numeric(
-                new_x[i], offset=xmin, dtype=np.float64)
+            xmin = x[i].values.min()
+            x[i] = x[i]._to_numeric(offset=xmin, dtype=np.float64)
+            new_x[i] = new_x[i]._to_numeric(offset=xmin, dtype=np.float64)
     return x, new_x
 
 

xarray/core/utils.py

@@ -603,40 +603,6 @@ def __len__(self):
         return len(self._data) - num_hidden
 
 
-def datetime_to_numeric(array, offset=None, datetime_unit=None, dtype=float):
-    """Convert an array containing datetime-like data to an array of floats.
-
-    Parameters
-    ----------
-    da : array
-        Input data
-    offset: Scalar with the same type of array or None
-        If None, subtract minimum values to reduce round off error
-    datetime_unit: None or any of {'Y', 'M', 'W', 'D', 'h', 'm', 's', 'ms',
-        'us', 'ns', 'ps', 'fs', 'as'}
-    dtype: target dtype
-
-    Returns
-    -------
-    array
-    """
-    from . import duck_array_ops
-
-    if offset is None:
-        offset = array.min()
-    array = array - offset
-
-    if datetime_unit:
-        array = array / np.timedelta64(1, datetime_unit)
-    # convert np.NaT to np.nan
-    if array.dtype.kind in 'mM':
-        if hasattr(array, 'isnull'):
-            return np.where(array.isnull(), np.nan, array.astype(dtype))
-        return np.where(duck_array_ops.isnull(array), np.nan,
-                        array.astype(dtype))
-    return array
-
-
 def get_temp_dimname(dims, new_dim):
     """ Get an new dimension name based on new_dim, that is not used in dims.
     If the same name exists, we add an underscore(s) in the head.

xarray/core/variable.py

@@ -1782,6 +1782,14 @@ def func(self, other):
             return self
         return func
 
+    def _to_numeric(self, offset=None, datetime_unit=None, dtype=float):
+        """ A (private) method to convert datetime array to numeric dtype
+        See duck_array_ops.datetime_to_numeric
+        """
+        numeric_array = duck_array_ops.datetime_to_numeric(
+            self.data, offset, datetime_unit, dtype)
+        return type(self)(self.dims, numeric_array, self._attrs)
+
 
 ops.inject_all_ops_and_reduce_methods(Variable)
 

xarray/tests/test_dataset.py

@@ -4671,7 +4671,7 @@ def test_differentiate_datetime(dask):
     actual = da.differentiate('x', edge_order=1, datetime_unit='D')
     expected_x = xr.DataArray(
         npcompat.gradient(
-            da, utils.datetime_to_numeric(da['x'], datetime_unit='D'),
+            da, da['x'].variable._to_numeric(datetime_unit='D'),
             axis=0, edge_order=1), dims=da.dims, coords=da.coords)
     assert_equal(expected_x, actual)
 
@@ -4705,7 +4705,7 @@ def test_differentiate_cftime(dask):
 
     actual = da.differentiate('time', edge_order=1, datetime_unit='D')
     expected_data = npcompat.gradient(
-        da, utils.datetime_to_numeric(da['time'], datetime_unit='D'),
+        da, da['time'].variable._to_numeric(datetime_unit='D'),
         axis=0, edge_order=1)
     expected = xr.DataArray(expected_data, coords=da.coords, dims=da.dims)
     assert_equal(expected, actual)

xarray/tests/test_duck_array_ops.py

@@ -7,16 +7,17 @@
 import pytest
 from numpy import array, nan
 
-from xarray import DataArray, Dataset, concat
+from xarray import DataArray, Dataset, concat, cftime_range
 from xarray.core import dtypes, duck_array_ops
 from xarray.core.duck_array_ops import (
     array_notnull_equiv, concatenate, count, first, gradient, last, mean,
     rolling_window, stack, where)
 from xarray.core.pycompat import dask_array_type
-from xarray.testing import assert_allclose, assert_equal
+from xarray.testing import assert_allclose, assert_equal, assert_identical
 
 from . import (
-    assert_array_equal, has_dask, has_np113, raises_regex, requires_dask)
+    assert_array_equal, has_dask, has_np113, raises_regex, requires_cftime,
+    requires_dask)
 
 
 class TestOps(object):
@@ -569,3 +570,42 @@ def test_docs():
             indicated dimension(s) removed.
         """)
     assert actual == expected
+
+
+def test_datetime_to_numeric_datetime64():
+    times = pd.date_range('2000', periods=5, freq='7D').values
+    result = duck_array_ops.datetime_to_numeric(times, datetime_unit='h')
+    expected = 24 * np.arange(0, 35, 7)
+    np.testing.assert_array_equal(result, expected)
+
+    offset = times[1]
+    result = duck_array_ops.datetime_to_numeric(
+        times, offset=offset, datetime_unit='h')
+    expected = 24 * np.arange(-7, 28, 7)
+    np.testing.assert_array_equal(result, expected)
+
+    dtype = np.float32
+    result = duck_array_ops.datetime_to_numeric(
+        times, datetime_unit='h', dtype=dtype)
+    expected = 24 * np.arange(0, 35, 7).astype(dtype)
+    np.testing.assert_array_equal(result, expected)
+
+
+@requires_cftime
+def test_datetime_to_numeric_cftime():
+    times = cftime_range('2000', periods=5, freq='7D').values
+    result = duck_array_ops.datetime_to_numeric(times, datetime_unit='h')
+    expected = 24 * np.arange(0, 35, 7)
+    np.testing.assert_array_equal(result, expected)
+
+    offset = times[1]
+    result = duck_array_ops.datetime_to_numeric(
+        times, offset=offset, datetime_unit='h')
+    expected = 24 * np.arange(-7, 28, 7)
+    np.testing.assert_array_equal(result, expected)
+
+    dtype = np.float32
+    result = duck_array_ops.datetime_to_numeric(
+        times, datetime_unit='h', dtype=dtype)
+    expected = 24 * np.arange(0, 35, 7).astype(dtype)
+    np.testing.assert_array_equal(result, expected)

xarray/tests/test_interp.py

@@ -571,3 +571,17 @@ def test_cftime_to_non_cftime_error():
 
     with pytest.raises(TypeError):
         da.interp(time=0.5)
+
+
+@requires_cftime
+@requires_scipy
+def test_datetime_interp_noerror():
+    # GH:2667
+    a = xr.DataArray(
+        np.arange(21).reshape(3, 7), dims=['x', 'time'],
+        coords={'x': [1, 2, 3],
+                'time': pd.date_range('01-01-2001', periods=7, freq='D')})
+    xi = xr.DataArray(
+        np.linspace(1, 3, 50), dims=['time'],
+        coords={'time': pd.date_range('01-01-2001', periods=50, freq='H')})
+    a.interp(x=xi, time=xi.time)  # should no error

xarray/tests/test_utils.py

@@ -279,42 +279,3 @@ def test_either_dict_or_kwargs():
 
     with pytest.raises(ValueError, match=r'foo'):
         result = either_dict_or_kwargs(dict(a=1), dict(a=1), 'foo')
-
-
-def test_datetime_to_numeric_datetime64():
-    times = pd.date_range('2000', periods=5, freq='7D')
-    da = xr.DataArray(times, coords=[times], dims=['time'])
-    result = utils.datetime_to_numeric(da, datetime_unit='h')
-    expected = 24 * xr.DataArray(np.arange(0, 35, 7), coords=da.coords)
-    assert_identical(result, expected)
-
-    offset = da.isel(time=1)
-    result = utils.datetime_to_numeric(da, offset=offset, datetime_unit='h')
-    expected = 24 * xr.DataArray(np.arange(-7, 28, 7), coords=da.coords)
-    assert_identical(result, expected)
-
-    dtype = np.float32
-    result = utils.datetime_to_numeric(da, datetime_unit='h', dtype=dtype)
-    expected = 24 * xr.DataArray(
-        np.arange(0, 35, 7), coords=da.coords).astype(dtype)
-    assert_identical(result, expected)
-
-
-@requires_cftime
-def test_datetime_to_numeric_cftime():
-    times = xr.cftime_range('2000', periods=5, freq='7D')
-    da = xr.DataArray(times, coords=[times], dims=['time'])
-    result = utils.datetime_to_numeric(da, datetime_unit='h')
-    expected = 24 * xr.DataArray(np.arange(0, 35, 7), coords=da.coords)
-    assert_identical(result, expected)
-
-    offset = da.isel(time=1)
-    result = utils.datetime_to_numeric(da, offset=offset, datetime_unit='h')
-    expected = 24 * xr.DataArray(np.arange(-7, 28, 7), coords=da.coords)
-    assert_identical(result, expected)
-
-    dtype = np.float32
-    result = utils.datetime_to_numeric(da, datetime_unit='h', dtype=dtype)
-    expected = 24 * xr.DataArray(
-        np.arange(0, 35, 7), coords=da.coords).astype(dtype)
-    assert_identical(result, expected)