Merge branch 'add-basic-zerodim-array-support' of https://github.com/immerrr/pandas into immerrr-add-basic-zerodim-array-support

Author: jreback

doc/source/whatsnew/v0.15.1.txt

@@ -187,6 +187,7 @@ Bug Fixes
 - Bug in selecting from a ``Categorical`` with ``.iloc`` (:issue:`8623`)
 - Bug in groupby-transform with a Categorical (:issue:`8623`)
 - Bug in duplicated/drop_duplicates with a Categorical (:issue:`8623`)
+- Bug in ``Categorical`` reflected comparison operator raising if the first argument was a numpy array scalar (e.g. np.int64) (:issue:`8658`)
 
 
 

pandas/core/categorical.py

@@ -42,7 +42,16 @@ def f(self, other):
                 # In other series, the leads to False, so do that here too
                 ret[na_mask] = False
             return ret
-        elif lib.isscalar(other):
+
+        # Numpy-1.9 and earlier may convert a scalar to a zerodim array during
+        # comparison operation when second arg has higher priority, e.g.
+        #
+        #     cat[0] < cat
+        #
+        # With cat[0], for example, being ``np.int64(1)`` by the time it gets
+        # into this function would become ``np.array(1)``.
+        other = lib.item_from_zerodim(other)
+        if lib.isscalar(other):
             if other in self.categories:
                 i = self.categories.get_loc(other)
                 return getattr(self._codes, op)(i)

pandas/core/common.py

@@ -58,10 +58,7 @@ class AmbiguousIndexError(PandasError, KeyError):
 def create_pandas_abc_type(name, attr, comp):
     @classmethod
     def _check(cls, inst):
-        result = getattr(inst, attr, None)
-        if result is None:
-            return False
-        return result in comp
+        return getattr(inst, attr, '_typ') in comp
     dct = dict(__instancecheck__=_check,
                __subclasscheck__=_check)
     meta = type("ABCBase", (type,), dct)
@@ -84,7 +81,7 @@ def _check(cls, inst):
 ABCSparseArray = create_pandas_abc_type("ABCSparseArray", "_subtyp",
                                         ('sparse_array', 'sparse_series'))
 ABCCategorical = create_pandas_abc_type("ABCCategorical","_typ",("categorical"))
-
+ABCPeriod = create_pandas_abc_type("ABCPeriod", "_typ", ("period",))
 
 class _ABCGeneric(type):
 

pandas/lib.pyx

@@ -4,6 +4,7 @@ import numpy as np
 
 from numpy cimport *
 
+np.import_array()
 
 cdef extern from "numpy/arrayobject.h":
     cdef enum NPY_TYPES:
@@ -235,7 +236,47 @@ cpdef checknull_old(object val):
         return util._checknull(val)
 
 def isscalar(object val):
-    return np.isscalar(val) or val is None or PyDateTime_Check(val) or PyDelta_Check(val)
+    """
+    Return True if given value is scalar.
+
+    This includes:
+    - numpy array scalar (e.g. np.int64)
+    - Python builtin numerics
+    - Python builtin byte arrays and strings
+    - None
+    - instances of datetime.datetime
+    - instances of datetime.timedelta
+    - Period
+
+    """
+
+    return (np.PyArray_IsAnyScalar(val)
+            # As of numpy-1.9, PyArray_IsAnyScalar misses bytearrays on Py3.
+            or PyBytes_Check(val)
+            or val is None
+            or PyDate_Check(val)
+            or PyDelta_Check(val)
+            or PyTime_Check(val)
+            or util.is_period_object(val))
+
+
+def item_from_zerodim(object val):
+    """
+    If the value is a zerodim array, return the item it contains.
+
+    Examples
+    --------
+    >>> item_from_zerodim(1)
+    1
+    >>> item_from_zerodim('foobar')
+    'foobar'
+    >>> item_from_zerodim(np.array(1))
+    1
+    >>> item_from_zerodim(np.array([1]))
+    array([1])
+
+    """
+    return util.unbox_if_zerodim(val)
 
 
 @cython.wraparound(False)

pandas/src/inference.pyx

@@ -16,6 +16,10 @@ def is_bool(object obj):
 def is_complex(object obj):
     return util.is_complex_object(obj)
 
+def is_period(object val):
+    """ Return a boolean if this is a Period object """
+    return util.is_period_object(val)
+
 _TYPE_MAP = {
     'int8': 'integer',
     'int16': 'integer',

pandas/src/numpy_helper.h

@@ -167,6 +167,21 @@ void set_array_not_contiguous(PyArrayObject *ao) {
 }
 
 
+// If arr is zerodim array, return a proper array scalar (e.g. np.int64).
+// Otherwise, return arr as is.
+PANDAS_INLINE PyObject*
+unbox_if_zerodim(PyObject* arr) {
+    if (PyArray_IsZeroDim(arr)) {
+        PyObject *ret;
+        ret = PyArray_ToScalar(PyArray_DATA(arr), arr);
+        return ret;
+    } else {
+        Py_INCREF(arr);
+        return arr;
+    }
+}
+
+
 // PANDAS_INLINE PyObject*
 // get_base_ndarray(PyObject* ap) {
 //   // if (!ap || (NULL == ap)) {

pandas/src/util.pxd

@@ -22,6 +22,7 @@ cdef extern from "numpy_helper.h":
     inline void transfer_object_column(char *dst, char *src, size_t stride,
                                        size_t length)
     object sarr_from_data(cnp.dtype, int length, void* data)
+    inline object unbox_if_zerodim(object arr)
 
 cdef inline object get_value_at(ndarray arr, object loc):
     cdef:
@@ -64,7 +65,6 @@ cdef inline int is_contiguous(ndarray arr):
 cdef inline is_array(object o):
     return cnp.PyArray_Check(o)
 
-
 cdef inline bint _checknull(object val):
     try:
         return val is None or (cpython.PyFloat_Check(val) and val != val)
@@ -82,3 +82,6 @@ cdef inline bint _checknull_old(object val):
 
 cdef inline bint _checknan(object val):
     return not cnp.PyArray_Check(val) and val != val
+
+cdef inline bint is_period_object(object val):
+    return getattr(val,'_typ','_typ') == 'period'

pandas/tests/test_categorical.py

@@ -917,6 +917,12 @@ def test_datetime_categorical_comparison(self):
         self.assert_numpy_array_equal(dt_cat > dt_cat[0], [False, True, True])
         self.assert_numpy_array_equal(dt_cat[0] < dt_cat, [False, True, True])
 
+    def test_reflected_comparison_with_scalars(self):
+        # GH8658
+        cat = pd.Categorical([1, 2, 3])
+        self.assert_numpy_array_equal(cat > cat[0], [False, True, True])
+        self.assert_numpy_array_equal(cat[0] < cat, [False, True, True])
+
 
 class TestCategoricalAsBlock(tm.TestCase):
     _multiprocess_can_split_ = True

pandas/tests/test_lib.py

@@ -0,0 +1,72 @@
+from datetime import datetime, timedelta, date, time
+
+import numpy as np
+
+import pandas as pd
+from pandas.lib import isscalar, item_from_zerodim
+import pandas.util.testing as tm
+
+
+class TestIsscalar(tm.TestCase):
+    def test_isscalar_builtin_scalars(self):
+        self.assertTrue(isscalar(None))
+        self.assertTrue(isscalar(True))
+        self.assertTrue(isscalar(False))
+        self.assertTrue(isscalar(0.))
+        self.assertTrue(isscalar(np.nan))
+        self.assertTrue(isscalar('foobar'))
+        self.assertTrue(isscalar(b'foobar'))
+        self.assertTrue(isscalar(u'foobar'))
+        self.assertTrue(isscalar(datetime(2014, 1, 1)))
+        self.assertTrue(isscalar(date(2014, 1, 1)))
+        self.assertTrue(isscalar(time(12, 0)))
+        self.assertTrue(isscalar(timedelta(hours=1)))
+        self.assertTrue(isscalar(pd.NaT))
+
+    def test_isscalar_builtin_nonscalars(self):
+        self.assertFalse(isscalar({}))
+        self.assertFalse(isscalar([]))
+        self.assertFalse(isscalar([1]))
+        self.assertFalse(isscalar(()))
+        self.assertFalse(isscalar((1,)))
+        self.assertFalse(isscalar(slice(None)))
+        self.assertFalse(isscalar(Ellipsis))
+
+    def test_isscalar_numpy_array_scalars(self):
+        self.assertTrue(isscalar(np.int64(1)))
+        self.assertTrue(isscalar(np.float64(1.)))
+        self.assertTrue(isscalar(np.int32(1)))
+        self.assertTrue(isscalar(np.object_('foobar')))
+        self.assertTrue(isscalar(np.str_('foobar')))
+        self.assertTrue(isscalar(np.unicode_(u'foobar')))
+        self.assertTrue(isscalar(np.bytes_(b'foobar')))
+        self.assertTrue(isscalar(np.datetime64('2014-01-01')))
+        self.assertTrue(isscalar(np.timedelta64(1, 'h')))
+
+    def test_isscalar_numpy_zerodim_arrays(self):
+        for zerodim in [np.array(1),
+                        np.array('foobar'),
+                        np.array(np.datetime64('2014-01-01')),
+                        np.array(np.timedelta64(1, 'h'))]:
+            self.assertFalse(isscalar(zerodim))
+            self.assertTrue(isscalar(item_from_zerodim(zerodim)))
+
+    def test_isscalar_numpy_arrays(self):
+        self.assertFalse(isscalar(np.array([])))
+        self.assertFalse(isscalar(np.array([[]])))
+        self.assertFalse(isscalar(np.matrix('1; 2')))
+
+    def test_isscalar_pandas_scalars(self):
+        self.assertTrue(isscalar(pd.Timestamp('2014-01-01')))
+        self.assertTrue(isscalar(pd.Timedelta(hours=1)))
+        self.assertTrue(isscalar(pd.Period('2014-01-01')))
+
+    def test_isscalar_pandas_containers(self):
+        self.assertFalse(isscalar(pd.Series()))
+        self.assertFalse(isscalar(pd.Series([1])))
+        self.assertFalse(isscalar(pd.DataFrame()))
+        self.assertFalse(isscalar(pd.DataFrame([[1]])))
+        self.assertFalse(isscalar(pd.Panel()))
+        self.assertFalse(isscalar(pd.Panel([[[1]]])))
+        self.assertFalse(isscalar(pd.Index([])))
+        self.assertFalse(isscalar(pd.Index([1])))

pandas/tseries/period.py

@@ -63,6 +63,7 @@ class Period(PandasObject):
     """
     __slots__ = ['freq', 'ordinal']
     _comparables = ['name','freqstr']
+    _typ = 'period'
 
     @classmethod
     def _from_ordinal(cls, ordinal, freq):
@@ -498,7 +499,6 @@ def strftime(self, fmt):
         base, mult = _gfc(self.freq)
         return tslib.period_format(self.ordinal, base, fmt)
 
-
 def _get_ordinals(data, freq):
     f = lambda x: Period(x, freq=freq).ordinal
     if isinstance(data[0], Period):