diff --git a/doc/source/whatsnew/v0.24.0.txt b/doc/source/whatsnew/v0.24.0.txt index 53a7396065a8d6..0c0afc892bb879 100644 --- a/doc/source/whatsnew/v0.24.0.txt +++ b/doc/source/whatsnew/v0.24.0.txt @@ -8,6 +8,54 @@ v0.24.0 New features ~~~~~~~~~~~~ +Integer NA Support +^^^^^^^^^^^^^^^^^^ + +TODO(update docs) +Pandas has gained the ability to hold integer dtypes with missing values. This long requested feature is enable thru the use of ``ExtensionTypes`` . Here is an example of usage. + +We can construct a ``Series`` with the specified dtype. The dtype string ``Int64`` is a pandas ``ExtensionDtype``. Specifying an list or array using the traditional missing value +marker of ``np.nan`` will infer to integer dtype. The display of the ``Series`` will also use the ``NaN`` to indicate missing values in string outputs. + +.. ipython:: python + + s = pd.Series([1, 2, np.nan], dtype='Int64') + s + + +Operations on these dtypes will propagate ``NaN`` as other pandas operations. + +.. ipython:: python + + # arithmetic + s + 1 + + # comparison + s == 1 + + # indexing + s.iloc[1:3] + + # operate with other dtypes + s + s.iloc[1:3] + +These dtypes can operate as part of ``DataFrames`` as well. + +.. ipython:: python + + df = pd.DataFrame({'A': s, 'B': [1, 2, 3], 'C': list('aab')}) + df + df.dtypes + + +These dtypes can be merged & reshaped & casted as well. + +.. ipython:: python + + pd.concat([df[['A']], df[['B', 'C']]], axis=1).dtypes + df['A'].astype(float) + + .. _whatsnew_0240.enhancements.other: Other Enhancements diff --git a/pandas/core/arrays/__init__.py b/pandas/core/arrays/__init__.py index f8adcf520c15ba..1c34c123e483c2 100644 --- a/pandas/core/arrays/__init__.py +++ b/pandas/core/arrays/__init__.py @@ -1,2 +1,6 @@ from .base import ExtensionArray # noqa from .categorical import Categorical # noqa +from .integer import ( # noqa + Int8Array, Int16Array, Int32Array, Int64Array, + UInt8Array, UInt16Array, UInt32Array, UInt64Array, + to_integer_array) diff --git a/pandas/core/arrays/integer.py b/pandas/core/arrays/integer.py new file mode 100644 index 00000000000000..e7c9b6c3dd5969 --- /dev/null +++ b/pandas/core/arrays/integer.py @@ -0,0 +1,489 @@ +import sys +import warnings +import numpy as np + +from pandas.core.dtypes.generic import ABCSeries, ABCIndexClass +from pandas.compat import set_function_name +from pandas.api.types import (is_integer, is_scalar, is_float, + is_float_dtype, is_integer_dtype, + is_object_dtype, + infer_dtype) +from pandas.core.arrays import ExtensionArray +from pandas.core.dtypes.base import ExtensionDtype +from pandas.core.dtypes.dtypes import registry +from pandas.core.dtypes.missing import isna, notna + +# available dtypes +_integer_dtypes = ['int8', 'int16', 'int32', 'int64'] +_integer_formatter = lambda x: x.capitalize() +_unsigned_dtypes = ['uint8', 'uint16', 'uint32', 'uint64'] +_unsigned_formatter = lambda x: "{}{}".format(x[0].upper(), x[1:].capitalize()) + + +class IntegerDtype(ExtensionDtype): + type = None + na_value = np.nan + kind = 'i' + is_integer = True + is_signed_integer = True + is_unsigned_integer = False + + @classmethod + def construct_from_string(cls, string): + if string == cls.name: + return cls() + else: + raise TypeError("Cannot construct a '{}' from " + "'{}'".format(cls, string)) + + @classmethod + def construct_from_string_strict(cls, string): + """ + Strict construction from a string, raise a TypeError if not + possible + """ + if string[0] == 'I': + return cls.construct_from_string(string) + raise TypeError("could not construct PeriodDtype") + + +class UnsignedIntegerDtype(IntegerDtype): + kind = 'u' + is_signed_integer = False + is_unsigned_integer = True + + @classmethod + def construct_from_string_strict(cls, string): + """ + Strict construction from a string, raise a TypeError if not + possible + """ + if string[0] == 'U': + return cls.construct_from_string(string) + raise TypeError("could not construct PeriodDtype") + + +def to_integer_array(values): + """ + Parameters + ---------- + values : 1D list-like + + Returns + ------- + infer and return an integer array + + Raises + ------ + TypeError if incompatible types + """ + values = np.array(values, copy=False) + kind = 'UInt' if values.dtype.kind == 'u' else 'Int' + array_type = "{}{}Array".format(kind, values.dtype.itemsize * 8) + try: + array_type = getattr(module, array_type) + except AttributeError: + raise TypeError("Incompatible dtype for {}".format(values.dtype)) + return array_type(values, copy=False) + + +def coerce_to_array(values, dtype, mask=None, copy=False): + """ + Coerce the input values array to numpy arrays with a mask + + Parameters + ---------- + values : 1D list-like + dtype : integer dtype + mask : boolean 1D array, optional + copy : boolean, default False + if True, copy the input + + Returns + ------- + tuple of (values, mask) + """ + values = np.array(values, copy=copy) + if is_object_dtype(values): + inferred_type = infer_dtype(values) + if inferred_type not in ['floating', 'integer', + 'mixed-integer', 'mixed-integer-float']: + raise TypeError("{} cannot be converted to an IntegerDtype".format( + values.dtype)) + + elif not (is_integer_dtype(values) or is_float_dtype(values)): + raise TypeError("{} cannot be converted to an IntegerDtype".format( + values.dtype)) + + if mask is None: + mask = isna(values) + else: + assert len(mask) == len(values) + + if not values.ndim == 1: + raise TypeError("values must be a 1D list-like") + if not mask.ndim == 1: + raise TypeError("mask must be a 1D list-like") + + # avoid float->int numpy conversion issues + if is_object_dtype(values): + mask |= isna(values) + + # we copy as need to coerce here + if mask.any(): + values = values.copy() + values[mask] = 1 + + values = values.astype(dtype.type) + else: + values = values.astype(dtype.type, copy=False) + + return values, mask + + +class IntegerArray(ExtensionArray): + """ + We represent an IntegerArray with 2 numpy arrays + - data: contains a numpy integer array of the appropriate dtype + - mask: a boolean array holding a mask on the data, False is missing + """ + + dtype = None + + def __init__(self, values, mask=None, copy=False): + self.data, self.mask = coerce_to_array( + values, dtype=self.dtype, mask=mask, copy=copy) + + @classmethod + def _from_sequence(cls, scalars, mask=None, copy=False): + return cls(scalars, mask=mask, copy=copy) + + @classmethod + def _from_factorized(cls, values, original): + return cls(values) + + def __getitem__(self, item): + if is_integer(item): + if self.mask[item]: + return self.dtype.na_value + return self.data[item] + return type(self)(self.data[item], mask=self.mask[item]) + + def _coerce_to_ndarray(self): + """ coerce to an ndarary, preserving my scalar types """ + + # TODO(jreback) make this better + data = self.data.astype(object) + data[self.mask] = self._na_value + return data + + def __array__(self, dtype=None): + """ + the array interface, return my values + We return an object array here to preserve our scalar values + """ + return self._coerce_to_ndarray() + + def __iter__(self): + """Iterate over elements of the array. + + """ + # This needs to be implemented so that pandas recognizes extension + # arrays as list-like. The default implementation makes successive + # calls to ``__getitem__``, which may be slower than necessary. + for i in range(len(self)): + if self.mask[i]: + yield self.dtype.na_value + else: + yield self.data[i] + + def _formatting_values(self): + # type: () -> np.ndarray + return self._coerce_to_ndarray() + + def take(self, indexer, allow_fill=False, fill_value=None): + from pandas.api.extensions import take + + # we always fill with 1 internally + # to avoid upcasting + data_fill_value = 1 if isna(fill_value) else fill_value + result = take(self.data, indexer, fill_value=data_fill_value, + allow_fill=allow_fill) + + mask = take(self.mask, indexer, fill_value=True, + allow_fill=allow_fill) + + # if we are filling + # we only fill where the indexer is null + # not existing missing values + # TODO(jreback) what if we have a non-na float as a fill value? + if allow_fill and notna(fill_value): + fill_mask = np.asarray(indexer) == -1 + result[fill_mask] = fill_value + mask = mask ^ fill_mask + + return self._from_sequence(result, mask=mask) + + def copy(self, deep=False): + if deep: + return type(self)( + self.data.copy(), mask=self.mask.copy()) + return type(self)(self) + + def __setitem__(self, key, value): + _is_scalar = is_scalar(value) + if _is_scalar: + value = [value] + value, mask = coerce_to_array(value, dtype=self.dtype) + + if _is_scalar: + value = value[0] + mask = mask[0] + + self.data[key] = value + self.mask[key] = mask + + def __len__(self): + return len(self.data) + + def __repr__(self): + + formatted = self._formatting_values() + return '{}({})'.format( + self.__class__.__name__, + formatted.tolist()) + + @property + def nbytes(self): + return self.data.nbytes + self.mask.nbytes + + def isna(self): + return self.mask + + @property + def _na_value(self): + return np.nan + + @classmethod + def _concat_same_type(cls, to_concat): + data = np.concatenate([x.data for x in to_concat]) + mask = np.concatenate([x.mask for x in to_concat]) + return cls(data, mask=mask) + + def astype(self, dtype, copy=True): + """Cast to a NumPy array with 'dtype'. + + Parameters + ---------- + dtype : str or dtype + Typecode or data-type to which the array is cast. + copy : bool, default True + Whether to copy the data, even if not necessary. If False, + a copy is made only if the old dtype does not match the + new dtype. + + Returns + ------- + array : ndarray + NumPy ndarray with 'dtype' for its dtype. + """ + data = self._coerce_to_ndarray() + return data.astype(dtype=dtype, copy=False) + + @property + def _ndarray_values(self): + # type: () -> np.ndarray + """Internal pandas method for lossy conversion to a NumPy ndarray. + + This method is not part of the pandas interface. + + The expectation is that this is cheap to compute, and is primarily + used for interacting with our indexers. + """ + return self.data + + def value_counts(self, dropna=True): + """ + Returns a Series containing counts of each category. + + Every category will have an entry, even those with a count of 0. + + Parameters + ---------- + dropna : boolean, default True + Don't include counts of NaN. + + Returns + ------- + counts : Series + + See Also + -------- + Series.value_counts + + """ + + from pandas import Index, Series + + # compute counts on the data with no nans + data = self.data[~self.mask] + value_counts = Index(data).value_counts() + + array = value_counts.values + index = value_counts.index + + # if we want nans, count the mask + if not dropna: + array = np.append(array, [self.mask.sum()]) + + # TODO(extension) + # should this be an and Index backed by the + # Array type? + index = index.astype(object).append(Index([np.nan])) + + return Series(array, index=index) + + def _values_for_argsort(self): + # type: () -> ndarray + """Return values for sorting. + + Returns + ------- + ndarray + The transformed values should maintain the ordering between values + within the array. + + See Also + -------- + ExtensionArray.argsort + """ + data = self.data.copy() + data[self.mask] = data.min() - 1 + return data + + @classmethod + def _make_comparison_op(cls, op): + def cmp_method(self, other): + + op_str = op.__name__ + mask = None + if isinstance(other, (np.ndarray, ABCIndexClass, ABCSeries)): + if other.ndim > 0 and len(self) != len(other): + raise ValueError('Lengths must match to compare') + elif isinstance(other, cls): + other, mask = other.data, other.mask + + # numpy will show a DeprecationWarning on invalid elementwise + # comparisons, this will raise in the future + with warnings.catch_warnings(record=True): + with np.errstate(all='ignore'): + result = op(self.data, np.asarray(other)) + + # nans propagate + if mask is None: + mask = self.mask + else: + mask = self.mask | mask + + result[mask] = True if op_str == 'ne' else False + return result + + name = '__{name}__'.format(name=op.__name__) + return set_function_name(cmp_method, name, cls) + + @classmethod + def _make_arithmetic_op(cls, op): + def integer_arithmetic_method(self, other): + + mask = None + if isinstance(other, (ABCSeries, ABCIndexClass)): + other = getattr(other, 'values', other) + elif isinstance(other, cls): + other, mask = other.data, other.mask + elif getattr(other, 'ndim', 0) > 1: + raise TypeError("can only perform ops with 1-d structures") + elif isinstance(other, IntegerArray): + pass + elif isinstance(other, np.ndarray): + if not other.ndim: + other = other.item() + elif other.ndim == 1: + if not (is_float_dtype(other) or is_integer_dtype(other)): + raise TypeError( + "can only perform ops with numeric values") + else: + if not (is_float(other) or is_integer(other)): + raise TypeError("can only perform ops with numeric values") + + # nans propagate + if mask is None: + mask = self.mask + else: + mask = self.mask | mask + + with np.errstate(all='ignore'): + result = op(self.data, other) + + # may need to fill infs + # and mask wraparound + if is_float_dtype(result): + mask |= (result == np.inf) | (result == -np.inf) + + return cls(result, mask=mask) + + name = '__{name}__'.format(name=op.__name__) + return set_function_name(integer_arithmetic_method, name, cls) + + +class UnsignedIntegerArray(IntegerArray): + pass + + +module = sys.modules[__name__] + + +# create the Dtype +types = [(_integer_dtypes, IntegerDtype, _integer_formatter), + (_unsigned_dtypes, UnsignedIntegerDtype, _unsigned_formatter)] +for dtypes, superclass, formatter in types: + + for dtype in dtypes: + + name = formatter(dtype) + classname = "{}Dtype".format(name) + attributes_dict = {'type': getattr(np, dtype), + 'name': name} + dtype_type = type(classname, (superclass, ), attributes_dict) + setattr(module, classname, dtype_type) + + # register + registry.register(dtype_type, dtype_type.construct_from_string_strict) + + +# create the Array +types = [(_integer_dtypes, IntegerArray, _integer_formatter), + (_unsigned_dtypes, UnsignedIntegerArray, _unsigned_formatter)] +for dtypes, superclass, formatter in types: + + for dtype in dtypes: + + dtype_type = getattr(module, "{}Dtype".format(formatter(dtype))) + classname = "{}Array".format(formatter(dtype)) + attributes_dict = {'dtype': dtype_type()} + array_type = type(classname, (superclass, ), attributes_dict) + setattr(module, classname, array_type) + + # add ops + array_type._add_numeric_methods_binary() + array_type._add_comparison_methods_binary() + + # set the Array type on the Dtype + dtype_type.array_type = array_type + + +def make_data(): + return (list(range(8)) + + [np.nan] + + list(range(10, 98)) + + [np.nan] + + [99, 100]) diff --git a/pandas/tests/extension/integer/__init__.py b/pandas/tests/extension/integer/__init__.py new file mode 100644 index 00000000000000..e69de29bb2d1d6 diff --git a/pandas/tests/extension/integer/test_integer.py b/pandas/tests/extension/integer/test_integer.py new file mode 100644 index 00000000000000..ecbfd74c12c235 --- /dev/null +++ b/pandas/tests/extension/integer/test_integer.py @@ -0,0 +1,411 @@ +import numpy as np +import pandas as pd +import pandas.util.testing as tm +import pytest + +from pandas.tests.extension import base +from pandas.api.types import is_integer + +from pandas.core.arrays import ( + to_integer_array, + Int8Array, Int16Array, Int32Array, Int64Array, + UInt8Array, UInt16Array, UInt32Array, UInt64Array) +from pandas.core.arrays.integer import ( + Int8Dtype, Int16Dtype, Int32Dtype, Int64Dtype, + UInt8Dtype, UInt16Dtype, UInt32Dtype, UInt64Dtype, + IntegerArray, make_data) + + +@pytest.fixture(params=[Int8Dtype, Int16Dtype, Int32Dtype, Int64Dtype, + UInt8Dtype, UInt16Dtype, UInt32Dtype, UInt64Dtype]) +def dtype(request): + return request.param() + + +@pytest.fixture(params=[Int8Array, Int16Array, Int32Array, Int64Array, + UInt8Array, UInt16Array, UInt32Array, UInt64Array]) +def arrays(request): + return request.param + + +@pytest.fixture +def data(arrays): + return arrays(make_data()) + + +@pytest.fixture +def data_missing(arrays): + return arrays([np.nan, 1]) + + +@pytest.fixture +def data_for_sorting(arrays): + return arrays([1, 2, 0]) + + +@pytest.fixture +def data_missing_for_sorting(arrays): + return arrays([1, np.nan, 0]) + + +@pytest.fixture +def na_cmp(): + # we are np.nan + return lambda x, y: np.isnan(x) and np.isnan(y) + + +@pytest.fixture +def na_value(): + return np.nan + + +@pytest.fixture +def data_for_grouping(arrays): + b = 1 + a = 0 + c = 2 + na = np.nan + return arrays([b, b, na, na, a, a, b, c]) + + +def test_dtypes(dtype): + # smoke tests on auto dtype construction + + if dtype.is_signed_integer: + assert np.dtype(dtype.type).kind == 'i' + else: + assert np.dtype(dtype.type).kind == 'u' + assert dtype.name is not None + + +class BaseInteger(object): + + def assert_series_equal(self, left, right, *args, **kwargs): + + left_na = left.isna() + right_na = right.isna() + + tm.assert_series_equal(left_na, right_na) + return tm.assert_series_equal(left[~left_na], + right[~right_na], + *args, **kwargs) + + def assert_frame_equal(self, left, right, *args, **kwargs): + # TODO(EA): select_dtypes + tm.assert_index_equal( + left.columns, right.columns, + exact=kwargs.get('check_column_type', 'equiv'), + check_names=kwargs.get('check_names', True), + check_exact=kwargs.get('check_exact', False), + check_categorical=kwargs.get('check_categorical', True), + obj='{obj}.columns'.format(obj=kwargs.get('obj', 'DataFrame'))) + + integers = (left.dtypes == 'integer').index + + for col in integers: + self.assert_series_equal(left[col], right[col], + *args, **kwargs) + + left = left.drop(columns=integers) + right = right.drop(columns=integers) + tm.assert_frame_equal(left, right, *args, **kwargs) + + +class TestDtype(BaseInteger, base.BaseDtypeTests): + + @pytest.mark.skip(reason="using multiple dtypes") + def test_is_dtype_unboxes_dtype(self): + # we have multiple dtypes, so skip + pass + + +class TestOps(BaseInteger, base.BaseOpsTests): + + def compare(self, s, op, other, exc=None): + + result = getattr(s, op)(other) + + # compute expected + mask = s.isna() + + # other array is an Integer + if isinstance(other, IntegerArray): + omask = getattr(other, 'mask', None) + mask = getattr(other, 'data', other) + if omask is not None: + mask |= omask + + # to compare properly, we convert the expected + # to float, mask to nans and convert infs + # if we have uints then we process as uints + # then conert to float + # and we ultimately want to create a IntArray + # for comparisons + rs = pd.Series(s.values.data) + expected = getattr(rs, op)(other) + + # truediv can make infs + if op in ['__truediv__', '__rtruediv__', '__rdiv__', '__div__']: + fill_value = np.nan + else: + fill_value = 0 + + try: + expected[(expected == np.inf) | (expected == -np.inf)] = fill_value + expected = expected.astype(s.dtype) + + except ValueError: + + expected = expected.astype(float) + expected[(expected == np.inf) | (expected == -np.inf)] = fill_value + expected = expected.astype(s.dtype) + + expected[mask] = np.nan + self.assert_series_equal(result, expected) + + def test_arith_integer_array(self, data, all_arithmetic_operators): + # we operate with a rhs of an integer array + + op = all_arithmetic_operators + s = pd.Series(data) + rhs = pd.Series([1] * len(data), dtype=data.dtype) + rhs.iloc[-1] = np.nan + + self.compare(s, op, rhs) + + def test_compare_scalar(self, data, all_compare_operators): + op = all_compare_operators + + # array + result = getattr(data, op)(0) + expected = getattr(data.data, op)(0) + + # fill the nan locations + expected[data.mask] = True if op == '__ne__' else False + + tm.assert_numpy_array_equal(result, expected) + + # series + s = pd.Series(data) + result = getattr(s, op)(0) + + expected = pd.Series(data.data) + expected = getattr(expected, op)(0) + + # fill the nan locations + expected[data.mask] = True if op == '__ne__' else False + + tm.assert_series_equal(result, expected) + + def test_error(self, data, all_arithmetic_operators): + # invalid ops + + op = all_arithmetic_operators + s = pd.Series(data) + ops = getattr(s, op) + opa = getattr(data, op) + + # invalid scalars + with pytest.raises(TypeError): + ops('foo') + with pytest.raises(TypeError): + ops(pd.Timestamp('20180101')) + + # invalid array-likes + with pytest.raises(TypeError): + ops(pd.Series('foo', index=s.index)) + + if op != '__rpow__': + # TODO(extension) + # rpow with a datetimelike coerces the integer array incorrectly + with pytest.raises(TypeError): + ops(pd.Series(pd.date_range('20180101', periods=len(s)))) + + # 2d + with pytest.raises(TypeError): + opa(pd.DataFrame({'A': s})) + with pytest.raises(TypeError): + opa(np.arange(len(s)).reshape(-1, len(s))) + + +class TestInterface(BaseInteger, base.BaseInterfaceTests): + pass + + +class TestConstructors(BaseInteger, base.BaseConstructorsTests): + + def test_from_dtype_from_float(self, data): + # construct from our dtype & string dtype + dtype = data.dtype + + # from float + expected = pd.Series(data) + result = pd.Series(np.array(data).astype('float'), dtype=str(dtype)) + self.assert_series_equal(result, expected) + + # from int / list + expected = pd.Series(data) + result = pd.Series(np.array(data).tolist(), dtype=str(dtype)) + self.assert_series_equal(result, expected) + + # from int / array + expected = pd.Series(data).dropna().reset_index(drop=True) + dropped = np.array(data.dropna()).astype(np.dtype((dtype.type))) + result = pd.Series(dropped, dtype=str(dtype)) + self.assert_series_equal(result, expected) + + +class TestReshaping(BaseInteger, base.BaseReshapingTests): + + def test_concat_mixed_dtypes(self, data): + # https://github.com/pandas-dev/pandas/issues/20762 + df1 = pd.DataFrame({'A': data[:3]}) + df2 = pd.DataFrame({"A": [1, 2, 3]}) + df3 = pd.DataFrame({"A": ['a', 'b', 'c']}).astype('category') + df4 = pd.DataFrame({"A": pd.SparseArray([1, 2, 3])}) + dfs = [df1, df2, df3, df4] + + # dataframes + result = pd.concat(dfs) + expected = pd.concat([x.astype(object) for x in dfs]) + self.assert_frame_equal(result, expected) + + # series + result = pd.concat([x['A'] for x in dfs]) + expected = pd.concat([x['A'].astype(object) for x in dfs]) + self.assert_series_equal(result, expected) + + result = pd.concat([df1, df2]) + expected = pd.concat([df1.astype('object'), df2.astype('object')]) + self.assert_frame_equal(result, expected) + + # concat of an Integer and Int coerces to object dtype + # TODO(jreback) once integrated this would + # be a result of Integer + result = pd.concat([df1['A'], df2['A']]) + expected = pd.concat([df1['A'].astype('object'), + df2['A'].astype('object')]) + self.assert_series_equal(result, expected) + + +class TestGetitem(BaseInteger, base.BaseGetitemTests): + pass + + +class TestMissing(BaseInteger, base.BaseMissingTests): + pass + + +class TestMethods(BaseInteger, base.BaseMethodsTests): + + @pytest.mark.xfail(reason="need a Index type with ExtensionArrays") + @pytest.mark.parametrize('dropna', [True, False]) + def test_value_counts(self, all_data, dropna): + all_data = all_data[:10] + if dropna: + other = np.array(all_data[~all_data.isna()]) + else: + other = all_data + + result = pd.Series(all_data).value_counts(dropna=dropna).sort_index() + expected = pd.Series(other).value_counts( + dropna=dropna).sort_index() + + self.assert_series_equal(result, expected) + + +class TestCasting(BaseInteger, base.BaseCastingTests): + pass + + +class TestGroupby(BaseInteger, base.BaseGroupbyTests): + pass + + +def test_frame_repr(data_missing): + + df = pd.DataFrame({'A': data_missing}) + result = repr(df) + expected = ' A\n0 NaN\n1 1' + assert result == expected + + +def test_conversions(data_missing): + + # astype to object series + df = pd.DataFrame({'A': data_missing}) + result = df['A'].astype('object') + expected = pd.Series(np.array([np.nan, 1], dtype=object), name='A') + tm.assert_series_equal(result, expected) + + # convert to object ndarray + # we assert that we are exactly equal + # including type conversions of scalars + result = df['A'].astype('object').values + expected = np.array([np.nan, 1], dtype=object) + tm.assert_numpy_array_equal(result, expected) + + for r, e in zip(result, expected): + if pd.isnull(r): + assert pd.isnull(e) + elif is_integer(r): + # PY2 can be int or long + assert r == e + assert is_integer(e) + else: + assert r == e + assert type(r) == type(e) + + +@pytest.mark.parametrize( + 'values', + [ + ['foo', 'bar'], + 'foo', + 1, + 1.0, + pd.date_range('20130101', periods=2), + np.array(['foo'])]) +def test_to_integer_array_error(values): + # error in converting existing arrays to IntegerArrays + with pytest.raises(TypeError): + to_integer_array(values) + + +@pytest.mark.parametrize( + 'values, expected', + [ + (np.array([1], dtype='int64'), Int64Array([1])), + (np.array([1, np.nan]), Int64Array([1, np.nan]))]) +def test_to_integer_array(values, expected): + # convert existing arrays to IntegerArrays + result = to_integer_array(values) + tm.assert_extension_array_equal(result, expected) + + +def test_cross_type_arithmetic(): + + df = pd.DataFrame({'A': pd.Series([1, 2, np.nan], dtype='Int64'), + 'B': pd.Series([1, np.nan, 3], dtype='UInt8'), + 'C': [1, 2, 3]}) + + result = df.A + df.C + expected = pd.Series([2, 4, np.nan], dtype='Int64') + tm.assert_series_equal(result, expected) + + result = (df.A + df.C) * 3 == 12 + expected = pd.Series([False, True, False]) + tm.assert_series_equal(result, expected) + + result = df.A + df.B + expected = pd.Series([2, np.nan, np.nan], dtype='Int64') + tm.assert_series_equal(result, expected) + + +# TODO(jreback) - these need testing / are broken + +# groupby + +# shift + +# set_index (destroys type)