Python pandas.Timedelta() Examples

def test_insert_index_timedelta64(self):
        obj = pd.TimedeltaIndex(['1 day', '2 day', '3 day', '4 day'])
        assert obj.dtype == 'timedelta64[ns]'

        # timedelta64 + timedelta64 => timedelta64
        exp = pd.TimedeltaIndex(['1 day', '10 day', '2 day', '3 day', '4 day'])
        self._assert_insert_conversion(obj, pd.Timedelta('10 day'),
                                       exp, 'timedelta64[ns]')

        # ToDo: must coerce to object
        msg = "cannot insert TimedeltaIndex with incompatible label"
        with pytest.raises(TypeError, match=msg):
            obj.insert(1, pd.Timestamp('2012-01-01'))

        # ToDo: must coerce to object
        msg = "cannot insert TimedeltaIndex with incompatible label"
        with pytest.raises(TypeError, match=msg):
            obj.insert(1, 1) 

def test_tdi_round(self):
        td = pd.timedelta_range(start='16801 days', periods=5, freq='30Min')
        elt = td[1]

        expected_rng = TimedeltaIndex([Timedelta('16801 days 00:00:00'),
                                       Timedelta('16801 days 00:00:00'),
                                       Timedelta('16801 days 01:00:00'),
                                       Timedelta('16801 days 02:00:00'),
                                       Timedelta('16801 days 02:00:00')])
        expected_elt = expected_rng[1]

        tm.assert_index_equal(td.round(freq='H'), expected_rng)
        assert elt.round(freq='H') == expected_elt

        msg = pd._libs.tslibs.frequencies.INVALID_FREQ_ERR_MSG
        with pytest.raises(ValueError, match=msg):
            td.round(freq='foo')
        with pytest.raises(ValueError, match=msg):
            elt.round(freq='foo')

        msg = "<MonthEnd> is a non-fixed frequency"
        with pytest.raises(ValueError, match=msg):
            td.round(freq='M')
        with pytest.raises(ValueError, match=msg):
            elt.round(freq='M') 

def test_to_csv_from_csv4(self):

        with ensure_clean('__tmp_to_csv_from_csv4__') as path:
            # GH 10833 (TimedeltaIndex formatting)
            dt = pd.Timedelta(seconds=1)
            df = pd.DataFrame({'dt_data': [i * dt for i in range(3)]},
                              index=pd.Index([i * dt for i in range(3)],
                                             name='dt_index'))
            df.to_csv(path)

            result = pd.read_csv(path, index_col='dt_index')
            result.index = pd.to_timedelta(result.index)
            # TODO: remove renaming when GH 10875 is solved
            result.index = result.index.rename('dt_index')
            result['dt_data'] = pd.to_timedelta(result['dt_data'])

            assert_frame_equal(df, result, check_index_type=True) 

def test_get_loc_datetimelike_overlapping(self, arrays):
        # GH 20636
        # overlapping = IntervalTree method with i8 conversion
        index = IntervalIndex.from_arrays(*arrays)

        value = index[0].mid + Timedelta('12 hours')
        result = np.sort(index.get_loc(value))
        expected = np.array([0, 1], dtype='intp')
        assert tm.assert_numpy_array_equal(result, expected)

        interval = Interval(index[0].left, index[1].right)
        result = np.sort(index.get_loc(interval))
        expected = np.array([0, 1, 2], dtype='intp')
        assert tm.assert_numpy_array_equal(result, expected)

    # To be removed, replaced by test_interval_new.py (see #16316, #16386) 

def test_get_indexer(self):
        idx = pd.to_timedelta(['0 days', '1 days', '2 days'])
        tm.assert_numpy_array_equal(idx.get_indexer(idx),
                                    np.array([0, 1, 2], dtype=np.intp))

        target = pd.to_timedelta(['-1 hour', '12 hours', '1 day 1 hour'])
        tm.assert_numpy_array_equal(idx.get_indexer(target, 'pad'),
                                    np.array([-1, 0, 1], dtype=np.intp))
        tm.assert_numpy_array_equal(idx.get_indexer(target, 'backfill'),
                                    np.array([0, 1, 2], dtype=np.intp))
        tm.assert_numpy_array_equal(idx.get_indexer(target, 'nearest'),
                                    np.array([0, 1, 1], dtype=np.intp))

        res = idx.get_indexer(target, 'nearest',
                              tolerance=Timedelta('1 hour'))
        tm.assert_numpy_array_equal(res, np.array([0, -1, 1], dtype=np.intp)) 

def test_isin_empty_datetimelike(self):
        # GH 15473
        df1_ts = DataFrame({'date':
                            pd.to_datetime(['2014-01-01', '2014-01-02'])})
        df1_td = DataFrame({'date':
                            [pd.Timedelta(1, 's'), pd.Timedelta(2, 's')]})
        df2 = DataFrame({'date': []})
        df3 = DataFrame()

        expected = DataFrame({'date': [False, False]})

        result = df1_ts.isin(df2)
        tm.assert_frame_equal(result, expected)
        result = df1_ts.isin(df3)
        tm.assert_frame_equal(result, expected)

        result = df1_td.isin(df2)
        tm.assert_frame_equal(result, expected)
        result = df1_td.isin(df3)
        tm.assert_frame_equal(result, expected)

    # Rounding 

def test_take(self):
        # GH 10295
        idx1 = timedelta_range('1 day', '31 day', freq='D', name='idx')

        for idx in [idx1]:
            result = idx.take([0])
            assert result == Timedelta('1 day')

            result = idx.take([-1])
            assert result == Timedelta('31 day')

            result = idx.take([0, 1, 2])
            expected = timedelta_range('1 day', '3 day', freq='D',
                                       name='idx')
            tm.assert_index_equal(result, expected)
            assert result.freq == expected.freq

            result = idx.take([0, 2, 4])
            expected = timedelta_range('1 day', '5 day', freq='2D',
                                       name='idx')
            tm.assert_index_equal(result, expected)
            assert result.freq == expected.freq

            result = idx.take([7, 4, 1])
            expected = timedelta_range('8 day', '2 day', freq='-3D',
                                       name='idx')
            tm.assert_index_equal(result, expected)
            assert result.freq == expected.freq

            result = idx.take([3, 2, 5])
            expected = TimedeltaIndex(['4 day', '3 day', '6 day'], name='idx')
            tm.assert_index_equal(result, expected)
            assert result.freq is None

            result = idx.take([-3, 2, 5])
            expected = TimedeltaIndex(['29 day', '3 day', '6 day'], name='idx')
            tm.assert_index_equal(result, expected)
            assert result.freq is None 

def test_astype(self):
        # GH 13149, GH 13209
        idx = TimedeltaIndex([1e14, 'NaT', NaT, np.NaN])

        result = idx.astype(object)
        expected = Index([Timedelta('1 days 03:46:40')] + [NaT] * 3,
                         dtype=object)
        tm.assert_index_equal(result, expected)

        result = idx.astype(int)
        expected = Int64Index([100000000000000] + [-9223372036854775808] * 3,
                              dtype=np.int64)
        tm.assert_index_equal(result, expected)

        result = idx.astype(str)
        expected = Index(str(x) for x in idx)
        tm.assert_index_equal(result, expected)

        rng = timedelta_range('1 days', periods=10)
        result = rng.astype('i8')
        tm.assert_index_equal(result, Index(rng.asi8))
        tm.assert_numpy_array_equal(rng.asi8, result.values) 

def test_compare_ticks_to_strs(cls):
    # GH#23524
    off = cls(19)

    # These tests should work with any strings, but we particularly are
    #  interested in "infer" as that comparison is convenient to make in
    #  Datetime/Timedelta Array/Index constructors
    assert not off == "infer"
    assert not "foo" == off

    for left, right in [("infer", off), (off, "infer")]:
        with pytest.raises(TypeError):
            left < right
        with pytest.raises(TypeError):
            left <= right
        with pytest.raises(TypeError):
            left > right
        with pytest.raises(TypeError):
            left >= right 

def test_dti_tdi_numeric_ops(self):
        # These are normally union/diff set-like ops
        tdi = TimedeltaIndex(['1 days', pd.NaT, '2 days'], name='foo')
        dti = pd.date_range('20130101', periods=3, name='bar')

        # TODO(wesm): unused?
        # td = Timedelta('1 days')
        # dt = Timestamp('20130101')

        result = tdi - tdi
        expected = TimedeltaIndex(['0 days', pd.NaT, '0 days'], name='foo')
        tm.assert_index_equal(result, expected)

        result = tdi + tdi
        expected = TimedeltaIndex(['2 days', pd.NaT, '4 days'], name='foo')
        tm.assert_index_equal(result, expected)

        result = dti - tdi  # name will be reset
        expected = DatetimeIndex(['20121231', pd.NaT, '20130101'])
        tm.assert_index_equal(result, expected) 

def test_td64arr_add_sub_td64_nat(self, box):
        # GH#23320 special handling for timedelta64("NaT")
        tdi = pd.TimedeltaIndex([NaT, Timedelta('1s')])
        other = np.timedelta64("NaT")
        expected = pd.TimedeltaIndex(["NaT"] * 2)

        obj = tm.box_expected(tdi, box)
        expected = tm.box_expected(expected, box)

        result = obj + other
        tm.assert_equal(result, expected)
        result = other + obj
        tm.assert_equal(result, expected)
        result = obj - other
        tm.assert_equal(result, expected)
        result = other - obj
        tm.assert_equal(result, expected) 

def test_take_fill_valid(self, timedelta_index):
        tdi = timedelta_index
        arr = TimedeltaArray(tdi)

        td1 = pd.Timedelta(days=1)
        result = arr.take([-1, 1], allow_fill=True, fill_value=td1)
        assert result[0] == td1

        now = pd.Timestamp.now()
        with pytest.raises(ValueError):
            # fill_value Timestamp invalid
            arr.take([0, 1], allow_fill=True, fill_value=now)

        with pytest.raises(ValueError):
            # fill_value Period invalid
            arr.take([0, 1], allow_fill=True, fill_value=now.to_period('D')) 

def test_tick_division(cls):
    off = cls(10)

    assert off / cls(5) == 2
    assert off / 2 == cls(5)
    assert off / 2.0 == cls(5)

    assert off / off.delta == 1
    assert off / off.delta.to_timedelta64() == 1

    assert off / Nano(1) == off.delta / Nano(1).delta

    if cls is not Nano:
        # A case where we end up with a smaller class
        result = off / 1000
        assert isinstance(result, offsets.Tick)
        assert not isinstance(result, cls)
        assert result.delta == off.delta / 1000

    if cls._inc < Timedelta(seconds=1):
        # Case where we end up with a bigger class
        result = off / .001
        assert isinstance(result, offsets.Tick)
        assert not isinstance(result, cls)
        assert result.delta == off.delta / .001 

def test_min_max(self):
        arr = TimedeltaArray._from_sequence([
            '3H', '3H', 'NaT', '2H', '5H', '4H',
        ])

        result = arr.min()
        expected = pd.Timedelta('2H')
        assert result == expected

        result = arr.max()
        expected = pd.Timedelta('5H')
        assert result == expected

        result = arr.min(skipna=False)
        assert result is pd.NaT

        result = arr.max(skipna=False)
        assert result is pd.NaT 

def test_timedelta64_conversions(self, m, unit):
        startdate = Series(pd.date_range('2013-01-01', '2013-01-03'))
        enddate = Series(pd.date_range('2013-03-01', '2013-03-03'))

        ser = enddate - startdate
        ser[2] = np.nan

        # op
        expected = Series([x / np.timedelta64(m, unit) for x in ser])
        result = ser / np.timedelta64(m, unit)
        tm.assert_series_equal(result, expected)

        # reverse op
        expected = Series([Timedelta(np.timedelta64(m, unit)) / x
                           for x in ser])
        result = np.timedelta64(m, unit) / ser
        tm.assert_series_equal(result, expected)

    # ------------------------------------------------------------------
    # Multiplication
    # organized with scalar others first, then array-like 

def test_slice_with_negative_step(self):
        ts = Series(np.arange(20), timedelta_range('0', periods=20, freq='H'))
        SLC = pd.IndexSlice

        def assert_slices_equivalent(l_slc, i_slc):
            assert_series_equal(ts[l_slc], ts.iloc[i_slc])
            assert_series_equal(ts.loc[l_slc], ts.iloc[i_slc])
            assert_series_equal(ts.loc[l_slc], ts.iloc[i_slc])

        assert_slices_equivalent(SLC[Timedelta(hours=7)::-1], SLC[7::-1])
        assert_slices_equivalent(SLC['7 hours'::-1], SLC[7::-1])

        assert_slices_equivalent(SLC[:Timedelta(hours=7):-1], SLC[:6:-1])
        assert_slices_equivalent(SLC[:'7 hours':-1], SLC[:6:-1])

        assert_slices_equivalent(SLC['15 hours':'7 hours':-1], SLC[15:6:-1])
        assert_slices_equivalent(SLC[Timedelta(hours=15):Timedelta(hours=7):-
                                     1], SLC[15:6:-1])
        assert_slices_equivalent(SLC['15 hours':Timedelta(hours=7):-1],
                                 SLC[15:6:-1])
        assert_slices_equivalent(SLC[Timedelta(hours=15):'7 hours':-1],
                                 SLC[15:6:-1])

        assert_slices_equivalent(SLC['7 hours':'15 hours':-1], SLC[:0]) 

def test_timedelta_assignment():
    # GH 8209
    s = Series([])
    s.loc['B'] = timedelta(1)
    tm.assert_series_equal(s, Series(Timedelta('1 days'), index=['B']))

    s = s.reindex(s.index.insert(0, 'A'))
    tm.assert_series_equal(s, Series(
        [np.nan, Timedelta('1 days')], index=['A', 'B']))

    result = s.fillna(timedelta(1))
    expected = Series(Timedelta('1 days'), index=['A', 'B'])
    tm.assert_series_equal(result, expected)

    s.loc['A'] = timedelta(1)
    tm.assert_series_equal(s, expected)

    # GH 14155
    s = Series(10 * [np.timedelta64(10, 'm')])
    s.loc[[1, 2, 3]] = np.timedelta64(20, 'm')
    expected = pd.Series(10 * [np.timedelta64(10, 'm')])
    expected.loc[[1, 2, 3]] = pd.Timedelta(np.timedelta64(20, 'm'))
    tm.assert_series_equal(s, expected) 

def test_constructor_from_index_dtlike(self, cast_as_obj, index):
        if cast_as_obj:
            result = pd.Index(index.astype(object))
        else:
            result = pd.Index(index)

        tm.assert_index_equal(result, index)

        if isinstance(index, pd.DatetimeIndex):
            assert result.tz == index.tz
            if cast_as_obj:
                # GH#23524 check that Index(dti, dtype=object) does not
                #  incorrectly raise ValueError, and that nanoseconds are not
                #  dropped
                index += pd.Timedelta(nanoseconds=50)
                result = pd.Index(index, dtype=object)
                assert result.dtype == np.object_
                assert list(result) == list(index) 

def test_tolerance_tz(self):
        # GH 14844
        left = pd.DataFrame(
            {'date': pd.date_range(start=pd.to_datetime('2016-01-02'),
                                   freq='D', periods=5,
                                   tz=pytz.timezone('UTC')),
             'value1': np.arange(5)})
        right = pd.DataFrame(
            {'date': pd.date_range(start=pd.to_datetime('2016-01-01'),
                                   freq='D', periods=5,
                                   tz=pytz.timezone('UTC')),
             'value2': list("ABCDE")})
        result = pd.merge_asof(left, right, on='date',
                               tolerance=pd.Timedelta('1 day'))

        expected = pd.DataFrame(
            {'date': pd.date_range(start=pd.to_datetime('2016-01-02'),
                                   freq='D', periods=5,
                                   tz=pytz.timezone('UTC')),
             'value1': np.arange(5),
             'value2': list("BCDEE")})
        assert_frame_equal(result, expected) 

def test_map_box(self):
        vals = [pd.Timestamp('2011-01-01'), pd.Timestamp('2011-01-02')]
        s = pd.Series(vals)
        assert s.dtype == 'datetime64[ns]'
        # boxed value must be Timestamp instance
        res = s.map(lambda x: '{0}_{1}_{2}'.format(x.__class__.__name__,
                                                   x.day, x.tz))
        exp = pd.Series(['Timestamp_1_None', 'Timestamp_2_None'])
        tm.assert_series_equal(res, exp)

        vals = [pd.Timestamp('2011-01-01', tz='US/Eastern'),
                pd.Timestamp('2011-01-02', tz='US/Eastern')]
        s = pd.Series(vals)
        assert s.dtype == 'datetime64[ns, US/Eastern]'
        res = s.map(lambda x: '{0}_{1}_{2}'.format(x.__class__.__name__,
                                                   x.day, x.tz))
        exp = pd.Series(['Timestamp_1_US/Eastern', 'Timestamp_2_US/Eastern'])
        tm.assert_series_equal(res, exp)

        # timedelta
        vals = [pd.Timedelta('1 days'), pd.Timedelta('2 days')]
        s = pd.Series(vals)
        assert s.dtype == 'timedelta64[ns]'
        res = s.map(lambda x: '{0}_{1}'.format(x.__class__.__name__, x.days))
        exp = pd.Series(['Timedelta_1', 'Timedelta_2'])
        tm.assert_series_equal(res, exp)

        # period
        vals = [pd.Period('2011-01-01', freq='M'),
                pd.Period('2011-01-02', freq='M')]
        s = pd.Series(vals)
        assert s.dtype == 'Period[M]'
        res = s.map(lambda x: '{0}_{1}'.format(x.__class__.__name__,
                                               x.freqstr))
        exp = pd.Series(['Period_M', 'Period_M'])
        tm.assert_series_equal(res, exp) 

def test_quantile(self):

        q = self.ts.quantile(0.1)
        assert q == np.percentile(self.ts.dropna(), 10)

        q = self.ts.quantile(0.9)
        assert q == np.percentile(self.ts.dropna(), 90)

        # object dtype
        q = Series(self.ts, dtype=object).quantile(0.9)
        assert q == np.percentile(self.ts.dropna(), 90)

        # datetime64[ns] dtype
        dts = self.ts.index.to_series()
        q = dts.quantile(.2)
        assert q == Timestamp('2000-01-10 19:12:00')

        # timedelta64[ns] dtype
        tds = dts.diff()
        q = tds.quantile(.25)
        assert q == pd.to_timedelta('24:00:00')

        # GH7661
        result = Series([np.timedelta64('NaT')]).sum()
        assert result == pd.Timedelta(0)

        msg = 'percentiles should all be in the interval \\[0, 1\\]'
        for invalid in [-1, 2, [0.5, -1], [0.5, 2]]:
            with pytest.raises(ValueError, match=msg):
                self.ts.quantile(invalid) 

def test_dti_cmp_object_dtype(self):
        # GH#22074
        dti = date_range('2000-01-01', periods=10, tz='Asia/Tokyo')

        other = dti.astype('O')

        result = dti == other
        expected = np.array([True] * 10)
        tm.assert_numpy_array_equal(result, expected)

        other = dti.tz_localize(None)
        with pytest.raises(TypeError):
            # tzawareness failure
            dti != other

        other = np.array(list(dti[:5]) + [Timedelta(days=1)] * 5)
        result = dti == other
        expected = np.array([True] * 5 + [False] * 5)
        tm.assert_numpy_array_equal(result, expected)

        with pytest.raises(TypeError):
            dti >= other


# ------------------------------------------------------------------
# Arithmetic 

def test_setitem_series_timedelta64(self, val, exp_dtype):
        obj = pd.Series([pd.Timedelta('1 day'),
                         pd.Timedelta('2 day'),
                         pd.Timedelta('3 day'),
                         pd.Timedelta('4 day')])
        assert obj.dtype == 'timedelta64[ns]'

        exp = pd.Series([pd.Timedelta('1 day'),
                         val,
                         pd.Timedelta('3 day'),
                         pd.Timedelta('4 day')])
        self._assert_setitem_series_conversion(obj, val, exp, exp_dtype) 

def test_at_iat_coercion(self):

        # as timestamp is not a tuple!
        dates = date_range('1/1/2000', periods=8)
        df = DataFrame(np.random.randn(8, 4),
                       index=dates,
                       columns=['A', 'B', 'C', 'D'])
        s = df['A']

        result = s.at[dates[5]]
        xp = s.values[5]
        assert result == xp

        # GH 7729
        # make sure we are boxing the returns
        s = Series(['2014-01-01', '2014-02-02'], dtype='datetime64[ns]')
        expected = Timestamp('2014-02-02')

        for r in [lambda: s.iat[1], lambda: s.iloc[1]]:
            result = r()
            assert result == expected

        s = Series(['1 days', '2 days'], dtype='timedelta64[ns]')
        expected = Timedelta('2 days')

        for r in [lambda: s.iat[1], lambda: s.iloc[1]]:
            result = r()
            assert result == expected 

def test_to_csv_with_dst_transitions(self):

        with ensure_clean('csv_date_format_with_dst') as path:
            # make sure we are not failing on transitions
            times = pd.date_range("2013-10-26 23:00", "2013-10-27 01:00",
                                  tz="Europe/London",
                                  freq="H",
                                  ambiguous='infer')

            for i in [times, times + pd.Timedelta('10s')]:
                time_range = np.array(range(len(i)), dtype='int64')
                df = DataFrame({'A': time_range}, index=i)
                df.to_csv(path, index=True)
                # we have to reconvert the index as we
                # don't parse the tz's
                result = read_csv(path, index_col=0)
                result.index = to_datetime(result.index, utc=True).tz_convert(
                    'Europe/London')
                assert_frame_equal(result, df)

        # GH11619
        idx = pd.date_range('2015-01-01', '2015-12-31',
                            freq='H', tz='Europe/Paris')
        df = DataFrame({'values': 1, 'idx': idx},
                       index=idx)
        with ensure_clean('csv_date_format_with_dst') as path:
            df.to_csv(path, index=True)
            result = read_csv(path, index_col=0)
            result.index = to_datetime(result.index, utc=True).tz_convert(
                'Europe/Paris')
            result['idx'] = to_datetime(result['idx'], utc=True).astype(
                'datetime64[ns, Europe/Paris]')
            assert_frame_equal(result, df)

        # assert working
        df.astype(str)

        with ensure_clean('csv_date_format_with_dst') as path:
            df.to_pickle(path)
            result = pd.read_pickle(path)
            assert_frame_equal(result, df) 

def test_tick_addition(kls, expected):
    offset = kls(3)
    result = offset + Timedelta(hours=2)
    assert isinstance(result, Timedelta)
    assert result == expected 

def test_delta_to_tick():
    delta = timedelta(3)

    tick = offsets._delta_to_tick(delta)
    assert (tick == offsets.Day(3))

    td = Timedelta(nanoseconds=5)
    tick = offsets._delta_to_tick(td)
    assert tick == Nano(5) 

def test_subplots_timeseries_y_axis(self):
        # GH16953
        data = {"numeric": np.array([1, 2, 5]),
                "timedelta": [pd.Timedelta(-10, unit="s"),
                              pd.Timedelta(10, unit="m"),
                              pd.Timedelta(10, unit="h")],
                "datetime_no_tz": [pd.to_datetime("2017-08-01 00:00:00"),
                                   pd.to_datetime("2017-08-01 02:00:00"),
                                   pd.to_datetime("2017-08-02 00:00:00")],
                "datetime_all_tz": [pd.to_datetime("2017-08-01 00:00:00",
                                                   utc=True),
                                    pd.to_datetime("2017-08-01 02:00:00",
                                                   utc=True),
                                    pd.to_datetime("2017-08-02 00:00:00",
                                                   utc=True)],
                "text": ["This", "should", "fail"]}
        testdata = DataFrame(data)

        ax_numeric = testdata.plot(y="numeric")
        assert (ax_numeric.get_lines()[0].get_data()[1] ==
                testdata["numeric"].values).all()
        ax_timedelta = testdata.plot(y="timedelta")
        assert (ax_timedelta.get_lines()[0].get_data()[1] ==
                testdata["timedelta"].values).all()
        ax_datetime_no_tz = testdata.plot(y="datetime_no_tz")
        assert (ax_datetime_no_tz.get_lines()[0].get_data()[1] ==
                testdata["datetime_no_tz"].values).all()
        ax_datetime_all_tz = testdata.plot(y="datetime_all_tz")
        assert (ax_datetime_all_tz.get_lines()[0].get_data()[1] ==
                testdata["datetime_all_tz"].values).all()
        with pytest.raises(TypeError):
            testdata.plot(y="text") 

def test_allow_exact_matches_and_tolerance2(self):
        # GH 13695
        df1 = pd.DataFrame({
            'time': pd.to_datetime(['2016-07-15 13:30:00.030']),
            'username': ['bob']})
        df2 = pd.DataFrame({
            'time': pd.to_datetime(['2016-07-15 13:30:00.000',
                                    '2016-07-15 13:30:00.030']),
            'version': [1, 2]})

        result = pd.merge_asof(df1, df2, on='time')
        expected = pd.DataFrame({
            'time': pd.to_datetime(['2016-07-15 13:30:00.030']),
            'username': ['bob'],
            'version': [2]})
        assert_frame_equal(result, expected)

        result = pd.merge_asof(df1, df2, on='time', allow_exact_matches=False)
        expected = pd.DataFrame({
            'time': pd.to_datetime(['2016-07-15 13:30:00.030']),
            'username': ['bob'],
            'version': [1]})
        assert_frame_equal(result, expected)

        result = pd.merge_asof(df1, df2, on='time', allow_exact_matches=False,
                               tolerance=pd.Timedelta('10ms'))
        expected = pd.DataFrame({
            'time': pd.to_datetime(['2016-07-15 13:30:00.030']),
            'username': ['bob'],
            'version': [np.nan]})
        assert_frame_equal(result, expected) 

def test_allow_exact_matches_and_tolerance(self):

        result = merge_asof(self.trades, self.quotes,
                            on='time',
                            by='ticker',
                            tolerance=Timedelta('100ms'),
                            allow_exact_matches=False)
        expected = self.allow_exact_matches_and_tolerance
        assert_frame_equal(result, expected)