Timedeltas are differences in times, expressed in difference units, e.g. days, hours, minutes,seconds. They can be both positive and negative.
Timedelta
is a subclass of datetime.timedelta
, and behaves in a similar manner,but allows compatibility with np.timedelta64
types as well as a host of custom representation,parsing, and attributes.
Parsing#
You can construct a Timedelta
scalar through various arguments, including ISO 8601 Duration strings.
In [1]: import datetime# stringsIn [2]: pd.Timedelta("1 days")Out[2]: Timedelta('1 days 00:00:00')In [3]: pd.Timedelta("1 days 00:00:00")Out[3]: Timedelta('1 days 00:00:00')In [4]: pd.Timedelta("1 days 2 hours")Out[4]: Timedelta('1 days 02:00:00')In [5]: pd.Timedelta("-1 days 2 min 3us")Out[5]: Timedelta('-2 days +23:57:59.999997')# like datetime.timedelta# note: these MUST be specified as keyword argumentsIn [6]: pd.Timedelta(days=1, seconds=1)Out[6]: Timedelta('1 days 00:00:01')# integers with a unitIn [7]: pd.Timedelta(1, unit="d")Out[7]: Timedelta('1 days 00:00:00')# from a datetime.timedelta/np.timedelta64In [8]: pd.Timedelta(datetime.timedelta(days=1, seconds=1))Out[8]: Timedelta('1 days 00:00:01')In [9]: pd.Timedelta(np.timedelta64(1, "ms"))Out[9]: Timedelta('0 days 00:00:00.001000')# negative Timedeltas have this string repr# to be more consistent with datetime.timedelta conventionsIn [10]: pd.Timedelta("-1us")Out[10]: Timedelta('-1 days +23:59:59.999999')# a NaTIn [11]: pd.Timedelta("nan")Out[11]: NaTIn [12]: pd.Timedelta("nat")Out[12]: NaT# ISO 8601 Duration stringsIn [13]: pd.Timedelta("P0DT0H1M0S")Out[13]: Timedelta('0 days 00:01:00')In [14]: pd.Timedelta("P0DT0H0M0.000000123S")Out[14]: Timedelta('0 days 00:00:00.000000123')
DateOffsets (Day, Hour, Minute, Second, Milli, Micro, Nano
) can also be used in construction.
In [15]: pd.Timedelta(pd.offsets.Second(2))Out[15]: Timedelta('0 days 00:00:02')
Further, operations among the scalars yield another scalar Timedelta
.
In [16]: pd.Timedelta(pd.offsets.Day(2)) + pd.Timedelta(pd.offsets.Second(2)) + pd.Timedelta( ....: "00:00:00.000123" ....: ) ....: Out[16]: Timedelta('2 days 00:00:02.000123')
to_timedelta#
Using the top-level pd.to_timedelta
, you can convert a scalar, array, list,or Series from a recognized timedelta format / value into a Timedelta
type.It will construct Series if the input is a Series, a scalar if the input isscalar-like, otherwise it will output a TimedeltaIndex
.
You can parse a single string to a Timedelta:
In [17]: pd.to_timedelta("1 days 06:05:01.00003")Out[17]: Timedelta('1 days 06:05:01.000030')In [18]: pd.to_timedelta("15.5us")Out[18]: Timedelta('0 days 00:00:00.000015500')
or a list/array of strings:
In [19]: pd.to_timedelta(["1 days 06:05:01.00003", "15.5us", "nan"])Out[19]: TimedeltaIndex(['1 days 06:05:01.000030', '0 days 00:00:00.000015500', NaT], dtype='timedelta64[ns]', freq=None)
The unit
keyword argument specifies the unit of the Timedelta if the inputis numeric:
In [20]: pd.to_timedelta(np.arange(5), unit="s")Out[20]: TimedeltaIndex(['0 days 00:00:00', '0 days 00:00:01', '0 days 00:00:02', '0 days 00:00:03', '0 days 00:00:04'], dtype='timedelta64[ns]', freq=None)In [21]: pd.to_timedelta(np.arange(5), unit="d")Out[21]: TimedeltaIndex(['0 days', '1 days', '2 days', '3 days', '4 days'], dtype='timedelta64[ns]', freq=None)
Warning
If a string or array of strings is passed as an input then the unit
keywordargument will be ignored. If a string without units is passed then the defaultunit of nanoseconds is assumed.
Timedelta limitations#
pandas represents Timedeltas
in nanosecond resolution using64 bit integers. As such, the 64 bit integer limits determinethe Timedelta
limits.
In [22]: pd.Timedelta.minOut[22]: Timedelta('-106752 days +00:12:43.145224193')In [23]: pd.Timedelta.maxOut[23]: Timedelta('106751 days 23:47:16.854775807')
Operations#
You can operate on Series/DataFrames and construct timedelta64[ns]
Series throughsubtraction operations on datetime64[ns]
Series, or Timestamps
.
In [24]: s = pd.Series(pd.date_range("2012-1-1", periods=3, freq="D"))In [25]: td = pd.Series([pd.Timedelta(days=i) for i in range(3)])In [26]: df = pd.DataFrame({"A": s, "B": td})In [27]: dfOut[27]: A B0 2012-01-01 0 days1 2012-01-02 1 days2 2012-01-03 2 daysIn [28]: df["C"] = df["A"] + df["B"]In [29]: dfOut[29]: A B C0 2012-01-01 0 days 2012-01-011 2012-01-02 1 days 2012-01-032 2012-01-03 2 days 2012-01-05In [30]: df.dtypesOut[30]: A datetime64[ns]B timedelta64[ns]C datetime64[ns]dtype: objectIn [31]: s - s.max()Out[31]: 0 -2 days1 -1 days2 0 daysdtype: timedelta64[ns]In [32]: s - datetime.datetime(2011, 1, 1, 3, 5)Out[32]: 0 364 days 20:55:001 365 days 20:55:002 366 days 20:55:00dtype: timedelta64[ns]In [33]: s + datetime.timedelta(minutes=5)Out[33]: 0 2012-01-01 00:05:001 2012-01-02 00:05:002 2012-01-03 00:05:00dtype: datetime64[ns]In [34]: s + pd.offsets.Minute(5)Out[34]: 0 2012-01-01 00:05:001 2012-01-02 00:05:002 2012-01-03 00:05:00dtype: datetime64[ns]In [35]: s + pd.offsets.Minute(5) + pd.offsets.Milli(5)Out[35]: 0 2012-01-01 00:05:00.0051 2012-01-02 00:05:00.0052 2012-01-03 00:05:00.005dtype: datetime64[ns]
Operations with scalars from a timedelta64[ns]
series:
In [36]: y = s - s[0]In [37]: yOut[37]: 0 0 days1 1 days2 2 daysdtype: timedelta64[ns]
Series of timedeltas with NaT
values are supported:
In [38]: y = s - s.shift()In [39]: yOut[39]: 0 NaT1 1 days2 1 daysdtype: timedelta64[ns]
Elements can be set to NaT
using np.nan
analogously to datetimes:
In [40]: y[1] = np.nanIn [41]: yOut[41]: 0 NaT1 NaT2 1 daysdtype: timedelta64[ns]
Operands can also appear in a reversed order (a singular object operated with a Series):
In [42]: s.max() - sOut[42]: 0 2 days1 1 days2 0 daysdtype: timedelta64[ns]In [43]: datetime.datetime(2011, 1, 1, 3, 5) - sOut[43]: 0 -365 days +03:05:001 -366 days +03:05:002 -367 days +03:05:00dtype: timedelta64[ns]In [44]: datetime.timedelta(minutes=5) + sOut[44]: 0 2012-01-01 00:05:001 2012-01-02 00:05:002 2012-01-03 00:05:00dtype: datetime64[ns]
min, max
and the corresponding idxmin, idxmax
operations are supported on frames:
In [45]: A = s - pd.Timestamp("20120101") - pd.Timedelta("00:05:05")In [46]: B = s - pd.Series(pd.date_range("2012-1-2", periods=3, freq="D"))In [47]: df = pd.DataFrame({"A": A, "B": B})In [48]: dfOut[48]: A B0 -1 days +23:54:55 -1 days1 0 days 23:54:55 -1 days2 1 days 23:54:55 -1 daysIn [49]: df.min()Out[49]: A -1 days +23:54:55B -1 days +00:00:00dtype: timedelta64[ns]In [50]: df.min(axis=1)Out[50]: 0 -1 days1 -1 days2 -1 daysdtype: timedelta64[ns]In [51]: df.idxmin()Out[51]: A 0B 0dtype: int64In [52]: df.idxmax()Out[52]: A 2B 0dtype: int64
min, max, idxmin, idxmax
operations are supported on Series as well. A scalar result will be a Timedelta
.
In [53]: df.min().max()Out[53]: Timedelta('-1 days +23:54:55')In [54]: df.min(axis=1).min()Out[54]: Timedelta('-1 days +00:00:00')In [55]: df.min().idxmax()Out[55]: 'A'In [56]: df.min(axis=1).idxmin()Out[56]: 0
You can fillna on timedeltas, passing a timedelta to get a particular value.
In [57]: y.fillna(pd.Timedelta(0))Out[57]: 0 0 days1 0 days2 1 daysdtype: timedelta64[ns]In [58]: y.fillna(pd.Timedelta(10, unit="s"))Out[58]: 0 0 days 00:00:101 0 days 00:00:102 1 days 00:00:00dtype: timedelta64[ns]In [59]: y.fillna(pd.Timedelta("-1 days, 00:00:05"))Out[59]: 0 -1 days +00:00:051 -1 days +00:00:052 1 days 00:00:00dtype: timedelta64[ns]
You can also negate, multiply and use abs
on Timedeltas
:
In [60]: td1 = pd.Timedelta("-1 days 2 hours 3 seconds")In [61]: td1Out[61]: Timedelta('-2 days +21:59:57')In [62]: -1 * td1Out[62]: Timedelta('1 days 02:00:03')In [63]: -td1Out[63]: Timedelta('1 days 02:00:03')In [64]: abs(td1)Out[64]: Timedelta('1 days 02:00:03')
Reductions#
Numeric reduction operation for timedelta64[ns]
will return Timedelta
objects. As usualNaT
are skipped during evaluation.
In [65]: y2 = pd.Series( ....: pd.to_timedelta(["-1 days +00:00:05", "nat", "-1 days +00:00:05", "1 days"]) ....: ) ....: In [66]: y2Out[66]: 0 -1 days +00:00:051 NaT2 -1 days +00:00:053 1 days 00:00:00dtype: timedelta64[ns]In [67]: y2.mean()Out[67]: Timedelta('-1 days +16:00:03.333333334')In [68]: y2.median()Out[68]: Timedelta('-1 days +00:00:05')In [69]: y2.quantile(0.1)Out[69]: Timedelta('-1 days +00:00:05')In [70]: y2.sum()Out[70]: Timedelta('-1 days +00:00:10')
Frequency conversion#
Timedelta Series and TimedeltaIndex
, and Timedelta
can be converted to other frequencies by astyping to a specific timedelta dtype.
In [71]: december = pd.Series(pd.date_range("20121201", periods=4))In [72]: january = pd.Series(pd.date_range("20130101", periods=4))In [73]: td = january - decemberIn [74]: td[2] += datetime.timedelta(minutes=5, seconds=3)In [75]: td[3] = np.nanIn [76]: tdOut[76]: 0 31 days 00:00:001 31 days 00:00:002 31 days 00:05:033 NaTdtype: timedelta64[ns]# to secondsIn [77]: td.astype("timedelta64[s]")Out[77]: 0 31 days 00:00:001 31 days 00:00:002 31 days 00:05:033 NaTdtype: timedelta64[s]
For timedelta64 resolutions other than the supported “s”, “ms”, “us”, “ns”,an alternative is to divide by another timedelta object. Note that division by the NumPy scalar is true division, while astyping is equivalent of floor division.
# to daysIn [78]: td / np.timedelta64(1, "D")Out[78]: 0 31.0000001 31.0000002 31.0035073 NaNdtype: float64
Dividing or multiplying a timedelta64[ns]
Series by an integer or integer Seriesyields another timedelta64[ns]
dtypes Series.
In [79]: td * -1Out[79]: 0 -31 days +00:00:001 -31 days +00:00:002 -32 days +23:54:573 NaTdtype: timedelta64[ns]In [80]: td * pd.Series([1, 2, 3, 4])Out[80]: 0 31 days 00:00:001 62 days 00:00:002 93 days 00:15:093 NaTdtype: timedelta64[ns]
Rounded division (floor-division) of a timedelta64[ns]
Series by a scalarTimedelta
gives a series of integers.
In [81]: td // pd.Timedelta(days=3, hours=4)Out[81]: 0 9.01 9.02 9.03 NaNdtype: float64In [82]: pd.Timedelta(days=3, hours=4) // tdOut[82]: 0 0.01 0.02 0.03 NaNdtype: float64
The mod (%) and divmod operations are defined for Timedelta
when operating with another timedelta-like or with a numeric argument.
In [83]: pd.Timedelta(hours=37) % datetime.timedelta(hours=2)Out[83]: Timedelta('0 days 01:00:00')# divmod against a timedelta-like returns a pair (int, Timedelta)In [84]: divmod(datetime.timedelta(hours=2), pd.Timedelta(minutes=11))Out[84]: (10, Timedelta('0 days 00:10:00'))# divmod against a numeric returns a pair (Timedelta, Timedelta)In [85]: divmod(pd.Timedelta(hours=25), 86400000000000)Out[85]: (Timedelta('0 days 00:00:00.000000001'), Timedelta('0 days 01:00:00'))
Attributes#
You can access various components of the Timedelta
or TimedeltaIndex
directly using the attributes days,seconds,microseconds,nanoseconds
. These are identical to the values returned by datetime.timedelta
, in that, for example, the .seconds
attribute represents the number of seconds >= 0 and < 1 day. These are signed according to whether the Timedelta
is signed.
These operations can also be directly accessed via the .dt
property of the Series
as well.
Note
Note that the attributes are NOT the displayed values of the Timedelta
. Use .components
to retrieve the displayed values.
For a Series
:
In [86]: td.dt.daysOut[86]: 0 31.01 31.02 31.03 NaNdtype: float64In [87]: td.dt.secondsOut[87]: 0 0.01 0.02 303.03 NaNdtype: float64
You can access the value of the fields for a scalar Timedelta
directly.
In [88]: tds = pd.Timedelta("31 days 5 min 3 sec")In [89]: tds.daysOut[89]: 31In [90]: tds.secondsOut[90]: 303In [91]: (-tds).secondsOut[91]: 86097
You can use the .components
property to access a reduced form of the timedelta. This returns a DataFrame
indexedsimilarly to the Series
. These are the displayed values of the Timedelta
.
In [92]: td.dt.componentsOut[92]: days hours minutes seconds milliseconds microseconds nanoseconds0 31.0 0.0 0.0 0.0 0.0 0.0 0.01 31.0 0.0 0.0 0.0 0.0 0.0 0.02 31.0 0.0 5.0 3.0 0.0 0.0 0.03 NaN NaN NaN NaN NaN NaN NaNIn [93]: td.dt.components.secondsOut[93]: 0 0.01 0.02 3.03 NaNName: seconds, dtype: float64
You can convert a Timedelta
to an ISO 8601 Duration string with the.isoformat
method
In [94]: pd.Timedelta( ....: days=6, minutes=50, seconds=3, milliseconds=10, microseconds=10, nanoseconds=12 ....: ).isoformat() ....: Out[94]: 'P6DT0H50M3.010010012S'
TimedeltaIndex#
To generate an index with time delta, you can use either the TimedeltaIndex orthe timedelta_range() constructor.
Using TimedeltaIndex
you can pass string-like, Timedelta
, timedelta
,or np.timedelta64
objects. Passing np.nan/pd.NaT/nat
will represent missing values.
In [95]: pd.TimedeltaIndex( ....: [ ....: "1 days", ....: "1 days, 00:00:05", ....: np.timedelta64(2, "D"), ....: datetime.timedelta(days=2, seconds=2), ....: ] ....: ) ....: Out[95]: TimedeltaIndex(['1 days 00:00:00', '1 days 00:00:05', '2 days 00:00:00', '2 days 00:00:02'], dtype='timedelta64[ns]', freq=None)
The string ‘infer’ can be passed in order to set the frequency of the index as theinferred frequency upon creation:
In [96]: pd.TimedeltaIndex(["0 days", "10 days", "20 days"], freq="infer")Out[96]: TimedeltaIndex(['0 days', '10 days', '20 days'], dtype='timedelta64[ns]', freq='10D')
Generating ranges of time deltas#
Similar to date_range(), you can construct regular ranges of a TimedeltaIndex
using timedelta_range(). The default frequency for timedelta_range
iscalendar day:
In [97]: pd.timedelta_range(start="1 days", periods=5)Out[97]: TimedeltaIndex(['1 days', '2 days', '3 days', '4 days', '5 days'], dtype='timedelta64[ns]', freq='D')
Various combinations of start
, end
, and periods
can be used withtimedelta_range
:
In [98]: pd.timedelta_range(start="1 days", end="5 days")Out[98]: TimedeltaIndex(['1 days', '2 days', '3 days', '4 days', '5 days'], dtype='timedelta64[ns]', freq='D')In [99]: pd.timedelta_range(end="10 days", periods=4)Out[99]: TimedeltaIndex(['7 days', '8 days', '9 days', '10 days'], dtype='timedelta64[ns]', freq='D')
The freq
parameter can passed a variety of frequency aliases:
In [100]: pd.timedelta_range(start="1 days", end="2 days", freq="30min")Out[100]: TimedeltaIndex(['1 days 00:00:00', '1 days 00:30:00', '1 days 01:00:00', '1 days 01:30:00', '1 days 02:00:00', '1 days 02:30:00', '1 days 03:00:00', '1 days 03:30:00', '1 days 04:00:00', '1 days 04:30:00', '1 days 05:00:00', '1 days 05:30:00', '1 days 06:00:00', '1 days 06:30:00', '1 days 07:00:00', '1 days 07:30:00', '1 days 08:00:00', '1 days 08:30:00', '1 days 09:00:00', '1 days 09:30:00', '1 days 10:00:00', '1 days 10:30:00', '1 days 11:00:00', '1 days 11:30:00', '1 days 12:00:00', '1 days 12:30:00', '1 days 13:00:00', '1 days 13:30:00', '1 days 14:00:00', '1 days 14:30:00', '1 days 15:00:00', '1 days 15:30:00', '1 days 16:00:00', '1 days 16:30:00', '1 days 17:00:00', '1 days 17:30:00', '1 days 18:00:00', '1 days 18:30:00', '1 days 19:00:00', '1 days 19:30:00', '1 days 20:00:00', '1 days 20:30:00', '1 days 21:00:00', '1 days 21:30:00', '1 days 22:00:00', '1 days 22:30:00', '1 days 23:00:00', '1 days 23:30:00', '2 days 00:00:00'], dtype='timedelta64[ns]', freq='30min')In [101]: pd.timedelta_range(start="1 days", periods=5, freq="2D5h")Out[101]: TimedeltaIndex(['1 days 00:00:00', '3 days 05:00:00', '5 days 10:00:00', '7 days 15:00:00', '9 days 20:00:00'], dtype='timedelta64[ns]', freq='53h')
Specifying start
, end
, and periods
will generate a range of evenly spacedtimedeltas from start
to end
inclusively, with periods
number of elementsin the resulting TimedeltaIndex
:
In [102]: pd.timedelta_range("0 days", "4 days", periods=5)Out[102]: TimedeltaIndex(['0 days', '1 days', '2 days', '3 days', '4 days'], dtype='timedelta64[ns]', freq=None)In [103]: pd.timedelta_range("0 days", "4 days", periods=10)Out[103]: TimedeltaIndex(['0 days 00:00:00', '0 days 10:40:00', '0 days 21:20:00', '1 days 08:00:00', '1 days 18:40:00', '2 days 05:20:00', '2 days 16:00:00', '3 days 02:40:00', '3 days 13:20:00', '4 days 00:00:00'], dtype='timedelta64[ns]', freq=None)
Using the TimedeltaIndex#
Similarly to other of the datetime-like indices, DatetimeIndex
and PeriodIndex
, you can useTimedeltaIndex
as the index of pandas objects.
In [104]: s = pd.Series( .....: np.arange(100), .....: index=pd.timedelta_range("1 days", periods=100, freq="h"), .....: ) .....: In [105]: sOut[105]: 1 days 00:00:00 01 days 01:00:00 11 days 02:00:00 21 days 03:00:00 31 days 04:00:00 4 ..4 days 23:00:00 955 days 00:00:00 965 days 01:00:00 975 days 02:00:00 985 days 03:00:00 99Freq: h, Length: 100, dtype: int64
Selections work similarly, with coercion on string-likes and slices:
In [106]: s["1 day":"2 day"]Out[106]: 1 days 00:00:00 01 days 01:00:00 11 days 02:00:00 21 days 03:00:00 31 days 04:00:00 4 ..2 days 19:00:00 432 days 20:00:00 442 days 21:00:00 452 days 22:00:00 462 days 23:00:00 47Freq: h, Length: 48, dtype: int64In [107]: s["1 day 01:00:00"]Out[107]: 1In [108]: s[pd.Timedelta("1 day 1h")]Out[108]: 1
Furthermore you can use partial string selection and the range will be inferred:
In [109]: s["1 day":"1 day 5 hours"]Out[109]: 1 days 00:00:00 01 days 01:00:00 11 days 02:00:00 21 days 03:00:00 31 days 04:00:00 41 days 05:00:00 5Freq: h, dtype: int64
Operations#
Finally, the combination of TimedeltaIndex
with DatetimeIndex
allow certain combination operations that are NaT preserving:
In [110]: tdi = pd.TimedeltaIndex(["1 days", pd.NaT, "2 days"])In [111]: tdi.to_list()Out[111]: [Timedelta('1 days 00:00:00'), NaT, Timedelta('2 days 00:00:00')]In [112]: dti = pd.date_range("20130101", periods=3)In [113]: dti.to_list()Out[113]: [Timestamp('2013-01-01 00:00:00'), Timestamp('2013-01-02 00:00:00'), Timestamp('2013-01-03 00:00:00')]In [114]: (dti + tdi).to_list()Out[114]: [Timestamp('2013-01-02 00:00:00'), NaT, Timestamp('2013-01-05 00:00:00')]In [115]: (dti - tdi).to_list()Out[115]: [Timestamp('2012-12-31 00:00:00'), NaT, Timestamp('2013-01-01 00:00:00')]
Conversions#
Similarly to frequency conversion on a Series
above, you can convert these indices to yield another Index.
In [116]: tdi / np.timedelta64(1, "s")Out[116]: Index([86400.0, nan, 172800.0], dtype='float64')In [117]: tdi.astype("timedelta64[s]")Out[117]: TimedeltaIndex(['1 days', NaT, '2 days'], dtype='timedelta64[s]', freq=None)
Scalars type ops work as well. These can potentially return a different type of index.
# adding or timedelta and date -> datelikeIn [118]: tdi + pd.Timestamp("20130101")Out[118]: DatetimeIndex(['2013-01-02', 'NaT', '2013-01-03'], dtype='datetime64[ns]', freq=None)# subtraction of a date and a timedelta -> datelike# note that trying to subtract a date from a Timedelta will raise an exceptionIn [119]: (pd.Timestamp("20130101") - tdi).to_list()Out[119]: [Timestamp('2012-12-31 00:00:00'), NaT, Timestamp('2012-12-30 00:00:00')]# timedelta + timedelta -> timedeltaIn [120]: tdi + pd.Timedelta("10 days")Out[120]: TimedeltaIndex(['11 days', NaT, '12 days'], dtype='timedelta64[ns]', freq=None)# division can result in a Timedelta if the divisor is an integerIn [121]: tdi / 2Out[121]: TimedeltaIndex(['0 days 12:00:00', NaT, '1 days 00:00:00'], dtype='timedelta64[ns]', freq=None)# or a float64 Index if the divisor is a TimedeltaIn [122]: tdi / tdi[0]Out[122]: Index([1.0, nan, 2.0], dtype='float64')
Resampling#
Similar to timeseries resampling, we can resample with a TimedeltaIndex
.
In [123]: s.resample("D").mean()Out[123]: 1 days 11.52 days 35.53 days 59.54 days 83.55 days 97.5Freq: D, dtype: float64