Timestamp should have alternate constructor for UTC millisecond timestamps

[wesm]

Was surprised at how difficult it was to convert a unix timestamp to a pandas Timestamp recently

[jreback]

I think an API like:

pd.to_datetime(values, unit='s')
Timestamp(value,unit='s')

might be all that we need, in that the unit is a hint that I am giving you something that this is in these units

e.g. right now the default is (so no conversion from int needed)

In [9]: Timestamp(1368431149000000000, unit='ns')
Out[9]: <Timestamp: 2013-05-13 07:45:49>

http://stackoverflow.com/questions/16517240/pandas-using-unix-epoch-timestamp-as-datetime-index

In [3]: np.array([1368431149, 1368431150]).astype('datetime64[s]')
Out[3]: array([2013-05-13 07:45:49, 2013-05-13 07:45:50], dtype=datetime64[s])

In [4]: pd.to_datetime(np.array([1368431149, 1368431150])*int(1e6))
Out[4]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[1970-01-16 20:07:11.149000, 1970-01-16 20:07:11.150000]
Length: 2, Freq: None, Timezone: None

In [5]: pd.to_datetime(np.array([1368431149, 1368431150])*int(1e9))
Out[5]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-05-13 07:45:49, 2013-05-13 07:45:50]
Length: 2, Freq: None, Timezone: None

[jreback]

@hayd what do u think about this API?

[hayd]

I think unit is descriptive, I like it.