BUG: Fixed float parsing with unit when using pd.to_datetime (GH13834)

[wcwagner]

• closes [BUG]: Wrong unix timestamp parsing with floating point using pd.to_datetime #13834
• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

[codecov-io]

Current coverage is 85.30% (diff: 100%)

Merging #13847 into master will decrease coverage by <.01%

@@             master     #13847   diff @@
==========================================
  Files           139        139          
  Lines         50157      50157          
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
- Hits          42785      42784     -1   
- Misses         7372       7373     +1   
  Partials          0          0          

Powered by Codecov. Last update b7abef4...54a26ee

[sinhrks]

#13834?

[chris-b1]

I think this should actually be casting='same_kind'. You can also add copy=False to avoid a copy in the case where the values are already int64.

In [35]: a = np.array([1, 2, 3], dtype='int32')

In [36]: a.astype('int64', casting='no')
TypeError                                 Traceback (most recent call last)
<ipython-input-36-efb09bba64c8> in <module>()
----> 1 a.astype('int64', casting='no')

TypeError: Cannot cast array from dtype('int32') to dtype('int64') according to the rule 'no'

In [37]: a.astype('int64', casting='same_kind')
Out[37]: array([1, 2, 3], dtype=int64)

[jreback]

@wcwagner can you update according to @chris-b1 suggestion

[wcwagner]

Having trouble w/ my laptop recently, so sorry for slow response. Please lmk if there's anything else, I will try to fix anything asap.

Thanks for the suggestion @chris-b1

[wcwagner]

@jreback everything good here?

[jreback]

be much more explict here, this is only when unit specified.

[jreback]

lgtm. ping when fixed / green.

[wcwagner]

@jreback

[jreback]

just about to merge this and I saw:
which don't look right

In [8]: pd.to_datetime(1000.1,unit='D').value
Out[8]: 86400000000000000

In [9]: pd.to_datetime(1000.1,unit='s').value
Out[9]: 1000000000000

[jreback]

@wcwagner nvm my last.....didn't compile the extension :<

thanks @wcwagner