Categorical(vals, cats) bad performance with NaNs

[Winand]

NaNs in datetime64 data values GREATLY reduce performance of Categorical(values, cats):

import pandas as pd
tmp=pd.Series(pd.DatetimeIndex(pd.np.datetime64('1995-01-01 00:00')+i for i in range(1000000)))
to = tmp.astype('category')
cats = to.cat.categorical._categories.values

%timeit pd.Categorical(tmp.values, cats)
1 loops, best of 3: 250 ms per loop

tmp[500000] = pd.NaT
%timeit pd.Categorical(tmp.values, cats)
1 loops, best of 3: 10.1 s per loop

tmp[tmp.isnull()] = pd.np.datetime64('0')
%timeit pd.Categorical(tmp.values, cats)
1 loops, best of 3: 251 ms per loop

Small issue with printing Categorical datetime64:

ds = pd.Categorical([pd.np.datetime64("2014-01-01"), pd.NaT])
>>> ds
[2014-01-01, 2014-01-01] <--- NO, -1 is NOT a category:-)
Categories (1, datetime64[ns]): [2014-01-01]
>>> ds.astype('datetime64')
array(['2014-01-01T03:00:00.000000000+0300', 'NaT'], dtype='datetime64[ns]')

Versions:

commit: None
python: 3.4.3.final.0
python-bits: 32
OS: Windows
OS-release: 7
machine: AMD64
processor: Intel64 Family 6 Model 58 Stepping 9, GenuineIntel
byteorder: little
LC_ALL: None
LANG: en

pandas: 0.17.1
nose: 1.3.7
pip: 7.1.2
setuptools: 18.7.1
Cython: 0.23.4
numpy: 1.9.3
scipy: 0.16.1
statsmodels: 0.6.1
IPython: 4.0.1
sphinx: 1.3.3
patsy: 0.4.1
dateutil: 2.4.2
pytz: 2015.7
blosc: 1.2.8
bottleneck: 1.0.0
tables: 3.2.2
numexpr: 2.4.6
matplotlib: 1.5.0
openpyxl: None
xlrd: 0.9.4
xlwt: None
xlsxwriter: 0.7.7
lxml: 3.5.0
bs4: 4.4.1
html5lib: 0.9999999
httplib2: None
apiclient: None
sqlalchemy: 1.0.9
pymysql: None
psycopg2: None
Jinja2: 2.8

[jreback]

pls show a copy pastable example and
pd.show_versions()

[jreback]

can u show using
%%timeit in ipython instead
it's much easier to read

[Winand]

At first i've tried to initialize like this:

to=pd.Series(pd.DatetimeIndex(range(1000000))).astype('category')
cats = to.cat.categorical._categories.values
tmp=pd.Series(pd.DatetimeIndex(range(1000000)))

but it gives wrong results in the 1st case (a bug?):

>>>c1
[1970-01-01, 1970-01-01, 1970-01-01, 1970-01-01, 1970-01-01, ..., 1970-01-01 00:00:00.000999, 1970-01-01 00:00:00.000999, 1970-01-01 00:00:00.000999, 1970-01-01 00:00:00.000999, 1970-01-01 00:00:00.000999]
Length: 1000000
Categories (1000000, datetime64[ns]): [1970-01-01 00:00:00.000000000, 1970-01-01 00:00:00.000000001,
                                       1970-01-01 00:00:00.000000002, 1970-01-01 00:00:00.000000003, ...,
                                       1970-01-01 00:00:00.000999996, 1970-01-01 00:00:00.000999997,
                                       1970-01-01 00:00:00.000999998, 1970-01-01 00:00:00.000999999]

>>>c2
[1970-01-01 00:00:00.000000000, 1970-01-01 00:00:00.000000001, 1970-01-01 00:00:00.000000002, 1970-01-01 00:00:00.000000003, 1970-01-01 00:00:00.000000004, ..., 1970-01-01 00:00:00.000999995, 1970-01-01 00:00:00.000999996, 1970-01-01 00:00:00.000999997, 1970-01-01 00:00:00.000999998, 1970-01-01 00:00:00.000999999]
Length: 1000000
Categories (1000000, datetime64[ns]): [1970-01-01 00:00:00.000000000, 1970-01-01 00:00:00.000000001,
                                       1970-01-01 00:00:00.000000002, 1970-01-01 00:00:00.000000003, ...,
                                       1970-01-01 00:00:00.000999996, 1970-01-01 00:00:00.000999997,
                                       1970-01-01 00:00:00.000999998, 1970-01-01 00:00:00.000999999]
equal? False

[jreback]

@Winand

#12128 should fix a multitude of categorical with NaT issues/perf.

was converting them to object dtype under the hood (bad) and not treating NaT like nan