PERF: fastpath on Period construction from PeriodIndex

[jreback]

vbench already in timeseries/period_setitem
very inefficient because on the boxing of the PeriodIndex (to an object index), it recomputes the freq for each Period (even though it is passed), need a fastpath on this type of construction

In [1]: rng = period_range('1/1/1990', freq='S', periods=20000)

In [2]: df = DataFrame(index=range(len(rng)))

In [3]: def f():
   ...:     df['col'] = rng
   ...:     
   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    40000    0.055    0.000    0.095    0.000 abc.py:128(__instancecheck__)
    20000    0.043    0.000    0.345    0.000 period.py:66(__init__)
    20000    0.037    0.000    0.381    0.000 period.py:683(<lambda>)
    80031    0.037    0.000    0.131    0.000 {isinstance}
    20000    0.034    0.000    0.223    0.000 frequencies.py:76(get_freq_code)
    60003    0.029    0.000    0.029    0.000 _weakrefset.py:68(__contains__)
    20000    0.028    0.000    0.063    0.000 frequencies.py:464(_base_and_stride)
    20000    0.028    0.000    0.028    0.000 {method 'match' of '_sre.SRE_Pattern' objects}
    20000    0.027    0.000    0.038    0.000 frequencies.py:767(_period_str_to_code)
    40000    0.023    0.000    0.144    0.000 common.py:1730(is_integer)
        1    0.016    0.016    0.398    0.398 {pandas.lib.map_infer}
    40007    0.010    0.000    0.010    0.000 {getattr}
    20000    0.009    0.000    0.012    0.000 frequencies.py:112(_get_freq_str)
    60000    0.008    0.000    0.008    0.000 {method 'get' of 'dict' objects}
    40000    0.006    0.000    0.006    0.000 {method 'group' of '_sre.SRE_Match' objects}
    20000    0.004    0.000    0.004    0.000 {method 'lower' of 'str' objects}
    20000    0.002    0.000    0.002    0.000 {method 'upper' of 'str' objects}
    20015    0.002    0.000    0.002    0.000 {len}

[jtratner]

@jreback did you resolve this already? I might have to deal with this soon anyways, as I'm about to tackle Period and DatetimeIndex fixups.

[jreback]

nope
it shouldn't be too hard; Period needs a fast path where it just sets stuff

it works now but does lots of work when u have a long PeriodIndex

[jreback]

IOW it reinterprets the freq code on every Period construction even when u r passing valid values (but it doesn't know that)

[jtratner]

You should also make sure you're fastpathing asobject (not sure if it matters for PI though)

[jreback]

not planning in this till 0.14

[jtratner]

got it

On Mon, Oct 14, 2013 at 11:10 PM, jreback notifications@github.com wrote:

not planning in this till 0.14

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/5155#issuecomment-26306475
.

[jreback]

revisit after #5148

[nehalecky]

Any relevance: #2949?

[jreback]

that's a separate issue...this is a pretty easy fix

[nehalecky]

Thanks, just wanted to throw it in the mix. I love (conceptually) PeriodIndexs, but they need a lot of tlc soon to be all they can be. :)