PERF: Regressions between 0.20.1 and 0.20.x

[TomAugspurger]

I'll add more later, I'm worried my local benchmarks aren't reliable, but I think these ones are real.

Interpolate

       before           after         ratio
     [20baf972]       [eefbaf71]
+        50.6±2ms            1.48s    29.27  frame_methods.Interpolate.time_interpolate
+     1.19±0.01ms       14.9±0.7ms    12.53  frame_methods.Interpolate.time_interpolate_some_good
+      2.45±0.1ms       13.6±0.8ms     5.57  frame_methods.Interpolate.time_interpolate_some_good_infer

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.

groupby

       before           after         ratio
     [20baf972]       [98162e04]
+     1.45±0.03ms      2.49±0.04ms     1.72  groupby.groupby_datetimetz.time_groupby_sum
+     2.71±0.03ms      4.53±0.09ms     1.67  groupby.groupby_nth.time_groupby_series_nth_any
+     1.59±0.02ms      2.51±0.04ms     1.58  groupby.groupby_nth.time_groupby_series_nth_none
+     3.69±0.03ms      4.95±0.09ms     1.34  groupby.groupby_multi.time_groupby_series_simple_cython
+     1.41±0.04ms      1.69±0.04ms     1.19  groupby.groupby_datetime.time_groupby_sum
+     1.91±0.03ms      2.20±0.01ms     1.15  groupby.GroupBySuite.time_cumprod('int', 10000)
+     1.61±0.03ms      1.80±0.02ms     1.12  groupby.GroupBySuite.time_count('int', 10000)
+      9.83±0.2ms      10.9±0.09ms     1.11  groupby.GroupBySuite.time_all('int', 100)
+     1.74±0.03ms      1.93±0.02ms     1.11  groupby.GroupBySuite.time_cummin('int', 10000)

[TomAugspurger]

cc @WBare for the interpolate regression, #16429 was the commit. Do you have chance to see if there's any easy optimizations to make? If not I can take a look later.

[jreback]

what causes the interpolate ones?

[TomAugspurger]

Looking into it. I have an optimization for it that should be ok. Will have it later.

…

On Jun 2, 2017, at 16:54, Jeff Reback ***@***.***> wrote: what causes the interpolate ones? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[TomAugspurger]

@jreback the groupby slowdown + 3.47±0.08ms 5.30±0.2ms 1.53 groupby.groupby_multi.time_groupby_series_simple_cython points to #16413

Thoughts? Does the increased perf elsewhere make up for it?

[jreback]

not sure what u mean
this is a perf improvement vs previous versions for other benchmarks

i would remeasure this

[TomAugspurger]

It affected time_groupby_series_simple_cython negatively

$ asv continuous d5a681bfa~1 d5a681bfa -b groupby.groupby_multi.time_groupby_series_simple_cython 

· Creating environments
· Discovering benchmarks
·· Uninstalling from conda-py3.6-Cython-matplotlib-numexpr-numpy-openpyxl-pytables-pytest-scipy-sqlalchemy-xlrd-xlsxwriter-xlwt
·· Installing into conda-py3.6-Cython-matplotlib-numexpr-numpy-openpyxl-pytables-pytest-scipy-sqlalchemy-xlrd-xlsxwriter-xlwt..
· Running 2 total benchmarks (2 commits * 1 environments * 1 benchmarks)
[  0.00%] · For pandas commit hash d5a681bf:
[  0.00%] ·· Building for conda-py3.6-Cython-matplotlib-numexpr-numpy-openpyxl-pytables-pytest-scipy-sqlalchemy-xlrd-xlsxwriter-xlwt..........................................................................................
[  0.00%] ·· Benchmarking conda-py3.6-Cython-matplotlib-numexpr-numpy-openpyxl-pytables-pytest-scipy-sqlalchemy-xlrd-xlsxwriter-xlwt
[ 50.00%] ··· Running groupby.groupby_multi.time_groupby_series_simple_cython                                                                                                                                                                                                                                    5.04±0.09ms
[ 50.00%] · For pandas commit hash 5fe042f5:
[ 50.00%] ·· Building for conda-py3.6-Cython-matplotlib-numexpr-numpy-openpyxl-pytables-pytest-scipy-sqlalchemy-xlrd-xlsxwriter-xlwt..........................................................................................
[ 50.00%] ·· Benchmarking conda-py3.6-Cython-matplotlib-numexpr-numpy-openpyxl-pytables-pytest-scipy-sqlalchemy-xlrd-xlsxwriter-xlwt
[100.00%] ··· Running groupby.groupby_multi.time_groupby_series_simple_cython                                                                                                                                                                                                                                    3.79±0.09ms       
       before           after         ratio
     [5fe042f5]       [d5a681bf]
+     3.79±0.09ms      5.04±0.09ms     1.33  groupby.groupby_multi.time_groupby_series_simple_cython

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.

[jreback]

ok create an issue and i will look
don't block for the release

[TomAugspurger]

Sounds good. Agreed that it's not worth blocking the release over.

[TomAugspurger]

Closed by #16592