PERF: perf of e6b99f4 vs v0.12.0

[jreback]

I have put boxes by the ones 2 look at, some/most of these were caused by #3482, but always worth taking a look.

• frame_xs_col, miniscule
• groupby_multi_python, CLN: series to now inherit from NDFrame #3482
• concat_series_axis1, series construction overhead (so ok)
• period_setitem, issue PERF: fastpath on Period construction from PeriodIndex #5155, will need to be address in 0.14
• frame_iloc_dups, miniscule
• reshape_stack_simple, fixed after PERF: calling get_indexer twice when a specialized indexer (e.g. pad) could be called instead #5178
• frame_iteritems, CLN: series to now inherit from NDFrame #3482
• frame_fancy_lookup, series construction overhead (so ok)
• sparse_frame_constructor, CLN: series to now inherit from NDFrame #3482
• timeseries_period_downsample_mean, PR improved daytimes conversion performance in period.c #5204
• series_timestamp_compare, PR PERF: remove auto-boxing on isnull/notnull #5154 (was boxing isnull)
• reindex_daterange_pad, PR PERF: calling get_indexer twice when a specialized indexer (e.g. pad) could be called instead #5178 (not using specialized indexer)
• reindex_daterange_backfill, PR PERF: calling get_indexer twice when a specialized indexer (e.g. pad) could be called instead #5178 (not using specialized indexer)

The good

-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------
plot_timeseries_period                       |  54.3883 | 1061.3217 |   0.0512 |
timeseries_to_datetime_YYYYMMDD              |   9.4270 |  42.3674 |   0.2225 |
write_store_table_mixed                      |  36.2830 | 116.2400 |   0.3121 |
packers_write_hdf_table                      |  29.6400 |  77.6550 |   0.3817 |
write_store_table_panel                      |  47.1393 |  88.3977 |   0.5333 |
write_store_table                            |  30.7643 |  56.7960 |   0.5417 |
read_csv_thou_vb                             |  17.3206 |  30.3980 |   0.5698 |
frame_getitem_single_column                  |  25.7526 |  40.0293 |   0.6433 |

The bad

-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------
frame_xs_col                                 |   0.0350 |   0.0233 |   1.5051 |
groupby_multi_python                         | 123.9320 |  81.8870 |   1.5135 |
concat_series_axis1                          |  93.0324 |  60.6890 |   1.5329 |
period_setitem                               | 212.0950 | 138.1503 |   1.5352 |
frame_iloc_dups                              |   0.3070 |   0.1993 |   1.5403 |
reshape_stack_simple                         |   1.9557 |   1.1647 |   1.6792 |
frame_iteritems                              |  37.6317 |  19.2017 |   1.9598 |
frame_fancy_lookup                           |   3.4526 |   1.6307 |   2.1173 |
sparse_frame_constructor                     |  12.2224 |   5.3250 |   2.2953 |
timeseries_period_downsample_mean            |  14.1153 |   5.4576 |   2.5863 |
series_timestamp_compare                     |   7.3427 |   1.8593 |   3.9491 |
reindex_daterange_pad                        |   2.8083 |   0.6460 |   4.3476 |
reindex_daterange_backfill                   |   2.8074 |   0.6456 |   4.3482 |
-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------

Ratio < 1.0 means the target commit is faster then the baseline.
Seed used: 1234

Target [e6b99f4] : Merge pull request #5134 from jreback/list_index

BUG: Treat a list/ndarray identically for iloc indexing with list-like (GH5006)
Base   [8c0a34f] : RLS: set released to True, edit release dates

[jtratner]

what's the comparison right before e6b99f4? If not fixing that leads to better performance, maybe it's worth the edge case bug?

[jreback]

that commit is just the latest master (as of last night). Their is probably not much to be done with these in any event. Believe me I have spent quite a bit of time on these. but worth a shot in any event.

[jtratner]

@jreback cool - I'm absolutely sure you know what's up 😄