ENH: Series.resample performance with datetime64[ns] #7754

[scari]

It fixes the performance drop.
See %timeit result here.
closes #7754

@jreback Would you review this PR? Thanks!

[jreback]

hmm, would need a benchmark, e.g. post a sample timeit for before/after (on a good sized test frame)

[scari]

import pandas as pd

intSeries = pd.Series(5, pd.date_range(start='2000-01-01', end='2000-01-08', freq='555000U'), dtype='int64')
timeSeries = intSeries.astype('datetime64[ns]')

%timeit intSeries.resample('1S', how='last')
%timeit timeSeries.resample('1S', how='last')
%prun intSeries.resample('1S', how='last')
%prun timeSeries.resample('1S', how='last')

Before

100 loops, best of 3: 18.8 ms per loop
1 loops, best of 3: 17 s per loop
449 function calls (441 primitive calls) in 0.021 seconds
[snip]
22378318 function calls (21773499 primitive calls) in 24.219 seconds
[snip]

After

100 loops, best of 3: 19 ms per loop
100 loops, best of 3: 20 ms per loop
449 function calls (441 primitive calls) in 0.020 seconds
[snip]
418 function calls (410 primitive calls) in 0.020 seconds
[snip]

[jreback]

ok, looks ok, can you add a vbench in vb_suite/timeseries.py. you can make similar to the above. but make smaller, so that the slowest time doesn't take too long (max 1s)

[scari]

Added test. @jreback please review again.

$ ./test_perf.sh -b HEAD~1 -t HEAD -r timeseries
-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------
timeseries_resample_datetime64               |   3.7426 | 2009.6250 |   0.0019 |

Target [dd6ffa3] : ENH: Series.resample performance with datetime64[ns] #7754
Base   [3128215] : cleanup test for GH10044

[scari]

Updated whatsnew/v0.17.0.txt.

[jreback]

looks good ping when green

[scari]

on green! :) @jreback

[jreback]

ty!