PERF: Regressions since 0.20.3

[TomAugspurger]

These are fairly reproducible on my machine:

       before           after         ratio
     [eac4d3f7]       [92db5c9c]
+           396μs            746μs     1.88  inference.DtypeInfer.time_uint32
+     2.63±0.02ms       3.93±0.1ms     1.49  indexing.IndexingMethods.time_take_dtindex
+     2.64±0.06ms      3.83±0.06ms     1.45  indexing.IndexingMethods.time_take_intindex

       before           after         ratio
     [eac4d3f7]       [92db5c9c]
+     1.43±0.01ms      1.74±0.07ms     1.22  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BDay', 1)
+     1.56±0.02ms      1.74±0.07ms     1.12  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('MonthBegin', 2)
+          56.0ms           61.8ms     1.10  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('FY5253Quarter_2', 2)

[jreback]

cc @jbrockmendel if u have any ideas here

[jbrockmendel]

I'll take a look at this.

[jbrockmendel]

Best guess is the change in _to_i8 (previously in _libs.index, now in tslib). Old:

cdef inline _to_i8(object val):
    cdef pandas_datetimestruct dts
    try:
        return val.value
    except AttributeError:
        if util.is_datetime64_object(val):
            return get_datetime64_value(val)
        elif PyDateTime_Check(val):
            tzinfo = getattr(val, 'tzinfo', None)
            # Save the original date value so we can get the utcoffset from it.
            ival = _pydatetime_to_dts(val, &dts)
            if tzinfo is not None and not is_utc(tzinfo):
                offset = get_utcoffset(tzinfo, val)
                ival -= tslib._delta_to_nanoseconds(offset)
            return ival
        return val

New:

cdef inline _to_i8(object val):
    cdef pandas_datetimestruct dts
    try:
        return val.value
    except AttributeError:
        if is_datetime64_object(val):
            return get_datetime64_value(val)
        elif PyDateTime_Check(val):
            return Timestamp(val).value
        return val

[jbrockmendel]

Hmm in %timeit reverting this makes a pretty big difference for the ind.get_loc(0) bench, but pretty small (though at least in the right direction) differences for time_take_dtindex and time_take_intindex

[jbrockmendel]

Does asv have something analogous to git bisect?

[jorisvandenbossche]

Maybe https://asv.readthedocs.io/en/latest/commands.html#asv-find ?

[TomAugspurger]

Sorry, those commit hashes in the original post look entirely incorrect... Let me rerun them.

[TomAugspurger]

Updated:

       before           after         ratio
       [888c4ec3]       [36c309ed]
!     4.01±0.02ms           failed      n/a  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('LastWeekOfMonth', 1)
!     4.39±0.01ms           failed      n/a  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('LastWeekOfMonth', 2)
!      9.17±0.1ms           failed      n/a  frame_methods.Formatting.time_frame_repr_wide
!      19.7±0.2ms           failed      n/a  frame_methods.Formatting.time_html_repr_trunc_mi
!      14.4±0.2ms           failed      n/a  frame_methods.Formatting.time_html_repr_trunc_si
!      10.8±0.1ms           failed      n/a  frame_methods.Formatting.time_repr_tall
!         117±2ms           failed      n/a  frame_methods.Formatting.time_to_html_mixed
!      14.1±0.1ms           failed      n/a  frame_methods.Formatting.time_to_string_floats
+      10.3±0.5ms            4.83s   469.35  inference.DtypeInfer.time_timedelta64_2
+     8.94±0.09ms            3.03s   339.45  inference.DtypeInfer.time_datetime64
+      29.8±0.4ms            5.99s   201.19  binary_ops.Timeseries.time_timestamp_ops_diff2
+        80.7±1ms            9.27s   114.89  binary_ops.TimeseriesTZ.time_timestamp_ops_diff2
+     3.36±0.03ms          278±2ms    82.71  timeseries.SeriesArithmetic.time_add_offset_delta
+      60.2±0.4ms            3.91s    64.91  inference.DtypeInfer.time_timedelta64_1
+      9.64±0.2ms          297±5ms    30.83  timeseries.SeriesArithmetic.time_add_offset_fast
+     6.18±0.09ms          122±1ms    19.77  categoricals.Categoricals3.time_rank_string_cat
+      36.6±0.4ms          576±1ms    15.75  packers.packers_read_stata.time_packers_read_stata
+     45.8±0.01ms            616ms    13.44  packers.packers_read_stata_with_validation.time_packers_read_stata_with_validation
+      28.5±0.2ms          308±2ms    10.84  packers.STATA.time_write_stata
+      40.5±0.4ms        323±0.9ms     7.97  packers.STATA.time_write_stata_with_validation
+      34.1±0.4ms          107±1ms     3.15  join_merge.Align.time_series_align_int64_index
+        631±10μs      1.51±0.01ms     2.39  indexing.Int64Indexing.time_getitem_array
+     7.13±0.04ms      16.4±0.09ms     2.30  timeseries.DatetimeIndex.time_infer_freq_none
+     3.62±0.05ms       7.34±0.1ms     2.03  indexing.Int64Indexing.time_getitem_lists
+     2.44±0.04μs      4.25±0.05μs     1.74  indexing.DataFrameIndexing.time_get_value
+         285±6μs         490±10μs     1.72  indexing.Int64Indexing.time_getitem_list_like
+      19.2±0.1ms       30.8±0.8ms     1.60  sparse.sparse_array_constructor.time_sparse_array_constructor_object_non_nan_fill_value_1percent
+      32.2±0.4ms       48.7±0.1ms     1.51  sparse.sparse_array_constructor.time_sparse_array_constructor_object_non_nan_fill_value_10percent
+     2.47±0.03ms      3.72±0.04ms     1.51  indexing.IndexingMethods.time_take_dtindex
+     2.48±0.07ms      3.73±0.03ms     1.50  indexing.IndexingMethods.time_take_intindex
+      6.12±0.1ms      9.04±0.05ms     1.48  binary_ops.Ops2.time_frame_int_div_by_zero
+     5.33±0.01ms      7.65±0.01ms     1.43  timeseries.DatetimeIndex.time_infer_freq_business
+        564±20ms          792±7ms     1.40  timeseries.SeriesArithmetic.time_add_offset_slow
+     2.95±0.09ms      4.05±0.03ms     1.37  strings.StringMethods.time_get
+     2.68±0.02ms      3.66±0.06ms     1.37  period.Algorithms.time_value_counts_pseries
+     2.09±0.03ms      2.87±0.02ms     1.37  period.Algorithms.time_drop_duplicates_pseries
+     6.84±0.07μs       8.89±0.1μs     1.30  timestamp.TimestampOps.time_replace_across_dst
+         252±2ms          322±5ms     1.28  rolling.SeriesRolling.time_rolling_quantile_median_l
+      7.91±0.1μs       10.1±0.1μs     1.28  timestamp.TimestampOps.time_replace_tz
+       252±0.6ms          322±1ms     1.28  rolling.DataframeRolling.time_rolling_quantile_median_l
+     1.25±0.01ms      1.58±0.01ms     1.26  categoricals.Categoricals.time_constructor_datetimes_with_nat
+         1.30±0s            1.61s     1.24  groupby.GroupBySuite.time_unique('float', 10000)
+      10.1±0.2μs       12.5±0.2μs     1.23  timestamp.TimestampProperties.time_is_quarter_end
+         842±4ms          1.02±0s     1.22  groupby.GroupBySuite.time_unique('int', 10000)
+         113±1ms          135±1ms     1.19  rolling.SeriesRolling.time_rolling_quantile_median
+         111±1ms          132±2ms     1.19  rolling.DataframeRolling.time_rolling_quantile_median
+     3.17±0.04ms      3.77±0.05ms     1.19  index_object.SetOperations.time_datetime_difference
+      9.66±0.1ms       11.4±0.1ms     1.18  groupby.GroupBySuite.time_unique('int', 100)
+      35.5±0.5μs       41.5±0.4μs     1.17  indexing.IntervalIndexing.time_getitem_scalar
+      14.5±0.2ms       16.9±0.4ms     1.17  groupby.GroupBySuite.time_unique('float', 100)
+      59.9±0.3ms       69.8±0.3ms     1.17  hdfstore_bench.HDF5.time_write_store_table
+        34.7±1ms       40.3±0.5ms     1.16  strings.StringMethods.time_join_split_expand
+     3.98±0.05ms      4.59±0.05ms     1.15  groupby.Transform.time_transform_ufunc
+         227±2μs          261±6μs     1.15  indexing.MultiIndexing.time_series_xs_mi_ix
+      10.5±0.1μs       12.1±0.2μs     1.15  indexing.Int64Indexing.time_iloc_scalar
+     1.68±0.01ms         1.93±0ms     1.14  indexing.MultiIndexing.time_multiindex_get_indexer
+         251±4μs          287±2μs     1.14  indexing.MultiIndexing.time_multiindex_large_get_loc
+     6.53±0.03μs      7.47±0.07μs     1.14  timestamp.TimestampOps.time_replace_None
+      41.3±0.7μs       46.8±0.5μs     1.13  indexing.Int64Indexing.time_loc_scalar
+      84.1±0.7μs       94.4±0.6μs     1.12  series_methods.series_clip.time_series_dropna_datetime
+           508ms            568ms     1.12  groupby.groupby_multi_index.time_groupby_multi_index
+       206±0.8ms          231±2ms     1.12  timeseries.ToDatetime.time_iso8601_tz_spaceformat
+         199±2ms        221±0.5ms     1.12  packers.Packers.time_packers_read_csv
+         606±3μs          674±5μs     1.11  groupby.GroupBySuite.time_value_counts('float', 100)
+       265±0.9μs          294±3μs     1.11  indexing.MultiIndexing.time_frame_xs_mi_ix
+      8.49±0.1ms         9.39±1ms     1.11  eval.Eval.time_mult('numexpr', 1)
+     8.07±0.04μs      8.91±0.09μs     1.10  timeseries.DatetimeIndex.time_timestamp_tzinfo_cons
+         209±2ms          231±3ms     1.10  groupby.GroupBySuite.time_describe('float', 100)
+          73.8ms           81.4ms     1.10  packers.JSON.time_write_json
+      58.2±0.7μs       64.0±0.7μs     1.10  indexing.IntervalIndexing.time_loc_scalar
+      31.3±0.2μs       34.4±0.6μs     1.10  indexing.Int64Indexing.time_ix_scalar

verifying for correctness and then I'll start digging into them.

[TomAugspurger]

+      10.3±0.5ms            4.83s   469.35  inference.DtypeInfer.time_timedelta64_2
+     8.94±0.09ms            3.03s   339.45  inference.DtypeInfer.time_datetime64

are real. _maybe_box_datetimelike looks suspicious (this is the RC):

ncalls	tottime	percall	cumtime	percall	filename:lineno(function)
2000000	6.369	3.185e-06	6.57	3.285e-06	common.py:170(_maybe_box_datetimelike)
2	0.6152	0.3076	7.185	3.593	base.py:813()
2000723/2000456	0.2013	1.006e-07	0.203	1.015e-07	~:0()

http://pandas.pydata.org/speed/pandas/#inference.DtypeInfer.time_datetime64 points to ~ eef810ef (cc @jreback)

[TomAugspurger]

DtypeInfer:

+      10.3±0.4ms            4.78s   464.27  inference.DtypeInfer.time_timedelta64_2
+     8.71±0.07ms            2.93s   336.02  inference.DtypeInfer.time_datetime64
+        63.4±2ms            3.84s    60.54  inference.DtypeInfer.time_timedelta64_1

SeriesArthmetic.time_add_offset*

560±20ms       before           after         ratio
     [888c4ec3]       [36c309ed]
+     3.38±0.08ms          298±9ms    87.90  timeseries.SeriesArithmetic.time_add_offset_delta
+     10.00±0.3ms          294±3ms    29.37  timeseries.SeriesArithmetic.time_add_offset_fast
+        560±20ms          803±2ms     1.43  timeseries.SeriesArithmetic.time_add_offset_slow

These are handled by #17980

[TomAugspurger]

categoricals.Categoricals3.time_rank_string_cat_ordered

5.12±0.08ms       before           after         ratio
     [888c4ec3]       [36c309ed]
+      7.22±0.5ms          139±6ms    19.25  categoricals.Categoricals3.time_rank_string_cat

Seems like this is ranking the objects, instead of codes.

This is handled by #17982

[TomAugspurger]

packers.*stata.*

46.1±0.01ms       before           after         ratio
     [888c4ec3]       [36c309ed]
+      36.5±0.3ms          601±6ms    16.47  packers.packers_read_stata.time_packers_read_stata
+     46.1±0.01ms            575ms    12.48  packers.packers_read_stata_with_validation.time_packers_read_stata_with_validation
+      30.2±0.7ms            346ms    11.48  packers.STATA.time_write_stata
+          42.5ms            339ms     7.96  packers.STATA.time_write_stata_with_validation

Fixed by #17980

[TomAugspurger]

join_merge:

36.1±0.7ms       before           after         ratio
     [888c4ec3]       [36c309ed]
+      36.1±0.7ms          112±4ms     3.09  join_merge.Align.time_series_align_int64_index

Ah dang, http://pandas.pydata.org/speed/pandas/#join_merge.Align.time_series_align_int64_index points to the import delay PR: 2310faa1

I think something strange is going on w/ the benchmark... On 0.20.3 and 0.21.0rc1, I'm seeing the numexpr import affecting the benchmark time. The first run is always ~120ms. Subsequent runs are ~25-30ms. I'l come back to this later.

[TomAugspurger]

indexing.Int64Indexing.time_getitem.*

[100.00%] ··· Running indexing.Int64Indexing.time_getitem_slice                                                                                   46.8±0.3μs       before           after         ratio
     [888c4ec3]       [36c309ed]
+         622±7μs      1.57±0.02ms     2.53  indexing.Int64Indexing.time_getitem_array
+     3.73±0.07ms       7.46±0.1ms     2.00  indexing.Int64Indexing.time_getitem_lists
+         301±7μs          492±3μs     1.63  indexing.Int64Indexing.time_getitem_list_like

Looks like #17295 for this one http://pandas.pydata.org/speed/pandas/#join_merge.Align.time_series_align_int64_index

cc @jreback not sure if there's anything to be done there.

[TomAugspurger]

indexing.DataFrameIndexing.time_get_value

394±2ms       before           after         ratio
     [888c4ec3]       [36c309ed]
+     2.49±0.03μs       4.40±0.3μs     1.77  indexing.DataFrameIndexing.time_get_value

I believe this was due to 3fae8dd

[TomAugspurger]

-b timeseries.DatetimeIndex.time_infer_

            before           after         ratio
     [888c4ec3]       [36c309ed]
+      8.68±0.7ms      16.2±0.07ms     1.87  timeseries.DatetimeIndex.time_infer_freq_none
+      5.53±0.2ms      7.80±0.05ms     1.41  timeseries.DatetimeIndex.time_infer_freq_business

[TomAugspurger]

sparse.sparse_array_constructor.time_sparse_array_constructor_object_non_nan_fill_value_

+      32.4±0.2ms         52.9±3ms     1.63  sparse.sparse_array_constructor.time_sparse_array_constructor_object_non_nan_fill_value_10percent
+     19.2±0.07ms       29.8±0.2ms     1.55  sparse.sparse_array_constructor.time_sparse_array_constructor_object_non_nan_fill_value_1percent

[TomAugspurger]

indexing.IndexingMethods.time_take_d

+     2.49±0.04ms      3.89±0.06ms     1.56  indexing.IndexingMethods.time_take_intindex
+     2.48±0.04ms      3.84±0.05ms     1.55  indexing.IndexingMethods.time_take_dtindex

[TomAugspurger]

+     2.16±0.03ms      3.00±0.08ms     1.39  period.Algorithms.time_drop_duplicates_pseries
+     2.77±0.05ms      3.41±0.03ms     1.23  period.Algorithms.time_value_counts_pseries

[jbrockmendel]

For morale purposes: is there any good news?

[TomAugspurger]

:) Yeah there were quite a few:

-         148±2ms          135±2ms     0.91  gil.nogil_rolling_algos_slow.time_nogil_rolling_median
-     1.77±0.01ms      1.61±0.01ms     0.91  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BQuarterBegin', 1)
-        68.4±1ms       62.1±0.4ms     0.91  hdfstore_bench.HDF5.time_write_store_table_mixed
-     1.51±0.05ms      1.37±0.02ms     0.91  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('Week', 1)
-         359±6ms          325±2ms     0.90  io_bench.read_json_lines.time_read_json_lines
-      11.5±0.3μs      10.4±0.05μs     0.90  period.Algorithms.time_drop_duplicates_pindex
-     3.85±0.04ms      3.48±0.03ms     0.90  stat_ops.stat_ops_level_series_sum_multiple.time_stat_ops_level_series_sum_multiple
-         104±1μs       93.8±0.7μs     0.90  frame_ctor.frame_get_numeric_data.time_frame_get_numeric_data
-     1.73±0.04ms      1.56±0.01ms     0.90  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('QuarterEnd', 2)
-       145±0.9ms          131±1ms     0.90  inference.to_numeric_downcast.time_downcast('string-float', None)
-     1.79±0.02ms      1.61±0.01ms     0.90  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BMonthBegin', 2)
-     4.32±0.04ms      3.89±0.03ms     0.90  timeseries.ToDatetime.time_iso8601_format_no_sep
-     3.27±0.05ms      2.94±0.02ms     0.90  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('FY5253_1', 1)
-     3.25±0.07ms      2.91±0.03ms     0.90  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('FY5253_1', 2)
-     14.5±0.09ms      12.9±0.06ms     0.89  io_sql.ReadSQLTypes.time_float_read_table_sqlalchemy
-     9.26±0.08ms      8.25±0.02ms     0.89  inference.to_numeric.time_from_str_coerce
-      1.57±0.1ms      1.40±0.02ms     0.89  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BusinessDay', 1)
-     1.82±0.03ms      1.62±0.03ms     0.89  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BMonthBegin', 1)
-          17.2ms           15.2ms     0.89  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('FY5253Quarter_1', 2)
-         330±1ms          293±4ms     0.89  replace.replace_convert.time_replace_series_timedelta
-           3.09s            2.74s     0.89  stat_ops.FrameOps.time_op('median', True, 'float', 1)
-      47.4±0.2ms       41.9±0.3ms     0.88  frame_ctor.FromDicts.time_frame_ctor_nested_dict_int64
-      46.8±0.4ms       41.2±0.5ms     0.88  frame_ctor.FromDicts.time_frame_ctor_nested_dict
-     1.18±0.02ms      1.04±0.01ms     0.88  indexing.DataFrameIndexing.time_loc_dups
-         114±4ms        100±0.6ms     0.88  frame_methods.Apply.time_apply_user_func
-     4.40±0.02ms      3.86±0.01ms     0.88  timeseries.ToDatetime.time_iso8601
-         331±3ms          290±4ms     0.88  replace.replace_convert.time_replace_series_timestamp
-     1.76±0.01ms      1.54±0.03ms     0.88  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BYearBegin', 2)
-        255±10μs          224±2μs     0.88  frame_methods.frame_boolean_row_select.time_frame_boolean_row_select
-     1.51±0.04ms      1.32±0.02ms     0.88  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('YearEnd', 2)
-     1.81±0.03ms      1.59±0.03ms     0.88  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BQuarterBegin', 2)
-     1.74±0.02ms      1.52±0.02ms     0.87  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('QuarterEnd', 1)
-        71.1±4μs       62.1±0.6μs     0.87  frame_methods.frame_xs_row.time_frame_xs_row
-     4.40±0.07ms      3.85±0.02ms     0.87  timeseries.ToDatetime.time_iso8601_format
-           6.20s            5.41s     0.87  groupby.GroupBySuite.time_mad('int', 10000)
-     4.40±0.05ms      3.83±0.01ms     0.87  timeseries.ToDatetime.time_iso8601_nosep
-     1.49±0.02ms      1.29±0.03ms     0.87  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('YearEnd', 1)
-        92.5±2μs         80.2±2μs     0.87  period.Properties.time_end_time
-     1.50±0.02ms      1.30±0.02ms     0.86  groupby.groupby_categorical.time_groupby_ordered_sort
-     1.51±0.01ms      1.30±0.02ms     0.86  groupby.groupby_categorical.time_groupby_sort
-     2.00±0.04ms      1.73±0.01ms     0.86  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BusinessHour', 2)
-     1.77±0.05ms      1.52±0.03ms     0.86  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BQuarterEnd', 2)
-         544±5ms          468±2ms     0.86  replace.replace_convert.time_replace_frame_timestamp
-         543±4ms          465±3ms     0.86  replace.replace_convert.time_replace_frame_timedelta
-      7.00±0.1ms      6.00±0.06ms     0.86  timeseries.DatetimeAccessor.time_dt_accessor_normalize
-     2.38±0.03ms      2.03±0.03ms     0.86  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('CustomBusinessHour', 1)
-         393±3μs          335±3μs     0.85  join_merge.Append.time_append_homogenous
-     1.82±0.06ms      1.55±0.04ms     0.85  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('CDay', 2)
-     2.42±0.04ms      2.06±0.04ms     0.85  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('CustomBusinessHour', 2)
-     1.54±0.01ms      1.29±0.02ms     0.84  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('YearBegin', 1)
-     1.87±0.06ms      1.57±0.02ms     0.84  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BQuarterEnd', 1)
-         1.05±0s          873±2ms     0.83  groupby.groupby_frame_apply.time_groupby_frame_df_copy_function
-         421±2ms          347±4ms     0.83  groupby.groupby_frame_apply.time_groupby_frame_apply_df_copy_overhead
-         377±1μs          308±2μs     0.82  algorithms.Algorithms.time_match_strings
-          71.2ms           58.0ms     0.81  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('FY5253Quarter_2', 2)
-         581±7μs        472±0.8μs     0.81  frame_methods.Iteration.time_iteritems_cached
-     1.69±0.04ms      1.35±0.01ms     0.80  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('Easter', 2)
-      26.3±0.1ms       21.0±0.5ms     0.80  sparse.sparse_series_to_coo.time_sparse_series_to_coo
-          43.8ms           35.0ms     0.80  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('FY5253Quarter_2', 1)
-           11.1s            8.85s     0.79  panel_methods.PanelMethods.time_pct_change_items
-      2.20±0.2ms      1.74±0.04ms     0.79  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('BusinessHour', 1)
-      12.5±0.2μs      9.72±0.09μs     0.78  period.period_standard_indexing.time_get_loc
-         204±2μs        156±0.9μs     0.76  timeseries.Offsets.time_custom_bmonthend_incr
-      13.1±0.3ms       9.91±0.1ms     0.76  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('FY5253_2', 1)
-      22.5±0.5ms       16.9±0.2ms     0.75  categoricals.Categoricals2.time_value_counts_dropna
-     2.96±0.02ms      2.21±0.05ms     0.75  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('CBMonthBegin', 2)
-        966±10μs          717±7μs     0.74  join_merge.Append.time_append_mixed
-      1.85±0.1ms      1.37±0.01ms     0.74  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('Easter', 1)
-      3.67±0.2ms      2.71±0.02ms     0.74  groupby.GroupBySuite.time_last('float', 10000)
-     4.56±0.07ms      3.37±0.05ms     0.74  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('CBMonthEnd', 1)
-      13.7±0.2ms      10.1±0.05ms     0.73  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('FY5253_2', 2)
-         848±9ns         620±10ns     0.73  timestamp.TimestampOps.time_to_pydatetime
-      4.72±0.2ms      3.45±0.08ms     0.73  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('CBMonthEnd', 2)
-     2.55±0.07ms           1.86ms     0.73  inference.DtypeInfer.time_float64
-      20.4±0.3ms       14.6±0.3ms     0.72  categoricals.Categoricals2.time_value_counts
-      2.58±0.3ms           1.81ms     0.70  inference.DtypeInfer.time_int64
-      32.7±0.8ms       22.6±0.7ms     0.69  binary_ops.Ops.time_frame_comparison(False, 1)
-         272±6μs          187±6μs     0.69  timeseries.Offsets.time_custom_bmonthbegin_incr_n
-      16.1±0.1ms      11.0±0.05ms     0.68  groupby.groupby_categorical.time_groupby_ordered_nosort
-         256±6μs          174±2μs     0.68  timeseries.Offsets.time_custom_bmonthend_incr_n
-      57.7±0.6ms       38.5±0.7ms     0.67  join_merge.Concat.time_concat_small_frames
-     3.95±0.05ms      2.63±0.09ms     0.67  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('CBMonthBegin', 1)
-     1.35±0.04ms         891±20μs     0.66  replace.replace_replacena.time_replace_replacena
-        32.5±2ms         21.4±1ms     0.66  binary_ops.Ops.time_frame_comparison(False, 'default')
-      10.7±0.1ms       6.98±0.1ms     0.65  groupby.groupby_categorical.time_groupby_nosort
-      3.68±0.2ms      2.39±0.03ms     0.65  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('WeekOfMonth', 2)
-           8.77s            5.47s     0.62  panel_methods.PanelMethods.time_pct_change_minor
-         297±2μs          181±2μs     0.61  timeseries.Offsets.time_custom_bmonthend_decr_n
-           9.15s            5.54s     0.61  panel_methods.PanelMethods.time_pct_change_major
-     8.82±0.06ms      5.31±0.05ms     0.60  groupby.groupby_categorical.time_groupby_extra_cat_nosort
-      4.04±0.3ms      2.42±0.06ms     0.60  frame_ctor.FrameConstructorDTIndexFromOffsets.time_frame_ctor('WeekOfMonth', 1)
-     1.84±0.01μs      1.09±0.02μs     0.59  period.Properties.time_is_leap_year
-        67.0±1ms       39.4±0.1ms     0.59  eval.Eval.time_chained_cmp('python', 'all')
-      28.5±0.6ms       16.6±0.5ms     0.58  frame_methods.frame_fancy_lookup.time_frame_fancy_lookup_all
-      26.1±0.2μs       14.7±0.2μs     0.56  timeseries.Offsets.time_timeseries_year_incr
-           8.89y            5.00y     0.56  io_bench.read_json_lines.peakmem_read_json_lines
-         266±3μs          147±3μs     0.55  timeseries.Offsets.time_custom_bmonthbegin_decr_n
-     1.82±0.02μs      1.00±0.01μs     0.55  period.Properties.time_daysinmonth
-     1.82±0.03μs          977±8ns     0.54  period.Properties.time_year
-     1.80±0.03μs         959±10ns     0.53  period.Properties.time_second
-     1.80±0.01μs         950±10ns     0.53  period.Properties.time_dayofweek
-     1.83±0.01μs         956±10ns     0.52  period.Properties.time_dayofyear
-     1.82±0.02μs         948±30ns     0.52  period.PeriodProperties.time_hour
-     1.79±0.01μs          931±3ns     0.52  period.Properties.time_month
-     1.82±0.01μs          942±7ns     0.52  period.PeriodProperties.time_second
-     1.81±0.03μs         925±20ns     0.51  period.Properties.time_quarter
-     1.83±0.02μs         933±10ns     0.51  period.PeriodProperties.time_day
-     1.85±0.01μs         943±20ns     0.51  period.PeriodProperties.time_quarter
-     1.81±0.02μs         919±10ns     0.51  period.Properties.time_minute
-     1.90±0.03μs         965±10ns     0.51  period.PeriodProperties.time_year
-     1.85±0.02μs         936±20ns     0.51  period.PeriodProperties.time_month
-      22.6±0.3μs      11.4±0.07μs     0.50  timeseries.Offsets.time_timeseries_year_apply
-     1.85±0.03μs          929±5ns     0.50  period.Properties.time_day
-     2.00±0.05μs         999±10ns     0.50  period.Properties.time_week
-     1.90±0.02μs         939±10ns     0.49  period.Properties.time_qyear
-     1.90±0.02μs         930±10ns     0.49  period.PeriodProperties.time_minute
-     1.96±0.03μs         912±10ns     0.47  period.Properties.time_hour
-     15.2±0.07ms      6.96±0.05ms     0.46  binary_ops.Ops.time_frame_comparison(True, 'default')
-       321±0.8ms          143±1ms     0.44  sparse.sparse_series_to_frame.time_sparse_series_to_frame
-      49.4±0.4ms       21.3±0.2ms     0.43  eval.Eval.time_chained_cmp('python', 1)
-     8.61±0.01ms      3.68±0.04ms     0.43  categoricals.Categoricals.time_union
-     11.1±0.05ms      2.74±0.06ms     0.25  binary_ops.Ops.time_frame_comparison(True, 1)
-        79.5±1ms       14.8±0.2ms     0.19  categoricals.Categoricals2.time_set_categories
-        3.45±0ms          439±3μs     0.13  timeseries.DatetimeIndex.time_timeseries_is_month_start
-           2.47s          231±2ms     0.09  sparse.sparse_frame_constructor.time_sparse_from_scipy
-     7.11±0.03μs          482±6ns     0.07  timestamp.TimestampProperties.time_weekday_name
-           2.46s          106±1ms     0.04  sparse.sparse_frame_constructor.time_sparse_from_dict
-        915±20μs       23.5±0.7μs     0.03  timeseries.DatetimeAccessor.time_dt_accessor
-       105±0.4ms      2.38±0.04ms     0.02  rolling.SeriesRolling.time_rolling_quantile_0
-         116±1ms      2.36±0.01ms     0.02  rolling.SeriesRolling.time_rolling_quantile_1
-       101±0.6ms      1.98±0.03ms     0.02  rolling.DataframeRolling.time_rolling_quantile_0
-      31.6±0.4μs          611±2ns     0.02  timestamp.TimestampOps.time_to_pydatetime_tz
-         123±2ms      1.99±0.03ms     0.02  rolling.DataframeRolling.time_rolling_quantile_1
-     10.1±0.09μs          147±1ns     0.01  timestamp.TimestampProperties.time_microsecond
-         197±1ms      2.55±0.03ms     0.01  rolling.SeriesRolling.time_rolling_quantile_0_l
-         201±1ms      2.13±0.04ms     0.01  rolling.DataframeRolling.time_rolling_quantile_0_l
-         260±1ms      2.49±0.04ms     0.01  rolling.SeriesRolling.time_rolling_quantile_1_l
-         263±2ms      2.17±0.02ms     0.01  rolling.DataframeRolling.time_rolling_quantile_1_l
-     3.51±0.02ms      4.59±0.04μs     0.00  hdfstore_bench.HDF5.time_store_repr
-     3.46±0.03ms      4.38±0.07μs     0.00  hdfstore_bench.HDF5.time_store_str
-      15.9±0.2ms      1.37±0.01μs     0.00  index_object.Range.time_min
-      16.1±0.3ms      1.35±0.01μs     0.00  index_object.Range.time_max
-      15.8±0.1ms      1.26±0.02μs     0.00  index_object.Range.time_max_trivial
-      16.0±0.1ms      1.23±0.02μs     0.00  index_object.Range.time_min_trivial

[jbrockmendel]

-b timeseries.DatetimeIndex.time_infer_

These may be improved by #17830 and/or #17746.

[TomAugspurger]

For period.Algorithms.time_drop_duplicates_pseries, it seems like Period.freqstr is being called somewhere inside _ensure_data / ensure_object. Could that be avoided?

[jbrockmendel]

Are you sure its freqstr in particular? It seems like drop_duplicates would make a bunch of calls to Period.__eq__, which then calls self.freq.__eq__(other.freq) which is very slow.

[TomAugspurger]

I think it's in Period.__hash__, but confirming now.

[TomAugspurger]

Hmm that may be it:

diff --git a/pandas/_libs/period.pyx b/pandas/_libs/period.pyx
index cf6ef91d3..f28ee2aac 100644
--- a/pandas/_libs/period.pyx
+++ b/pandas/_libs/period.pyx
@@ -718,7 +718,7 @@ cdef class _Period(object):
                             (type(self).__name__, type(other).__name__))

     def __hash__(self):
-        return hash((self.ordinal, self.freqstr))
+        return hash((self.ordinal, self.freq))

     def _add_delta(self, other):
         if isinstance(other, (timedelta, np.timedelta64, offsets.Tick)):
(

master:

In [1]: import pandas as pd

In [2]: p = pd.Period("2010", freq="M")

In [3]: %timeit hash(p)
807 ns ± 8.42 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

with that diff:

In [1]: import pandas as pd

In [2]: p = pd.Period("2010", freq="M")

In [3]: %timeit hash(p)
4.84 µs ± 42.1 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

Let's see if that breaks anything.

[jbrockmendel]

Can you try that with a handful of different frequencies? I'd be shocked if hash(freq) is faster than hash(freq.freqstr)

[TomAugspurger]

FWIW, it doesn't seem promising w.r.t. the benchmark (see the 2nd and 4th) lines. I rebuilt the c extensions between these two. My branch with the new hash is on top, master is on bottom.

 ◰³ pandas-dev  ~/E/p/l/p/s/pandas   period-freq-hash  asv_bench  asv dev -b 'period.Algorithms.time_[drop|value]'    4584ms  Wed Oct 25 14:12:40 2017
· Discovering benchmarks
· Running 4 total benchmarks (1 commits * 1 environments * 4 benchmarks)
[  0.00%] ·· Building for existing-py_Users_taugspurger_Envs_pandas-dev_bin_python3.6
[  0.00%] ·· Benchmarking existing-py_Users_taugspurger_Envs_pandas-dev_bin_python3.6
[ 25.00%] ··· Running period.Algorithms.time_drop_duplicates_pindex                                                                                   49.0μs
[ 50.00%] ··· Running period.Algorithms.time_drop_duplicates_pseries                                                                                  18.5ms
[ 75.00%] ··· Running period.Algorithms.time_value_counts_pindex                                                                                       402μs
[100.00%] ··· Running period.Algorithms.time_value_counts_pseries                                                                                     22.8ms
 ◰³ pandas-dev  ~/E/p/l/p/s/pandas   period-freq-hash  asv_bench  asv dev -b 'period.Algorithms.time_[drop|value]'        6s  Wed Oct 25 14:13:37 2017
· Discovering benchmarks
· Running 4 total benchmarks (1 commits * 1 environments * 4 benchmarks)
[  0.00%] ·· Building for existing-py_Users_taugspurger_Envs_pandas-dev_bin_python3.6
[  0.00%] ·· Benchmarking existing-py_Users_taugspurger_Envs_pandas-dev_bin_python3.6
[ 25.00%] ··· Running period.Algorithms.time_drop_duplicates_pindex                                                                                   48.0μs
[ 50.00%] ··· Running period.Algorithms.time_drop_duplicates_pseries                                                                                  3.14ms
[ 75.00%] ··· Running period.Algorithms.time_value_counts_pindex                                                                                       405μs
[100.00%] ··· Running period.Algorithms.time_value_counts_pseries                                                                                     3.43ms

[TomAugspurger]

@jreback if you have time to look at the indexing ones, I'd appreciate it: #17861 (comment) and #17861 (comment)

I'm not sure there's a whole lot to be done though. The first is because we have to check for missing values with list-like indexers now, to raise the warning.

[TomAugspurger]

Oh, I'm dumb. Mixed up nanoseconds and microseconds in #17861 (comment)

[TomAugspurger]

I think we decided in #17574 that the sparse changes were OK?

[jorisvandenbossche]

@jreback if you have time to look at the indexing ones, I'd appreciate it: #17861 (comment) and #17861 (comment)

The series getitem ones (s[np.arange(10000)]) are due to changing reindex to loc, and apparently loc is slower.

[TomAugspurger]

I think everything has been addressed.