Closed jorisvandenbossche closed 1 year ago
I did a full benchmark run on a dedicated machine comparing v1.2.0rc0 with v1.1.5.
The top results:
[b5958ee1] [7688d3cf] <v1.1.5^0> <v1.2.0rc0^0> + 4.25±0.05μs 366±9ms 86187.42 index_object.IndexEquals.time_non_object_equals_multiindex + 50.2±4μs 104±5ms 2075.23 indexing.NumericSeriesIndexing.time_getitem_scalar(<class 'pandas.core.indexes.numeric.Float64Index'>, 'nonunique_monotonic_inc') + 120±8μs 102±5ms 848.38 indexing.NumericSeriesIndexing.time_loc_scalar(<class 'pandas.core.indexes.numeric.Float64Index'>, 'nonunique_monotonic_inc') + 268±20μs 103±5ms 383.87 indexing.NumericSeriesIndexing.time_loc_slice(<class 'pandas.core.indexes.numeric.Float64Index'>, 'nonunique_monotonic_inc') + 289±40μs 111±8ms 382.15 hash_functions.NumericSeriesIndexing.time_loc_slice(<class 'pandas.core.indexes.numeric.Float64Index'>, 1000000) + 273±20μs 102±5ms 374.69 indexing.NumericSeriesIndexing.time_getitem_slice(<class 'pandas.core.indexes.numeric.Float64Index'>, 'nonunique_monotonic_inc') + 1.75±0.1ms 103±5ms 58.77 indexing.NumericSeriesIndexing.time_loc_list_like(<class 'pandas.core.indexes.numeric.Float64Index'>, 'nonunique_monotonic_inc') + 2.11±0.1ms 103±5ms 48.69 indexing.NumericSeriesIndexing.time_getitem_list_like(<class 'pandas.core.indexes.numeric.Float64Index'>, 'nonunique_monotonic_inc') + 7.38±0.7ms 109±7ms 14.80 hash_functions.NumericSeriesIndexingShuffled.time_loc_slice(<class 'pandas.core.indexes.numeric.Float64Index'>, 1000000) + 3.66±0.2μs 13.4±0.4μs 3.66 index_cached_properties.IndexCache.time_is_all_dates('Float64Index') + 336±2ms 1.22±0s 3.65 groupby.TransformEngine.time_series_numba(True) + 287±2ms 1.02±0s 3.55 groupby.AggEngine.time_series_numba(True) + 289±2ms 1.02±0s 3.52 groupby.AggEngine.time_dataframe_numba(True) + 3.64±0.2μs 12.8±0.5μs 3.52 index_cached_properties.IndexCache.time_is_all_dates('IntervalIndex') + 1.89±0.09μs 6.49±0.1μs 3.44 index_cached_properties.IndexCache.time_is_all_dates('PeriodIndex') + 3.80±0.2μs 12.7±0.4μs 3.35 index_cached_properties.IndexCache.time_is_all_dates('UInt64Index') + 3.27±0.1μs 10.9±0.2μs 3.33 index_cached_properties.IndexCache.time_is_all_dates('MultiIndex') + 728±30ns 2.39±0.06μs 3.28 index_cached_properties.IndexCache.time_is_all_dates('RangeIndex') + 2.02±0.09μs 6.49±0.1μs 3.21 index_cached_properties.IndexCache.time_is_all_dates('DatetimeIndex') + 719±30ns 2.27±0.06μs 3.15 index_cached_properties.IndexCache.time_is_all_dates('Int64Index') + 4.11±0.2μs 12.4±0.3μs 3.02 index_cached_properties.IndexCache.time_is_all_dates('TimedeltaIndex') + 430±3ms 1.23±0s 2.85 groupby.TransformEngine.time_dataframe_numba(True) + 7.44±0.08ms 17.1±2ms 2.29 hash_functions.UniqueAndFactorizeArange.time_unique(6) + 75.3±1ms 172±1ms 2.28 hash_functions.IsinWithArangeSorted.time_isin(<class 'numpy.float64'>, 1000000) + 7.64±0.04ms 17.1±2ms 2.24 hash_functions.UniqueAndFactorizeArange.time_unique(5) + 1.62±0.04ms 3.51±0.02ms 2.16 arithmetic.Timeseries.time_series_timestamp_compare(None) + 1.60±0.03ms 3.45±0.04ms 2.16 arithmetic.Timeseries.time_timestamp_series_compare(None) + 1.62±0.03ms 3.49±0.09ms 2.16 arithmetic.Timeseries.time_timestamp_series_compare('US/Eastern') + 10.9±0.7ms 23.2±1ms 2.14 hash_functions.UniqueAndFactorizeArange.time_factorize(6) + 11.1±0.5ms 23.2±1ms 2.10 hash_functions.UniqueAndFactorizeArange.time_factorize(5) + 1.63±0.04ms 3.41±0.03ms 2.09 arithmetic.Timeseries.time_series_timestamp_compare('US/Eastern') + 4.11±0.1ms 8.52±0.09ms 2.08 index_object.SetDisjoint.time_datetime_difference_disjoint + 74.5±0.3ms 153±1ms 2.05 replace.ReplaceList.time_replace_list_one_match(True) + 124±0.9ms 248±0.4ms 2.00 gil.ParallelDatetimeFields.time_datetime_to_period
The first, biggest regression in the IndexEquals benchmark is probably already fixed in https://github.com/pandas-dev/pandas/pull/38560
IndexEquals
I did a full benchmark run on a dedicated machine comparing v1.2.0rc0 with v1.1.5.
The top results:
The first, biggest regression in the
IndexEquals
benchmark is probably already fixed in https://github.com/pandas-dev/pandas/pull/38560Full results:
``` before after ratio [b5958ee1] [7688d3cf]