REF: remove libreduction.apply_frame_axis0 by jbrockmendel · Pull Request #42992 · pandas-dev/pandas (original) (raw)

time asv continuous -E virtualenv -f 1.01 --record-samples --append-samples master HEAD -b groupby
[...]
       before           after         ratio
     [860ff03e]       [83dacf98]
     <master>       <ref-libreduction>
+      5.27±0.4ms         12.1±1ms     2.29  groupby.Apply.time_scalar_function_single_col(4)
+      16.1±0.8ms         34.8±2ms     2.16  groupby.Apply.time_scalar_function_multi_col(4)
+         141±5μs          146±5μs     1.03  groupby.GroupByMethods.time_dtype_as_group('float', 'head', 'transformation')
+         129±7μs          133±8μs     1.03  groupby.GroupByMethods.time_dtype_as_group('float', 'cumcount', 'transformation')
+         130±5μs          133±6μs     1.03  groupby.GroupByMethods.time_dtype_as_group('float', 'cumcount', 'direct')
+         159±7μs          163±6μs     1.03  groupby.GroupByMethods.time_dtype_as_group('uint', 'cumprod', 'transformation')
+         142±6μs          146±4μs     1.03  groupby.GroupByMethods.time_dtype_as_group('float', 'head', 'direct')
+        61.4±3μs         62.8±4μs     1.02  groupby.GroupByMethods.time_dtype_as_group('float', 'cummax', 'transformation')
+        53.7±3μs         54.9±4μs     1.02  groupby.GroupByMethods.time_dtype_as_field('uint', 'any', 'direct')
+        74.2±7μs         75.9±2μs     1.02  groupby.GroupByMethods.time_dtype_as_group('uint', 'count', 'transformation')
+         195±6μs          199±5μs     1.02  groupby.GroupByMethods.time_dtype_as_group('float', 'sum', 'transformation')
+        94.2±4μs         96.1±3μs     1.02  groupby.GroupByMethods.time_dtype_as_group('uint', 'last', 'transformation')
+        54.8±2μs         55.8±1μs     1.02  groupby.GroupByMethods.time_dtype_as_group('uint', 'any', 'direct')
+        94.0±3μs         95.6±2μs     1.02  groupby.GroupByMethods.time_dtype_as_group('uint', 'last', 'direct')
+         308±9μs         312±20μs     1.01  groupby.GroupByMethods.time_dtype_as_group('uint', 'sem', 'direct')
+         156±5μs         158±10μs     1.01  groupby.GroupByMethods.time_dtype_as_group('float', 'bfill', 'transformation')
-         281±6ms         277±20ms     0.99  groupby.AggEngine.time_dataframe_numba(False)
-        511±10ms         504±20ms     0.99  groupby.AggEngine.time_series_numba(True)
-        511±20ms         503±20ms     0.99  groupby.AggEngine.time_dataframe_numba(True)
-      1.58±0.2ms      1.55±0.03ms     0.98  rolling.ExpandingMethods.time_expanding_groupby('Series', 'float', 'median')
-      1.96±0.2ms      1.92±0.06ms     0.98  groupby.CountMultiInt.time_multi_int_count
-        60.4±1ms       59.1±0.9ms     0.98  groupby.GroupByMethods.time_dtype_as_field('uint', 'skew', 'direct')
-        122±10ms          119±3ms     0.98  groupby.MultiColumn.time_lambda_sum
-        282±10ms         276±10ms     0.98  groupby.AggEngine.time_series_numba(False)
-        20.2±2ms       19.4±0.7ms     0.96  groupby.AggEngine.time_dataframe_cython(False)
-        39.7±2ms         37.8±4ms     0.95  groupby.Apply.time_copy_function_multi_col(5)
-     1.23±0.09ms      1.06±0.04ms     0.86  groupby.FillNA.time_df_bfill
-      1.23±0.1ms      1.03±0.09ms     0.84  groupby.FillNA.time_df_ffill
-        397±10ms          223±7ms     0.56  groupby.Apply.time_copy_overhead_single_col(4)
-      1.20±0.03s         604±30ms     0.50  groupby.Apply.time_copy_function_multi_col(4)

Updated with a larger sample size, looks more reasonable