diff options
Diffstat (limited to 'benchmarks')
| -rw-r--r-- | benchmarks/benchmarks/bench_core.py | 4 | ||||
| -rw-r--r-- | benchmarks/benchmarks/bench_function_base.py | 49 | ||||
| -rw-r--r-- | benchmarks/benchmarks/bench_lib.py | 103 | ||||
| -rw-r--r-- | benchmarks/benchmarks/bench_linalg.py | 2 | ||||
| -rw-r--r-- | benchmarks/benchmarks/bench_overrides.py | 16 | ||||
| -rw-r--r-- | benchmarks/benchmarks/bench_records.py | 43 | ||||
| -rw-r--r-- | benchmarks/benchmarks/bench_ufunc.py | 2 |
7 files changed, 185 insertions, 34 deletions
diff --git a/benchmarks/benchmarks/bench_core.py b/benchmarks/benchmarks/bench_core.py index 26cffcab1..194ce3218 100644 --- a/benchmarks/benchmarks/bench_core.py +++ b/benchmarks/benchmarks/bench_core.py @@ -97,8 +97,8 @@ class Temporaries(Benchmark): class CorrConv(Benchmark): - params = [[50, 1000, 1e5], - [10, 100, 1000, 1e4], + params = [[50, 1000, int(1e5)], + [10, 100, 1000, int(1e4)], ['valid', 'same', 'full']] param_names = ['size1', 'size2', 'mode'] diff --git a/benchmarks/benchmarks/bench_function_base.py b/benchmarks/benchmarks/bench_function_base.py index 9ef03262b..64e578680 100644 --- a/benchmarks/benchmarks/bench_function_base.py +++ b/benchmarks/benchmarks/bench_function_base.py @@ -96,35 +96,46 @@ class Select(Benchmark): class Sort(Benchmark): - def setup(self): - self.e = np.arange(10000, dtype=np.float32) - self.o = np.arange(10001, dtype=np.float32) + params = [ + ['quick', 'merge', 'heap'], + ['float32', 'int32', 'uint32'] + ] + param_names = ['kind', 'dtype'] + + def setup(self, kind, dtype): + self.e = np.arange(10000, dtype=dtype) + self.o = np.arange(10001, dtype=dtype) np.random.seed(25) np.random.shuffle(self.o) # quicksort implementations can have issues with equal elements - self.equal = np.ones(10000) - self.many_equal = np.sort(np.arange(10000) % 10) + self.equal = np.ones(10000, dtype=dtype) + self.many_equal = np.sort(np.arange(10000) % 10).astype(dtype) + + try: + np.sort(self.e, kind=kind) + except TypeError: + raise NotImplementedError() - def time_sort(self): - np.sort(self.e) + def time_sort(self, kind, dtype): + np.sort(self.e, kind=kind) - def time_sort_random(self): - np.sort(self.o) + def time_sort_random(self, kind, dtype): + np.sort(self.o, kind=kind) - def time_sort_inplace(self): - self.e.sort() + def time_sort_inplace(self, kind, dtype): + self.e.sort(kind=kind) - def time_sort_equal(self): - self.equal.sort() + def time_sort_equal(self, kind, dtype): + self.equal.sort(kind=kind) - def time_sort_many_equal(self): - self.many_equal.sort() + def time_sort_many_equal(self, kind, dtype): + self.many_equal.sort(kind=kind) - def time_argsort(self): - self.e.argsort() + def time_argsort(self, kind, dtype): + self.e.argsort(kind=kind) - def time_argsort_random(self): - self.o.argsort() + def time_argsort_random(self, kind, dtype): + self.o.argsort(kind=kind) class SortWorst(Benchmark): diff --git a/benchmarks/benchmarks/bench_lib.py b/benchmarks/benchmarks/bench_lib.py index e6c91a27c..f65a96dad 100644 --- a/benchmarks/benchmarks/bench_lib.py +++ b/benchmarks/benchmarks/bench_lib.py @@ -9,20 +9,109 @@ import numpy as np class Pad(Benchmark): - """Benchmarks for `numpy.pad`.""" + """Benchmarks for `numpy.pad`. + + When benchmarking the pad function it is useful to cover scenarios where + the ratio between the size of the input array and the output array differs + significantly (original area vs. padded area). This allows to evaluate for + which scenario a padding algorithm is optimized. Furthermore involving + large range of array sizes ensures that the effects of CPU-bound caching is + visible. + + The table below shows the sizes of the arrays involved in this benchmark: + + +-----------------+----------+-----------+-----------+-----------------+ + | shape | original | padded: 1 | padded: 8 | padded: (0, 32) | + +=================+==========+===========+===========+=================+ + | (2 ** 22,) | 32 MiB | 32.0 MiB | 32.0 MiB | 32.0 MiB | + +-----------------+----------+-----------+-----------+-----------------+ + | (1024, 1024) | 8 MiB | 8.03 MiB | 8.25 MiB | 8.51 MiB | + +-----------------+----------+-----------+-----------+-----------------+ + | (256, 256, 1) | 256 KiB | 786 KiB | 5.08 MiB | 11.6 MiB | + +-----------------+----------+-----------+-----------+-----------------+ + | (4, 4, 4, 4) | 2 KiB | 10.1 KiB | 1.22 MiB | 12.8 MiB | + +-----------------+----------+-----------+-----------+-----------------+ + | (1, 1, 1, 1, 1) | 8 B | 1.90 MiB | 10.8 MiB | 299 MiB | + +-----------------+----------+-----------+-----------+-----------------+ + """ param_names = ["shape", "pad_width", "mode"] params = [ - [(1000,), (10, 100), (10, 10, 10)], - [1, 3, (0, 5)], + # Shape of the input arrays + [(2 ** 22,), (1024, 1024), (256, 128, 1), + (4, 4, 4, 4), (1, 1, 1, 1, 1)], + # Tested pad widths + [1, 8, (0, 32)], + # Tested modes: mean, median, minimum & maximum use the same code path + # reflect & symmetric share a lot of their code path ["constant", "edge", "linear_ramp", "mean", "reflect", "wrap"], ] def setup(self, shape, pad_width, mode): - # avoid np.zeros or np.empty's lazy allocation. - # np.full causes pagefaults to occur during setup - # instead of during the benchmark - self.array = np.full(shape, 0) + # Make sure to fill the array to make the OS page fault + # in the setup phase and not the timed phase + self.array = np.full(shape, fill_value=1, dtype=np.float64) def time_pad(self, shape, pad_width, mode): np.pad(self.array, pad_width, mode) + +class Nan(Benchmark): + """Benchmarks for nan functions""" + + param_names = ["array_size", "percent_nans"] + params = [ + # sizes of the 1D arrays + [200, int(2e5)], + # percent of np.nan in arrays + [0, 0.1, 2., 50., 90.], + ] + + def setup(self, array_size, percent_nans): + np.random.seed(123) + # produce a randomly shuffled array with the + # approximate desired percentage np.nan content + base_array = np.random.uniform(size=array_size) + base_array[base_array < percent_nans / 100.] = np.nan + self.arr = base_array + + def time_nanmin(self, array_size, percent_nans): + np.nanmin(self.arr) + + def time_nanmax(self, array_size, percent_nans): + np.nanmax(self.arr) + + def time_nanargmin(self, array_size, percent_nans): + np.nanargmin(self.arr) + + def time_nanargmax(self, array_size, percent_nans): + np.nanargmax(self.arr) + + def time_nansum(self, array_size, percent_nans): + np.nansum(self.arr) + + def time_nanprod(self, array_size, percent_nans): + np.nanprod(self.arr) + + def time_nancumsum(self, array_size, percent_nans): + np.nancumsum(self.arr) + + def time_nancumprod(self, array_size, percent_nans): + np.nancumprod(self.arr) + + def time_nanmean(self, array_size, percent_nans): + np.nanmean(self.arr) + + def time_nanvar(self, array_size, percent_nans): + np.nanvar(self.arr) + + def time_nanstd(self, array_size, percent_nans): + np.nanstd(self.arr) + + def time_nanmedian(self, array_size, percent_nans): + np.nanmedian(self.arr) + + def time_nanquantile(self, array_size, percent_nans): + np.nanquantile(self.arr, q=0.2) + + def time_nanpercentile(self, array_size, percent_nans): + np.nanpercentile(self.arr, q=50) diff --git a/benchmarks/benchmarks/bench_linalg.py b/benchmarks/benchmarks/bench_linalg.py index a65d510be..5c44162a2 100644 --- a/benchmarks/benchmarks/bench_linalg.py +++ b/benchmarks/benchmarks/bench_linalg.py @@ -106,4 +106,4 @@ class Lstsq(Benchmark): self.b = get_indexes_rand()[:100].astype(np.float64) def time_numpy_linalg_lstsq_a__b_float64(self): - np.linalg.lstsq(self.a, self.b) + np.linalg.lstsq(self.a, self.b, rcond=-1) diff --git a/benchmarks/benchmarks/bench_overrides.py b/benchmarks/benchmarks/bench_overrides.py index 2cb94c95c..58572d07d 100644 --- a/benchmarks/benchmarks/bench_overrides.py +++ b/benchmarks/benchmarks/bench_overrides.py @@ -2,7 +2,15 @@ from __future__ import absolute_import, division, print_function from .common import Benchmark -from numpy.core.overrides import array_function_dispatch +try: + from numpy.core.overrides import array_function_dispatch +except ImportError: + # Don't fail at import time with old Numpy versions + def array_function_dispatch(*args, **kwargs): + def wrap(*args, **kwargs): + return None + return wrap + import numpy as np @@ -16,10 +24,10 @@ def mock_broadcast_to(array, shape, subok=False): def _concatenate_dispatcher(arrays, axis=None, out=None): - for array in arrays: - yield array if out is not None: - yield out + arrays = list(arrays) + arrays.append(out) + return arrays @array_function_dispatch(_concatenate_dispatcher) diff --git a/benchmarks/benchmarks/bench_records.py b/benchmarks/benchmarks/bench_records.py new file mode 100644 index 000000000..41a6dd775 --- /dev/null +++ b/benchmarks/benchmarks/bench_records.py @@ -0,0 +1,43 @@ +from __future__ import absolute_import, division, print_function +import os + +from .common import Benchmark + +import numpy as np + + +class Records(Benchmark): + def setup(self): + self.l50 = np.arange(1000) + self.fields_number = 10000 + self.arrays = [self.l50 for _ in range(self.fields_number)] + self.formats = [self.l50.dtype.str for _ in range(self.fields_number)] + self.formats_str = ','.join(self.formats) + self.dtype_ = np.dtype( + [ + ('field_{}'.format(i), self.l50.dtype.str) + for i in range(self.fields_number) + ] + ) + self.buffer = self.l50.tostring() * self.fields_number + + def time_fromarrays_w_dtype(self): + np.core.records.fromarrays(self.arrays, dtype=self.dtype_) + + def time_fromarrays_wo_dtype(self): + np.core.records.fromarrays(self.arrays) + + def time_fromarrays_formats_as_list(self): + np.core.records.fromarrays(self.arrays, formats=self.formats) + + def time_fromarrays_formats_as_string(self): + np.core.records.fromarrays(self.arrays, formats=self.formats_str) + + def time_fromstring_w_dtype(self): + np.core.records.fromstring(self.buffer, dtype=self.dtype_) + + def time_fromstring_formats_as_list(self): + np.core.records.fromstring(self.buffer, formats=self.formats) + + def time_fromstring_formats_as_string(self): + np.core.records.fromstring(self.buffer, formats=self.formats_str) diff --git a/benchmarks/benchmarks/bench_ufunc.py b/benchmarks/benchmarks/bench_ufunc.py index a7e385f70..62e70782d 100644 --- a/benchmarks/benchmarks/bench_ufunc.py +++ b/benchmarks/benchmarks/bench_ufunc.py @@ -15,7 +15,7 @@ ufuncs = ['abs', 'absolute', 'add', 'arccos', 'arccosh', 'arcsin', 'arcsinh', 'isinf', 'isnan', 'isnat', 'lcm', 'ldexp', 'left_shift', 'less', 'less_equal', 'log', 'log10', 'log1p', 'log2', 'logaddexp', 'logaddexp2', 'logical_and', 'logical_not', 'logical_or', - 'logical_xor', 'maximum', 'minimum', 'mod', 'modf', 'multiply', + 'logical_xor', 'matmul', 'maximum', 'minimum', 'mod', 'modf', 'multiply', 'negative', 'nextafter', 'not_equal', 'positive', 'power', 'rad2deg', 'radians', 'reciprocal', 'remainder', 'right_shift', 'rint', 'sign', 'signbit', 'sin', 'sinh', 'spacing', 'sqrt', |