1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
|
import matplotlib
matplotlib.use("agg")
from collections import OrderedDict
import contextlib
import io
from matplotlib import pyplot as plt
import numpy as np
import os
import pyperf
import subprocess
import sys
@contextlib.contextmanager
def stdoutIO(stdout=None):
old = sys.stdout
if stdout is None:
stdout = io.StringIO()
sys.stdout = stdout
yield stdout
sys.stdout = old
def run_perf(args, json_name):
if os.path.exists(json_name):
return
args = args + ["-o", json_name]
print(args)
p = subprocess.Popen(args, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
out, err = p.communicate()
print(out.decode())
print(err.decode())
def make_plot(extension):
ratios = OrderedDict()
stddevs = OrderedDict()
benchmarks = OrderedDict()
np_bench = pyperf.Benchmark.load(open("{}_{}".format("numpy", extension)))
np_mean = np_bench.mean()
np_stddev = np_bench.stdev()
for package in setup:
if package == "numpy":
continue
benchmarks[package] = pyperf.Benchmark.load(open(f"{package}_{extension}"))
mean = benchmarks[package].mean()
stddev = benchmarks[package].stdev()
ratios[package] = mean / np_mean
stddevs[package] = ratios[package] * np.sqrt(
(np_stddev / np_mean) ** 2 + (stddev / mean) ** 2
)
fig, ax = plt.subplots()
packages = list(ratios.keys())
ax.bar(packages, ratios.values(), yerr=stddevs.values())
fig.suptitle(extension.replace(".json", "").replace("_", " ").title())
ax.set_ylabel("numpy overhead (x time for numpy); lower is better")
plt.savefig(extension.replace(".json", ".png"))
plt.close(fig)
if ratios["unyt"] != min(ratios.values()):
rvalues = list(ratios.values())
svalues = list(stddevs.values())
unyt_index = packages.index("unyt")
min_index = rvalues.index(min(rvalues))
if ratios["unyt"] > 3 * svalues[min_index] + rvalues[min_index]:
for package in ratios:
script = get_script(benchmarks, package)
with stdoutIO() as s:
exec(script)
res = s.getvalue().replace("\n", "")
print(
"{}: {} +- {} ({})".format(
package, ratios[package], stddevs[package], res
)
)
print(get_script(benchmarks, "unyt"))
def get_script(benchmarks, package):
meta = benchmarks[package].get_metadata()
setup_s = meta["timeit_setup"][1:-1]
bench_s = "print(" + meta["timeit_stmt"][1:-1] + ")"
script = setup_s + "; " + bench_s
script = script.replace("; ", "\n")
return script
setup = OrderedDict(
[
("numpy", "import numpy as np"),
("pint", "from pint import UnitRegistry; u = UnitRegistry()"),
("astropy", "import astropy.units as u"),
("unyt", "import unyt as u"),
("quantities", "import quantities as u"),
]
)
base_args = ["python", "-m", "pyperf", "timeit"]
shared_setup = "import numpy as np; import operator"
base_setups = OrderedDict(
[
("small_list", "data = [1., 2., 3.]"),
("small_tuple", "data = (1., 2., 3.)"),
("small_array", "data = np.array([1., 2., 3.])"),
("big_list", "data = (np.arange(1e6)+1).tolist()"),
("big_array", "data = (np.arange(1e6)+1)"),
]
)
op_ufuncs = OrderedDict(
[
("operator.add", "np.add"),
("operator.sub", "np.subtract"),
("operator.mul", "np.multiply"),
("operator.truediv", "np.true_divide"),
("operator.eq", "np.equal"),
]
)
for bs in base_setups:
for package in sorted(setup):
print(package)
setup_s = "; ".join([shared_setup, setup[package], base_setups[bs]])
args = base_args + ["-s", setup_s + " "]
if package == "numpy":
args.append("np.array(data)")
else:
args.append("data*u.g")
json_name = f"{package}_{bs}_create.json"
run_perf(args, json_name)
if "list" in bs or "tuple" in bs:
continue
args = base_args + ["-s", setup_s + "; data=np.asarray(data); out=data.copy()"]
if package == "numpy":
args[-1] += "; "
else:
if package != "pint":
args[-1] += "*u.g"
args[-1] += "; data = data*u.g "
args.append("data**2")
json_name = f"{package}_{bs}_square.json"
run_perf(args, json_name)
args[-1] = "np.power(data, 2)"
json_name = f"{package}_{bs}_npsquare.json"
run_perf(args, json_name)
args[-1] = "np.power(data, 2, out=out)"
json_name = f"{package}_{bs}_npsquareout.json"
run_perf(args, json_name)
args[-1] = "data**0.5"
json_name = f"{package}_{bs}_sqrt.json"
run_perf(args, json_name)
args[-1] = "np.sqrt(data)"
json_name = f"{package}_{bs}_npsqrt.json"
run_perf(args, json_name)
args[-1] = "np.sqrt(data, out=out)"
json_name = f"{package}_{bs}_npsqrtout.json"
run_perf(args, json_name)
make_plot(f"{bs}_create.json")
if "list" not in bs and "tuple" not in bs:
make_plot(f"{bs}_square.json")
make_plot(f"{bs}_npsquare.json")
make_plot(f"{bs}_npsquareout.json")
make_plot(f"{bs}_sqrt.json")
make_plot(f"{bs}_npsqrt.json")
make_plot(f"{bs}_npsqrtout.json")
for bs in base_setups:
if "list" in bs or "tuple" in bs:
continue
for op, ufunc in op_ufuncs.items():
for bench, bench_name in [
(op + r"(data1, data2)", op + "12.json"),
(op + r"(data2, data1)", op + "21.json"),
(ufunc + r"(data1, data2)", ufunc + "12.json"),
(ufunc + r"(data2, data1)", ufunc + "21.json"),
(ufunc + r"(data1, data2, out=out)", ufunc + "12out.json"),
(ufunc + r"(data2, data1, out=out)", ufunc + "21out.json"),
]:
for unit_choice in [("g", "g"), ("kg", "g")]:
for package in sorted(setup):
print(package)
setup_s = (
"; ".join([shared_setup, setup[package], base_setups[bs]])
+ "; "
)
if "out" in bench:
if package not in ("pint", "numpy") and "equal" not in bench:
setup_s += f"out=data*u.{unit_choice[0]}; "
else:
setup_s += "out=np.array(data); "
if package == "numpy":
setup_s += "; ".join(
[r"data1 = np.array(data)", r"data2 = np.array(data)"]
)
if unit_choice[0] != unit_choice[1]:
_bench = bench.replace("data1", ".001*data1")
else:
setup_s += "; ".join(
[
f"data1 = data*u.{unit_choice[0]}",
f"data2 = data*u.{unit_choice[1]}",
]
)
_bench = bench
args = base_args + ["-s", setup_s + " "]
json_name = "{}_{}_{}{}".format(
package, bs, unit_choice[0], unit_choice[1]
)
run_perf(args + [_bench], json_name + "_" + bench_name)
make_plot(f"{bs}_{unit_choice[0]}{unit_choice[1]}_{bench_name}")
|