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
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
|
import random
import time
import sys
import itertools
import multiprocessing
import array, numpy, rpy2.robjects as ro
import rpy2.robjects.packages
from rpy2.robjects import Formula
from functools import reduce
n_loops = (1, 10, 20, 30, 40)
def setup_func(kind):
#-- setup_sum-begin
n = 20000
x_list = [random.random() for i in range(n)]
module = None
if kind == "array.array":
import array as module
res = module.array('f', x_list)
elif kind == "numpy.array":
import numpy as module
res = module.array(x_list, 'f')
elif kind == "FloatVector":
import rpy2.robjects as module
res = module.FloatVector(x_list)
elif kind == "FloatSexpVector":
import rpy2.rinterface as module
module.initr()
res = module.FloatSexpVector(x_list)
elif kind == "FloatSexpVector-memoryview-array":
import rpy2.rinterface as module
module.initr()
tmp = module.FloatSexpVector(x_list)
mv = tmp.memoryview()
res = array.array(mv.format, mv)
elif kind == "list":
res = x_list
elif kind == "R":
import rpy2.robjects as module
res = module.rinterface.FloatSexpVector(x_list)
module.globalenv['x'] = res
res = None
#-- setup_sum-end
else:
raise ValueError("Unknown kind '%s'" %kind)
return (res, module)
def python_reduce(x):
total = reduce(lambda x, y: x+y, x)
return total
#-- purepython_sum-begin
def python_sum(x):
total = 0.0
for elt in x:
total += elt
return total
#-- purepython_sum-end
def test_sumr(queue, func, n, setup_func):
array, module = setup_func("R")
r_loopsum = """
#-- purer_sum-begin
function(x)
{
total = 0;
for (elt in x) {
total <- total + elt
}
}
#-- purer_sum-end
"""
rdo_loopsum = """
f <- %s
for (i in 1:%i) {
res <- f(x)
}
"""
rcode = rdo_loopsum %(r_loopsum, n)
time_beg = time.time()
#print(rcode)
module.r(rcode)
time_end = time.time()
queue.put(time_end - time_beg)
def test_sumr_compile(queue, func, n, setup_func):
array, module = setup_func("R")
r_loopsum = """
#-- purer_sum-begin
function(x)
{
total = 0;
for (elt in x) {
total <- total + elt
}
}
#-- purer_sum-end
"""
rdo_loopsum = """
f <- %s
for (i in 1:%i) {
res <- f(x)
}
"""
compiler = ro.packages.importr("compiler")
rcode = rdo_loopsum %("compiler::cmpfun("+r_loopsum+")", n)
time_beg = time.time()
#print(rcode)
module.r(rcode)
time_end = time.time()
queue.put(time_end - time_beg)
def test_sumr_builtin(queue, func, n, setup_func):
array, module = setup_func("R")
r_sum = """
sum
"""
rdo_loopsum = """
f <- %s
for (i in 1:%i) {
res <- f(x)
}
"""
rcode = rdo_loopsum %(r_sum, n)
time_beg = time.time()
#print(rcode)
module.r(rcode)
time_end = time.time()
queue.put(time_end - time_beg)
def test_sum(queue, func, array_type, n, setup_func):
array, module = setup_func(array_type)
time_beg = time.time()
for i in range(n):
res = func(array)
time_end = time.time()
queue.put(time_end - time_beg)
q = multiprocessing.Queue()
combos = [(label_func, func, label_sequence, n_loop) \
for label_func, func in (("pure python", python_sum), \
("reduce python", python_reduce), \
("builtin python", sum))
for label_sequence in ("FloatSexpVector",
"FloatSexpVector-memoryview-array",
"FloatVector",
"list",
"array.array",
"numpy.array")
for n_loop in n_loops]
times = []
# python
for label_func, func, label_sequence, n_loop in combos:
p = multiprocessing.Process(target = test_sum, args = (q, func,
label_sequence,
n_loop,
setup_func))
p.start()
res = q.get()
p.join()
times.append(res)
combos_r = [(label_func, None, None, n_loop) \
for label_func, func in (("R", None), \
("R compiled", None), \
("R builtin", None))
for n_loop in n_loops]
times_r = []
# pure R
for n_loop in n_loops:
p = multiprocessing.Process(target = test_sumr, args = (q, func,
n_loop,
setup_func))
p.start()
res = q.get()
p.join()
times_r.append(res)
# pure R compiled
for n_loop in n_loops:
p = multiprocessing.Process(target = test_sumr_compile,
args = (q, func,
n_loop,
setup_func))
p.start()
res = q.get()
p.join()
times_r.append(res)
# R builtin
for n_loop in n_loops:
p = multiprocessing.Process(target = test_sumr_builtin, args = (q, func,
n_loop,
setup_func))
p.start()
res = q.get()
p.join()
times_r.append(res)
from rpy2.robjects.vectors import DataFrame, FloatVector, StrVector, IntVector
d = {}
d['code'] = StrVector([x[0] for x in combos]) + StrVector([x[0] for x in combos_r])
d['sequence'] = StrVector([x[-2] for x in combos]) + StrVector([x[0] for x in combos_r])
d['time'] = FloatVector([x for x in times]) + FloatVector(times_r)
d['n_loop'] = IntVector([x[-1] for x in combos]) + IntVector([x[3] for x in combos_r])
d['group'] = StrVector([d['code'][x] + ':' + d['sequence'][x] for x in range(len(d['n_loop']))])
dataf = DataFrame(d)
from rpy2.robjects.lib import ggplot2
p = ggplot2.ggplot(dataf) + \
ggplot2.geom_line(ggplot2.aes_string(x="n_loop",
y="time",
colour="code")) + \
ggplot2.geom_point(ggplot2.aes_string(x="n_loop",
y="time",
colour="code")) + \
ggplot2.facet_wrap(Formula('~sequence')) + \
ggplot2.scale_y_sqrt('running time') + \
ggplot2.scale_x_continuous('repeated n times', ) + \
ggplot2.xlim(0, max(n_loops)) + \
ggplot2.labs(title = "Benchmark (running time)")
from rpy2.robjects.packages import importr
grdevices = importr('grDevices')
grdevices.png('../../_static/benchmark_sum.png',
width=812, height=612,
type="cairo")
p.plot()
grdevices.dev_off()
#base = importr("base")
stats = importr('stats')
nlme = importr("nlme")
fit = nlme.lmList(Formula('time ~ n_loop | group'), data = dataf,
na_action = stats.na_exclude)
# scale to R's slope
speedup = [stats.coef(fit).rx("R:R", True)[1][0]/x for x in stats.coef(fit).rx2("n_loop")]
# workaround an issue in class mapping
df = DataFrame(stats.coef(fit))
import os
pad_len = (15, 45, 10)
header = ("Function", "Sequence", "Speedup")
res = [' '.join(["=" * pad_len[x] for x in range(3)]),
' '.join([header[x].ljust(pad_len[x]) for x in range(3)]),
' '.join(["=" * pad_len[x] for x in range(3)])]
for coef_name, sp in zip(df.rownames, speedup):
row_content = coef_name.split(':')
row_content.append('%.2f' %sp)
res.append(' '.join([row_content[x].ljust(pad_len[x]) for x in range(3)]))
res.append(' '.join(["=" * pad_len[x] for x in range(3)]))
print(os.linesep.join(res))
|
