flori-bullshit 0.1.0
Sign up to get free protection for your applications and to get access to all the features.
- data/COPYING +340 -0
- data/Rakefile +95 -0
- data/VERSION +1 -0
- data/bullshit.gemspec +25 -0
- data/examples/compare.rb +11 -0
- data/examples/fibonacci.rb +75 -0
- data/examples/iteration.rb +116 -0
- data/examples/josephus.rb +136 -0
- data/examples/sorting.rb +363 -0
- data/install.rb +15 -0
- data/lib/bullshit.rb +2244 -0
- data/lib/bullshit/version.rb +8 -0
- data/make_doc.rb +4 -0
- data/tests/test_analysis.rb +321 -0
- data/tests/test_bullshit.rb +222 -0
- data/tests/test_continued_fraction.rb +40 -0
- data/tests/test_distribution.rb +69 -0
- data/tests/test_functions.rb +33 -0
- data/tests/test_newton_bisection.rb +28 -0
- data/tests/test_window.rb +34 -0
- metadata +94 -0
data/install.rb
ADDED
@@ -0,0 +1,15 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
require 'rbconfig'
|
4
|
+
require 'fileutils'
|
5
|
+
include FileUtils::Verbose
|
6
|
+
|
7
|
+
include Config
|
8
|
+
|
9
|
+
file = 'lib/bullshit.rb'
|
10
|
+
libdir = CONFIG["sitelibdir"]
|
11
|
+
install(file, libdir, :mode => 0755)
|
12
|
+
mkdir_p subdir = File.join(libdir, 'bullshit')
|
13
|
+
for f in Dir['lib/bullshit/*.rb']
|
14
|
+
install(f, subdir)
|
15
|
+
end
|
data/lib/bullshit.rb
ADDED
@@ -0,0 +1,2244 @@
|
|
1
|
+
# = Bullshit - Benchmarking in Ruby
|
2
|
+
#
|
3
|
+
# == Description
|
4
|
+
#
|
5
|
+
# == Usage
|
6
|
+
#
|
7
|
+
# == Author
|
8
|
+
#
|
9
|
+
# Florian Frank mailto:flori@ping.de
|
10
|
+
#
|
11
|
+
# == License
|
12
|
+
#
|
13
|
+
# This is free software; you can redistribute it and/or modify it under the
|
14
|
+
# terms of the GNU General Public License Version 2 as published by the Free
|
15
|
+
# Software Foundation: www.gnu.org/copyleft/gpl.html
|
16
|
+
#
|
17
|
+
# == Download
|
18
|
+
#
|
19
|
+
# The latest version of this library can be downloaded at
|
20
|
+
#
|
21
|
+
# * http://rubyforge.org/frs/?group_id=8323
|
22
|
+
#
|
23
|
+
# The homepage of this library is located at
|
24
|
+
#
|
25
|
+
# * http://bullshit.rubyforge.org
|
26
|
+
#
|
27
|
+
# == Example
|
28
|
+
#
|
29
|
+
|
30
|
+
require 'dslkit'
|
31
|
+
require 'enumerator'
|
32
|
+
|
33
|
+
begin
|
34
|
+
require 'bullshit/version'
|
35
|
+
rescue LoadError
|
36
|
+
end
|
37
|
+
|
38
|
+
# Module that includes all constants of the bullshit library.
|
39
|
+
module Bullshit
|
40
|
+
COLUMNS = 79 # Number of columns in the output.
|
41
|
+
|
42
|
+
NAME_COLUMN_SIZE = 5 # Number of columns used for row names.
|
43
|
+
|
44
|
+
Infinity = 1.0 / 0 # Refers to floating point infinity.
|
45
|
+
|
46
|
+
RUBY_DESCRIPTION = "ruby %s (%s patchlevel %s) [%s]" %
|
47
|
+
[ RUBY_VERSION, RUBY_RELEASE_DATE, RUBY_PATCHLEVEL, RUBY_PLATFORM ]
|
48
|
+
|
49
|
+
# This class implements a continued fraction of the form:
|
50
|
+
#
|
51
|
+
# b_1
|
52
|
+
# a_0 + -------------------------
|
53
|
+
# b_2
|
54
|
+
# a_1 + --------------------
|
55
|
+
# b_3
|
56
|
+
# a_2 + ---------------
|
57
|
+
# b_4
|
58
|
+
# a_3 + ----------
|
59
|
+
# b_5
|
60
|
+
# a_4 + -----
|
61
|
+
# ...
|
62
|
+
#
|
63
|
+
class ContinuedFraction
|
64
|
+
# Creates a continued fraction instance. With the defaults for_a { 1 } and
|
65
|
+
# for_b { 1 } it approximates the golden ration phi if evaluated.
|
66
|
+
def initialize
|
67
|
+
@a = proc { 1.0 }
|
68
|
+
@b = proc { 1.0 }
|
69
|
+
end
|
70
|
+
|
71
|
+
# Creates a ContinuedFraction instances and passes its arguments to a call
|
72
|
+
# to for_a.
|
73
|
+
def self.for_a(arg = nil, &block)
|
74
|
+
new.for_a(arg, &block)
|
75
|
+
end
|
76
|
+
|
77
|
+
# Creates a ContinuedFraction instances and passes its arguments to a call
|
78
|
+
# to for_b.
|
79
|
+
def self.for_b(arg = nil, &block)
|
80
|
+
new.for_b(arg, &block)
|
81
|
+
end
|
82
|
+
|
83
|
+
# This method either takes a block or an argument +arg+. The argument +arg+
|
84
|
+
# has to respond to an integer index n >= 0 and return the value a_n. The
|
85
|
+
# block has to return the value for a_n when +n+ is passed as the first
|
86
|
+
# argument to the block. If a_n is dependent on an +x+ value (see the call
|
87
|
+
# method) the +x+ will be the second argument of the block.
|
88
|
+
def for_a(arg = nil, &block)
|
89
|
+
if arg and !block
|
90
|
+
@a = arg
|
91
|
+
elsif block and !arg
|
92
|
+
@a = block
|
93
|
+
else
|
94
|
+
raise ArgumentError, "exactly one argument or one block required"
|
95
|
+
end
|
96
|
+
self
|
97
|
+
end
|
98
|
+
|
99
|
+
# This method either takes a block or an argument +arg+. The argument +arg+
|
100
|
+
# has to respond to an integer index n >= 1 and return the value b_n. The
|
101
|
+
# block has to return the value for b_n when +n+ is passed as the first
|
102
|
+
# argument to the block. If b_n is dependent on an +x+ value (see the call
|
103
|
+
# method) the +x+ will be the second argument of the block.
|
104
|
+
def for_b(arg = nil, &block)
|
105
|
+
if arg and !block
|
106
|
+
@b = arg
|
107
|
+
elsif block and !arg
|
108
|
+
@b = block
|
109
|
+
else
|
110
|
+
raise ArgumentError, "exactly one argument or one block required"
|
111
|
+
end
|
112
|
+
self
|
113
|
+
end
|
114
|
+
|
115
|
+
# Returns the value for a_n or a_n(x).
|
116
|
+
def a(n, x = nil)
|
117
|
+
result = if x
|
118
|
+
@a[n, x]
|
119
|
+
else
|
120
|
+
@a[n]
|
121
|
+
end and result.to_f
|
122
|
+
end
|
123
|
+
|
124
|
+
# Returns the value for b_n or b_n(x).
|
125
|
+
def b(n, x = nil)
|
126
|
+
result = if x
|
127
|
+
@b[n, x]
|
128
|
+
else
|
129
|
+
@b[n]
|
130
|
+
end and result.to_f
|
131
|
+
end
|
132
|
+
|
133
|
+
# Evaluates the continued fraction for the value +x+ (if any) with the
|
134
|
+
# accuracy +epsilon+ and +max_iterations+ as the maximum number of
|
135
|
+
# iterations using the Wallis-method with scaling.
|
136
|
+
def call(x = nil, epsilon = 1E-16, max_iterations = 1 << 31)
|
137
|
+
c_0, c_1 = 1.0, a(0, x)
|
138
|
+
c_1 == nil and return 0 / 0.0
|
139
|
+
d_0, d_1 = 0.0, 1.0
|
140
|
+
result = c_1 / d_1
|
141
|
+
n = 0
|
142
|
+
error = 1 / 0.0
|
143
|
+
$DEBUG and warn "n=%u, a=%f, b=nil, c=%f, d=%f result=%f, error=nil" %
|
144
|
+
[ n, c_1, c_1, d_1, result ]
|
145
|
+
while n < max_iterations and error > epsilon
|
146
|
+
n += 1
|
147
|
+
a_n, b_n = a(n, x), b(n, x)
|
148
|
+
a_n and b_n or break
|
149
|
+
c_2 = a_n * c_1 + b_n * c_0
|
150
|
+
d_2 = a_n * d_1 + b_n * d_0
|
151
|
+
if c_2.infinite? or d_2.infinite?
|
152
|
+
if a_n != 0
|
153
|
+
c_2 = c_1 + (b_n / a_n * c_0)
|
154
|
+
d_2 = d_1 + (b_n / a_n * d_0)
|
155
|
+
elsif b_n != 0
|
156
|
+
c_2 = (a_n / b_n * c_1) + c_0
|
157
|
+
d_2 = (a_n / b_n * d_1) + d_0
|
158
|
+
else
|
159
|
+
raise Errno::ERANGE
|
160
|
+
end
|
161
|
+
end
|
162
|
+
r = c_2 / d_2
|
163
|
+
error = (r / result - 1).abs
|
164
|
+
|
165
|
+
result = r
|
166
|
+
|
167
|
+
$DEBUG and warn "n=%u, a=%f, b=%f, c=%f, d=%f, result=%f, error=%.16f" %
|
168
|
+
[ n, a_n, b_n, c_1, d_1, result, error ]
|
169
|
+
|
170
|
+
c_0, c_1 = c_1, c_2
|
171
|
+
d_0, d_1 = d_1, d_2
|
172
|
+
end
|
173
|
+
n >= max_iterations and raise Errno::ERANGE
|
174
|
+
result
|
175
|
+
end
|
176
|
+
|
177
|
+
alias [] call
|
178
|
+
|
179
|
+
# Returns this continued fraction as a Proc object which takes the same
|
180
|
+
# arguments like its call method does.
|
181
|
+
def to_proc
|
182
|
+
proc { |*a| call(*a) }
|
183
|
+
end
|
184
|
+
end
|
185
|
+
|
186
|
+
module ModuleFunctions
|
187
|
+
module_function
|
188
|
+
|
189
|
+
# Return the angle +degree+ in radians.
|
190
|
+
def angle(degree)
|
191
|
+
Math.tan(Math::PI * degree / 180)
|
192
|
+
end
|
193
|
+
|
194
|
+
# Return the percentage number as a value in the range 0..1.
|
195
|
+
def percent(number)
|
196
|
+
number / 100.0
|
197
|
+
end
|
198
|
+
|
199
|
+
# Let a window of size +window_size+ slide over the array +array+ and yield
|
200
|
+
# to the window array.
|
201
|
+
def array_window(array, window_size)
|
202
|
+
window_size < 1 and raise ArgumentError, "window_size = #{window_size} < 1"
|
203
|
+
window_size = window_size.to_i
|
204
|
+
window_size += 1 if window_size % 2 == 0
|
205
|
+
radius = window_size / 2
|
206
|
+
array.each_index do |i|
|
207
|
+
ws = window_size
|
208
|
+
from = i - radius
|
209
|
+
negative_from = false
|
210
|
+
if from < 0
|
211
|
+
negative_from = true
|
212
|
+
ws += from
|
213
|
+
from = 0
|
214
|
+
end
|
215
|
+
a = array[from, ws]
|
216
|
+
if (diff = window_size - a.size) > 0
|
217
|
+
mean = a.inject(0.0) { |s, x| s + x } / a.size
|
218
|
+
a = if negative_from
|
219
|
+
[ mean ] * diff + a
|
220
|
+
else
|
221
|
+
a + [ mean ] * diff
|
222
|
+
end
|
223
|
+
end
|
224
|
+
yield a
|
225
|
+
end
|
226
|
+
nil
|
227
|
+
end
|
228
|
+
end
|
229
|
+
|
230
|
+
# An excpeption raised by the bullshit library.
|
231
|
+
class BullshitException < StandardError; end
|
232
|
+
|
233
|
+
# A Clock instance is used to take measurements while benchmarking.
|
234
|
+
class Clock
|
235
|
+
TIMES = [ :real, :total, :user, :system ]
|
236
|
+
|
237
|
+
ALL_COLUMNS = [ :scatter ] + TIMES + [ :repeat ]
|
238
|
+
|
239
|
+
TIMES_MAX = TIMES.map { |t| t.to_s.size }.max
|
240
|
+
|
241
|
+
# Returns a Clock instance for CaseMethod instance +bc_method+.
|
242
|
+
def initialize(bc_method)
|
243
|
+
@bc_method = bc_method
|
244
|
+
@times = Hash.new { |h, k| h[k] = [] }
|
245
|
+
@repeat = 0
|
246
|
+
@scatter = 0
|
247
|
+
end
|
248
|
+
|
249
|
+
# Use a Clock instance to measure the time necessary to do
|
250
|
+
# bc_method.case.iterations repetitions of +bc_method+.
|
251
|
+
def self.repeat(bc_method)
|
252
|
+
clock = new(bc_method)
|
253
|
+
bs = clock.case.batch_size.abs
|
254
|
+
bs = 1 if !bs or bs < 0
|
255
|
+
clock.case.iterations.times do
|
256
|
+
bc_method.before_run
|
257
|
+
$DEBUG and warn "Calling #{bc_method.name}."
|
258
|
+
clock.inc_scatter
|
259
|
+
clock.measure do
|
260
|
+
bs.times { yield }
|
261
|
+
end
|
262
|
+
bc_method.after_run
|
263
|
+
end
|
264
|
+
clock
|
265
|
+
end
|
266
|
+
|
267
|
+
# Use a Clock instance to measure how many repetitions of +bc_method+ can
|
268
|
+
# be done in bc_method.case.duration seconds (a float value). If the
|
269
|
+
# bc_method.case.batch_size is >1 duration is multiplied by batch_size
|
270
|
+
# because the measured times are averaged by batch_size.
|
271
|
+
def self.time(bc_method)
|
272
|
+
clock = new(bc_method)
|
273
|
+
duration = clock.case.duration.abs
|
274
|
+
if bs = clock.case.batch_size and bs > 1
|
275
|
+
duration *= bs
|
276
|
+
end
|
277
|
+
until_at = Time.now + duration
|
278
|
+
bs = clock.case.batch_size.abs
|
279
|
+
bs = 1 if !bs or bs < 0
|
280
|
+
begin
|
281
|
+
bc_method.before_run
|
282
|
+
$DEBUG and warn "Calling #{bc_method.name}."
|
283
|
+
clock.inc_scatter
|
284
|
+
clock.measure do
|
285
|
+
bs.times { yield }
|
286
|
+
end
|
287
|
+
bc_method.after_run
|
288
|
+
end until clock.time > until_at
|
289
|
+
clock
|
290
|
+
end
|
291
|
+
|
292
|
+
# Iterate over the +range+ of the RangeCase instance of this +bc_method+
|
293
|
+
# and take measurements (including scattering).
|
294
|
+
def self.scale_range(bc_method)
|
295
|
+
clock = new(bc_method)
|
296
|
+
my_case = clock.case
|
297
|
+
bs = my_case.batch_size.abs
|
298
|
+
bs = 1 if !bs or bs < 0
|
299
|
+
for a in my_case.range
|
300
|
+
begin
|
301
|
+
my_case.args = (a.dup rescue a).freeze
|
302
|
+
clock.inc_scatter
|
303
|
+
my_case.scatter.times do
|
304
|
+
bc_method.before_run
|
305
|
+
$DEBUG and warn "Calling #{bc_method.name}."
|
306
|
+
clock.measure do
|
307
|
+
bs.times { yield }
|
308
|
+
end
|
309
|
+
bc_method.after_run
|
310
|
+
end
|
311
|
+
ensure
|
312
|
+
my_case.args = nil
|
313
|
+
end
|
314
|
+
end
|
315
|
+
clock
|
316
|
+
end
|
317
|
+
|
318
|
+
# The benchmark case class this clock belongs to (via bc_method).
|
319
|
+
def case
|
320
|
+
@bc_method.case.class
|
321
|
+
end
|
322
|
+
|
323
|
+
# Returns the benchmark method for this Clock instance.
|
324
|
+
attr_reader :bc_method
|
325
|
+
|
326
|
+
# Number of repetitions this clock has measured.
|
327
|
+
attr_accessor :repeat
|
328
|
+
|
329
|
+
# Last time object used for real time measurement.
|
330
|
+
attr_reader :time
|
331
|
+
|
332
|
+
# Return all the slopes of linear regressions computed during data
|
333
|
+
# truncation phase.
|
334
|
+
attr_reader :slopes
|
335
|
+
|
336
|
+
# Add the array +times+ to this clock's time measurements. +times+ consists
|
337
|
+
# of the time measurements in float values in order of TIMES.
|
338
|
+
def <<(times)
|
339
|
+
r = times.shift
|
340
|
+
@repeat += 1 if @times[:repeat].last != r
|
341
|
+
@times[:repeat] << r
|
342
|
+
TIMES.zip(times) { |t, time| @times[t] << time.to_f }
|
343
|
+
self
|
344
|
+
end
|
345
|
+
|
346
|
+
# Returns a Hash of Analysis object for all of TIMES's time keys.
|
347
|
+
def analysis
|
348
|
+
@analysis ||= Hash.new do |h, time|
|
349
|
+
time = time.to_sym
|
350
|
+
times = @times[time]
|
351
|
+
h[time] = Analysis.new(times)
|
352
|
+
end
|
353
|
+
end
|
354
|
+
|
355
|
+
# Return true, if other's mean value is indistinguishable from this
|
356
|
+
# object's mean after filtering out the noise from the measurements with a
|
357
|
+
# Welch's t-Test. This mean's that differences in the mean of both clocks
|
358
|
+
# might not inidicate a real performance difference and may be caused by
|
359
|
+
# chance.
|
360
|
+
def cover?(other)
|
361
|
+
time = self.case.compare_time.to_sym
|
362
|
+
analysis[time].cover?(other.analysis[time], self.case.covering.alpha_level.abs)
|
363
|
+
end
|
364
|
+
|
365
|
+
# Return column names in relation to Clock#to_a method.
|
366
|
+
def self.to_a
|
367
|
+
%w[ #scatter ] + TIMES + %w[ repeat ]
|
368
|
+
end
|
369
|
+
|
370
|
+
# Returns the measurements as an array of arrays.
|
371
|
+
def to_a
|
372
|
+
if @repeat >= 1
|
373
|
+
(::Bullshit::Clock::ALL_COLUMNS).map do |t|
|
374
|
+
analysis[t].measurements
|
375
|
+
end.transpose
|
376
|
+
else
|
377
|
+
[]
|
378
|
+
end
|
379
|
+
end
|
380
|
+
|
381
|
+
# Takes the times an returns an array, consisting of the times in the order
|
382
|
+
# of enumerated in the TIMES constant.
|
383
|
+
def take_time
|
384
|
+
@time, times = Time.now, Process.times
|
385
|
+
user_time = times.utime + times.cutime # user time of this process and its children
|
386
|
+
system_time = times.stime + times.cstime # system time of this process and its children
|
387
|
+
total_time = user_time + system_time # total time of this process and its children
|
388
|
+
[ @time.to_f, total_time, user_time, system_time ]
|
389
|
+
end
|
390
|
+
|
391
|
+
# Increment scatter counter by one.
|
392
|
+
def inc_scatter
|
393
|
+
@scatter += 1
|
394
|
+
end
|
395
|
+
|
396
|
+
# Take a single measurement. This method should be called with the code to
|
397
|
+
# benchmark in a block.
|
398
|
+
def measure
|
399
|
+
before = take_time
|
400
|
+
yield
|
401
|
+
after = take_time
|
402
|
+
@repeat += 1
|
403
|
+
@times[:repeat] << @repeat
|
404
|
+
@times[:scatter] << @scatter
|
405
|
+
bs = self.case.batch_size.abs
|
406
|
+
if bs and bs > 1
|
407
|
+
TIMES.each_with_index { |t, i| @times[t] << (after[i] - before[i]) / bs }
|
408
|
+
else
|
409
|
+
TIMES.each_with_index { |t, i| @times[t] << after[i] - before[i] }
|
410
|
+
end
|
411
|
+
@analysis = nil
|
412
|
+
end
|
413
|
+
|
414
|
+
# Returns the sample standard deviation for the +time+ (one of TIMES'
|
415
|
+
# symbols).
|
416
|
+
def sample_standard_deviation(time)
|
417
|
+
analysis[time.to_sym].sample_standard_deviation
|
418
|
+
end
|
419
|
+
|
420
|
+
# Returns the sample standard deviation for the +time+ (one of TIMES'
|
421
|
+
# symbols) in percentage of its arithmetic mean.
|
422
|
+
def sample_standard_deviation_percentage(time)
|
423
|
+
analysis[time.to_sym].sample_standard_deviation_percentage
|
424
|
+
end
|
425
|
+
|
426
|
+
# Returns the minimum for the +time+ (one of TIMES' symbols).
|
427
|
+
def min(time)
|
428
|
+
analysis[time.to_sym].min
|
429
|
+
end
|
430
|
+
|
431
|
+
# Returns the maximum for the +time+ (one of TIMES' symbols).
|
432
|
+
def max(time)
|
433
|
+
analysis[time.to_sym].max
|
434
|
+
end
|
435
|
+
|
436
|
+
# Returns the median of the +time+ values (one of TIMES' symbols).
|
437
|
+
def median(time)
|
438
|
+
analysis[time.to_sym].median
|
439
|
+
end
|
440
|
+
|
441
|
+
# Returns the +p+-percentile of the +time+ values (one of TIMES' symbols).
|
442
|
+
def percentile(time, p = 50)
|
443
|
+
analysis[time.to_sym].percentile p
|
444
|
+
end
|
445
|
+
|
446
|
+
# Returns the q value for the Ljung-Box statistic of this +time+'s
|
447
|
+
# analysis.detect_autocorrelation method.
|
448
|
+
def detect_autocorrelation(time)
|
449
|
+
analysis[time.to_sym].detect_autocorrelation(
|
450
|
+
self.case.autocorrelation.max_lags.to_i,
|
451
|
+
self.case.autocorrelation.alpha_level.abs)
|
452
|
+
end
|
453
|
+
|
454
|
+
# Return a result hash with the number of :very_low, :low, :high, and
|
455
|
+
# :very_high outliers, determined by the box plotting algorithm run with
|
456
|
+
# :median and :iqr parameters. If no outliers were found or the iqr is less
|
457
|
+
# than epsilon, nil is returned.
|
458
|
+
def detect_outliers(time)
|
459
|
+
analysis[time.to_sym].detect_outliers(self.case.outliers_factor.abs)
|
460
|
+
end
|
461
|
+
|
462
|
+
TIMES.each do |time|
|
463
|
+
define_method(time) { analysis[time].sum }
|
464
|
+
end
|
465
|
+
|
466
|
+
# Seconds per call (mean)
|
467
|
+
def call_time_mean
|
468
|
+
mean(self.case.compare_time)
|
469
|
+
end
|
470
|
+
|
471
|
+
# Calls per second of the +call_time_type+, e. g. :call_time_mean or
|
472
|
+
# :call_time_median.
|
473
|
+
def calls(call_time_type)
|
474
|
+
__send__(call_time_type) ** -1
|
475
|
+
end
|
476
|
+
|
477
|
+
# Calls per second (mean)
|
478
|
+
def calls_mean
|
479
|
+
call_time_mean ** -1
|
480
|
+
end
|
481
|
+
|
482
|
+
# Seconds per call (median)
|
483
|
+
def call_time_median
|
484
|
+
median(self.case.compare_time)
|
485
|
+
end
|
486
|
+
|
487
|
+
# Calls per second (median)
|
488
|
+
def calls_median
|
489
|
+
call_time_median ** -1
|
490
|
+
end
|
491
|
+
|
492
|
+
# Returns the arithmetic mean of +time+.
|
493
|
+
def arithmetic_mean(time)
|
494
|
+
analysis[time.to_sym].mean
|
495
|
+
end
|
496
|
+
|
497
|
+
alias mean arithmetic_mean
|
498
|
+
|
499
|
+
# Returns the harmonic mean of +time+.
|
500
|
+
def harmonic_mean(time)
|
501
|
+
analysis[time.to_sym].harmonic_mean
|
502
|
+
end
|
503
|
+
|
504
|
+
# Returns the geometric mean of +time+.
|
505
|
+
def geometric_mean(time)
|
506
|
+
analysis[time.to_sym].geometric_mean
|
507
|
+
end
|
508
|
+
|
509
|
+
# The times which should be displayed in the output.
|
510
|
+
def self.times
|
511
|
+
TIMES.map { |t| t.to_s }
|
512
|
+
end
|
513
|
+
|
514
|
+
# Return the Histogram for the +time+ values.
|
515
|
+
def histogram(time)
|
516
|
+
analysis[time.to_sym].histogram(self.case.histogram.bins)
|
517
|
+
end
|
518
|
+
|
519
|
+
# Return the array of autocorrelation values for +time+.
|
520
|
+
def autocorrelation(time)
|
521
|
+
analysis[time.to_sym].autocorrelation
|
522
|
+
end
|
523
|
+
|
524
|
+
# Returns the arrays for the autocorrelation plot, the first array for the
|
525
|
+
# numbers of lag measured, the second for the autocorrelation value.
|
526
|
+
def autocorrelation_plot(time)
|
527
|
+
r = autocorrelation time
|
528
|
+
start = @times[:repeat].first
|
529
|
+
ende = (start + r.size)
|
530
|
+
(start...ende).to_a.zip(r)
|
531
|
+
end
|
532
|
+
|
533
|
+
# Return the result of CaseMethod#file_path for this clock's bc_method.
|
534
|
+
def file_path(*args)
|
535
|
+
@bc_method.file_path(*args)
|
536
|
+
end
|
537
|
+
|
538
|
+
# Truncate the measurements stored in this clock starting from the integer
|
539
|
+
# +offset+.
|
540
|
+
def truncate_data(offset)
|
541
|
+
for t in ALL_COLUMNS
|
542
|
+
times = @times[t]
|
543
|
+
@times[t] = @times[t][offset, times.size]
|
544
|
+
@repeat = @times[t].size
|
545
|
+
end
|
546
|
+
@analysis = nil
|
547
|
+
self
|
548
|
+
end
|
549
|
+
|
550
|
+
# Find an offset from the start of the measurements in this clock to
|
551
|
+
# truncate the initial data until a stable state has been reached and
|
552
|
+
# return it as an integer.
|
553
|
+
def find_truncation_offset
|
554
|
+
truncation = self.case.truncate_data
|
555
|
+
slope_angle = self.case.truncate_data.slope_angle.abs
|
556
|
+
time = self.case.compare_time.to_sym
|
557
|
+
ms = analysis[time].measurements.reverse
|
558
|
+
offset = ms.size - 1
|
559
|
+
@slopes = []
|
560
|
+
ModuleFunctions.array_window(ms, truncation.window_size) do |data|
|
561
|
+
lr = LinearRegression.new(data)
|
562
|
+
a = lr.a
|
563
|
+
@slopes << [ offset, a ]
|
564
|
+
a.abs > slope_angle and break
|
565
|
+
offset -= 1
|
566
|
+
end
|
567
|
+
offset < 0 ? 0 : offset
|
568
|
+
end
|
569
|
+
end
|
570
|
+
|
571
|
+
# A histogram gives an overview of measurement time values.
|
572
|
+
class Histogram
|
573
|
+
# Create a Histogram for +clock+ using the measurements for +time+.
|
574
|
+
def initialize(analysis, bins)
|
575
|
+
@analysis = analysis
|
576
|
+
@bins = bins
|
577
|
+
@result = compute
|
578
|
+
end
|
579
|
+
|
580
|
+
# Number of bins for this Histogram.
|
581
|
+
attr_reader :bins
|
582
|
+
|
583
|
+
# Return the computed histogram as an array of arrays.
|
584
|
+
def to_a
|
585
|
+
@result
|
586
|
+
end
|
587
|
+
|
588
|
+
# Display this histogram to +output+, +width+ is the parameter for
|
589
|
+
# +prepare_display+
|
590
|
+
def display(output = $stdout, width = 50)
|
591
|
+
d = prepare_display(width)
|
592
|
+
for l, bar, r in d
|
593
|
+
output << "%11.5f -|%s\n" % [ (l + r) / 2.0, "*" * bar ]
|
594
|
+
end
|
595
|
+
self
|
596
|
+
end
|
597
|
+
|
598
|
+
private
|
599
|
+
|
600
|
+
# Returns an array of tuples (l, c, r) where +l+ is the left bin edge, +c+
|
601
|
+
# the +width+-normalized frequence count value, and +r+ the right bin
|
602
|
+
# edge. +width+ is usually an integer number representing the width of a
|
603
|
+
# histogram bar.
|
604
|
+
def prepare_display(width)
|
605
|
+
r = @result.reverse
|
606
|
+
factor = width.to_f / (r.transpose[1].max)
|
607
|
+
r.map { |l, c, r| [ l, (c * factor).round, r ] }
|
608
|
+
end
|
609
|
+
|
610
|
+
# Computes the histogram and returns it as an array of tuples (l, c, r).
|
611
|
+
def compute
|
612
|
+
@analysis.measurements.empty? and return []
|
613
|
+
last_r = -Infinity
|
614
|
+
min = @analysis.min
|
615
|
+
max = @analysis.max
|
616
|
+
step = (max - min) / bins.to_f
|
617
|
+
Array.new(bins) do |i|
|
618
|
+
l = min + i * step
|
619
|
+
r = min + (i + 1) * step
|
620
|
+
c = 0
|
621
|
+
@analysis.measurements.each do |x|
|
622
|
+
x > last_r and (x <= r || i == bins - 1) and c += 1
|
623
|
+
end
|
624
|
+
last_r = r
|
625
|
+
[ l, c, r ]
|
626
|
+
end
|
627
|
+
end
|
628
|
+
end
|
629
|
+
|
630
|
+
# This class is used to find the root of a function with Newton's bisection
|
631
|
+
# method.
|
632
|
+
class NewtonBisection
|
633
|
+
# Creates a NewtonBisection instance for +function+, a one-argument block.
|
634
|
+
def initialize(&function)
|
635
|
+
@function = function
|
636
|
+
end
|
637
|
+
|
638
|
+
# The function, passed into the constructor.
|
639
|
+
attr_reader :function
|
640
|
+
|
641
|
+
# Return a bracket around a root, starting from the initial +range+. The
|
642
|
+
# method returns nil, if no such bracket around a root could be found after
|
643
|
+
# +n+ tries with the scaling +factor+.
|
644
|
+
def bracket(range = -1..1, n = 50, factor = 1.6)
|
645
|
+
x1, x2 = range.first.to_f, range.last.to_f
|
646
|
+
x1 >= x2 and raise ArgumentError, "bad initial range #{range}"
|
647
|
+
f1, f2 = @function[x1], @function[x2]
|
648
|
+
n.times do
|
649
|
+
f1 * f2 < 0 and return x1..x2
|
650
|
+
if f1.abs < f2.abs
|
651
|
+
f1 = @function[x1 += factor * (x1 - x2)]
|
652
|
+
else
|
653
|
+
f2 = @function[x2 += factor * (x2 - x1)]
|
654
|
+
end
|
655
|
+
end
|
656
|
+
return
|
657
|
+
end
|
658
|
+
|
659
|
+
# Find the root of function in +range+ and return it. The method raises a
|
660
|
+
# BullshitException, if no such root could be found after +n+ tries and in
|
661
|
+
# the +epsilon+ environment.
|
662
|
+
def solve(range = nil, n = 1 << 16, epsilon = 1E-16)
|
663
|
+
if range
|
664
|
+
x1, x2 = range.first.to_f, range.last.to_f
|
665
|
+
x1 >= x2 and raise ArgumentError, "bad initial range #{range}"
|
666
|
+
elsif range = bracket
|
667
|
+
x1, x2 = range.first, range.last
|
668
|
+
else
|
669
|
+
raise ArgumentError, "bracket could not be determined"
|
670
|
+
end
|
671
|
+
f = @function[x1]
|
672
|
+
fmid = @function[x2]
|
673
|
+
f * fmid >= 0 and raise ArgumentError, "root must be bracketed in #{range}"
|
674
|
+
root = if f < 0
|
675
|
+
dx = x2 - x1
|
676
|
+
x1
|
677
|
+
else
|
678
|
+
dx = x1 - x2
|
679
|
+
x2
|
680
|
+
end
|
681
|
+
n.times do
|
682
|
+
fmid = @function[xmid = root + (dx *= 0.5)]
|
683
|
+
fmid < 0 and root = xmid
|
684
|
+
dx.abs < epsilon or fmid == 0 and return root
|
685
|
+
end
|
686
|
+
raise BullshitException, "too many iterations (#{n})"
|
687
|
+
end
|
688
|
+
end
|
689
|
+
|
690
|
+
module Functions
|
691
|
+
module_function
|
692
|
+
|
693
|
+
include Math
|
694
|
+
extend Math
|
695
|
+
|
696
|
+
LANCZOS_COEFFICIENTS = [
|
697
|
+
0.99999999999999709182,
|
698
|
+
57.156235665862923517,
|
699
|
+
-59.597960355475491248,
|
700
|
+
14.136097974741747174,
|
701
|
+
-0.49191381609762019978,
|
702
|
+
0.33994649984811888699e-4,
|
703
|
+
0.46523628927048575665e-4,
|
704
|
+
-0.98374475304879564677e-4,
|
705
|
+
0.15808870322491248884e-3,
|
706
|
+
-0.21026444172410488319e-3,
|
707
|
+
0.21743961811521264320e-3,
|
708
|
+
-0.16431810653676389022e-3,
|
709
|
+
0.84418223983852743293e-4,
|
710
|
+
-0.26190838401581408670e-4,
|
711
|
+
0.36899182659531622704e-5,
|
712
|
+
]
|
713
|
+
|
714
|
+
HALF_LOG_2_PI = 0.5 * log(2 * Math::PI)
|
715
|
+
|
716
|
+
# Returns the natural logarithm of Euler gamma function value for +x+ using
|
717
|
+
# the Lanczos approximation.
|
718
|
+
if method_defined?(:lgamma)
|
719
|
+
def log_gamma(x)
|
720
|
+
lgamma(x).first
|
721
|
+
end
|
722
|
+
else
|
723
|
+
def log_gamma(x)
|
724
|
+
if x.nan? || x <= 0
|
725
|
+
0 / 0.0
|
726
|
+
else
|
727
|
+
sum = 0.0
|
728
|
+
(LANCZOS_COEFFICIENTS.size - 1).downto(1) do |i|
|
729
|
+
sum += LANCZOS_COEFFICIENTS[i] / (x + i)
|
730
|
+
end
|
731
|
+
sum += LANCZOS_COEFFICIENTS[0]
|
732
|
+
tmp = x + 607.0 / 128 + 0.5
|
733
|
+
(x + 0.5) * log(tmp) - tmp + HALF_LOG_2_PI + log(sum / x)
|
734
|
+
end
|
735
|
+
rescue Errno::ERANGE, Errno::EDOM
|
736
|
+
0 / 0.0
|
737
|
+
end
|
738
|
+
end
|
739
|
+
|
740
|
+
# Returns the natural logarithm of the beta function value for +(a, b)+.
|
741
|
+
def log_beta(a, b)
|
742
|
+
log_gamma(a) + log_gamma(b) - log_gamma(a + b)
|
743
|
+
rescue Errno::ERANGE, Errno::EDOM
|
744
|
+
0 / 0.0
|
745
|
+
end
|
746
|
+
|
747
|
+
# Return an approximation value of Euler's regularized beta function for
|
748
|
+
# +x+, +a+, and +b+ with an error <= +epsilon+, but only iterate
|
749
|
+
# +max_iterations+-times.
|
750
|
+
def beta_regularized(x, a, b, epsilon = 1E-16, max_iterations = 1 << 16)
|
751
|
+
x, a, b = x.to_f, a.to_f, b.to_f
|
752
|
+
case
|
753
|
+
when a.nan? || b.nan? || x.nan? || a <= 0 || b <= 0 || x < 0 || x > 1
|
754
|
+
0 / 0.0
|
755
|
+
when x > (a + 1) / (a + b + 2)
|
756
|
+
1 - beta_regularized(1 - x, b, a, epsilon, max_iterations)
|
757
|
+
else
|
758
|
+
fraction = ContinuedFraction.for_b do |n, x|
|
759
|
+
if n % 2 == 0
|
760
|
+
m = n / 2.0
|
761
|
+
(m * (b - m) * x) / ((a + (2 * m) - 1) * (a + (2 * m)))
|
762
|
+
else
|
763
|
+
m = (n - 1) / 2.0
|
764
|
+
-((a + m) * (a + b + m) * x) / ((a + 2 * m) * (a + 2 * m + 1))
|
765
|
+
end
|
766
|
+
end
|
767
|
+
exp(a * log(x) + b * log(1.0 - x) - log(a) - log_beta(a, b)) /
|
768
|
+
fraction[x, epsilon, max_iterations]
|
769
|
+
end
|
770
|
+
rescue Errno::ERANGE, Errno::EDOM
|
771
|
+
0 / 0.0
|
772
|
+
end
|
773
|
+
|
774
|
+
# Return an approximation of the regularized gammaP function for +x+ and
|
775
|
+
# +a+ with an error of <= +epsilon+, but only iterate
|
776
|
+
# +max_iterations+-times.
|
777
|
+
def gammaP_regularized(x, a, epsilon = 1E-16, max_iterations = 1 << 16)
|
778
|
+
x, a = x.to_f, a.to_f
|
779
|
+
case
|
780
|
+
when a.nan? || x.nan? || a <= 0 || x < 0
|
781
|
+
0 / 0.0
|
782
|
+
when x == 0
|
783
|
+
0.0
|
784
|
+
when 1 <= a && a < x
|
785
|
+
1 - gammaQ_regularized(x, a, epsilon, max_iterations)
|
786
|
+
else
|
787
|
+
n = 0
|
788
|
+
an = 1 / a
|
789
|
+
sum = an
|
790
|
+
while an.abs > epsilon && n < max_iterations
|
791
|
+
n += 1
|
792
|
+
an *= x / (a + n)
|
793
|
+
sum += an
|
794
|
+
end
|
795
|
+
if n >= max_iterations
|
796
|
+
raise Errno::ERANGE
|
797
|
+
else
|
798
|
+
exp(-x + a * log(x) - log_gamma(a)) * sum
|
799
|
+
end
|
800
|
+
end
|
801
|
+
rescue Errno::ERANGE, Errno::EDOM
|
802
|
+
0 / 0.0
|
803
|
+
end
|
804
|
+
|
805
|
+
# Return an approximation of the regularized gammaQ function for +x+ and
|
806
|
+
# +a+ with an error of <= +epsilon+, but only iterate
|
807
|
+
# +max_iterations+-times.
|
808
|
+
def gammaQ_regularized(x, a, epsilon = 1E-16, max_iterations = 1 << 16)
|
809
|
+
x, a = x.to_f, a.to_f
|
810
|
+
case
|
811
|
+
when a.nan? || x.nan? || a <= 0 || x < 0
|
812
|
+
0 / 0.0
|
813
|
+
when x == 0
|
814
|
+
1.0
|
815
|
+
when a > x || a < 1
|
816
|
+
1 - gammaP_regularized(x, a, epsilon, max_iterations)
|
817
|
+
else
|
818
|
+
fraction = ContinuedFraction.for_a do |n, x|
|
819
|
+
(2 * n + 1) - a + x
|
820
|
+
end.for_b do |n, x|
|
821
|
+
n * (a - n)
|
822
|
+
end
|
823
|
+
exp(-x + a * log(x) - log_gamma(a)) *
|
824
|
+
fraction[x, epsilon, max_iterations] ** -1
|
825
|
+
end
|
826
|
+
rescue Errno::ERANGE, Errno::EDOM
|
827
|
+
0 / 0.0
|
828
|
+
end
|
829
|
+
|
830
|
+
ROOT2 = sqrt(2)
|
831
|
+
|
832
|
+
A = -8 * (Math::PI - 3) / (3 * Math::PI * (Math::PI - 4))
|
833
|
+
|
834
|
+
# Returns an approximate value for the error function's value for +x+.
|
835
|
+
def erf(x)
|
836
|
+
r = sqrt(1 - exp(-x ** 2 * (4 / Math::PI + A * x ** 2) / (1 + A * x ** 2)))
|
837
|
+
x < 0 ? -r : r
|
838
|
+
end unless method_defined?(:erf)
|
839
|
+
end
|
840
|
+
|
841
|
+
# This class is used to compute the T-Distribution.
|
842
|
+
class TDistribution
|
843
|
+
include Functions
|
844
|
+
|
845
|
+
# Returns a TDistribution instance for the degrees of freedom +df+.
|
846
|
+
def initialize(df)
|
847
|
+
@df = df
|
848
|
+
end
|
849
|
+
|
850
|
+
# Degrees of freedom.
|
851
|
+
attr_reader :df
|
852
|
+
|
853
|
+
# Returns the cumulative probability (p-value) of the TDistribution for the
|
854
|
+
# t-value +x+.
|
855
|
+
def probability(x)
|
856
|
+
if x == 0
|
857
|
+
0.5
|
858
|
+
else
|
859
|
+
t = beta_regularized(@df / (@df + x ** 2.0), 0.5 * @df, 0.5)
|
860
|
+
if x < 0.0
|
861
|
+
0.5 * t
|
862
|
+
else
|
863
|
+
1 - 0.5 * t
|
864
|
+
end
|
865
|
+
end
|
866
|
+
end
|
867
|
+
|
868
|
+
# Returns the inverse cumulative probability (t-value) of the TDistribution
|
869
|
+
# for the probability +p+.
|
870
|
+
def inverse_probability(p)
|
871
|
+
case
|
872
|
+
when p <= 0
|
873
|
+
-1 / 0.0
|
874
|
+
when p >= 1
|
875
|
+
1 / 0.0
|
876
|
+
else
|
877
|
+
begin
|
878
|
+
bisect = NewtonBisection.new { |x| probability(x) - p }
|
879
|
+
range = bisect.bracket(-10..10)
|
880
|
+
bisect.solve(range, 1_000_000)
|
881
|
+
rescue
|
882
|
+
0 / 0.0
|
883
|
+
end
|
884
|
+
end
|
885
|
+
end
|
886
|
+
end
|
887
|
+
|
888
|
+
# This class is used to compute the Normal Distribution.
|
889
|
+
class NormalDistribution
|
890
|
+
include Functions
|
891
|
+
|
892
|
+
# Creates a NormalDistribution instance for the values +mu+ and +sigma+.
|
893
|
+
def initialize(mu = 0.0, sigma = 1.0)
|
894
|
+
@mu, @sigma = mu.to_f, sigma.to_f
|
895
|
+
end
|
896
|
+
|
897
|
+
attr_reader :mu
|
898
|
+
|
899
|
+
attr_reader :sigma
|
900
|
+
|
901
|
+
# Returns the cumulative probability (p-value) of the NormalDistribution
|
902
|
+
# for the value +x+.
|
903
|
+
def probability(x)
|
904
|
+
0.5 * (1 + erf((x - @mu) / (@sigma * ROOT2)))
|
905
|
+
end
|
906
|
+
|
907
|
+
# Returns the inverse cumulative probability value of the
|
908
|
+
# NormalDistribution for the probability +p+.
|
909
|
+
def inverse_probability(p)
|
910
|
+
case
|
911
|
+
when p <= 0
|
912
|
+
-1 / 0.0
|
913
|
+
when p >= 1
|
914
|
+
1 / 0.0
|
915
|
+
when p == 0.5 # This is a bit sloppy, maybe improve this later.
|
916
|
+
@mu
|
917
|
+
else
|
918
|
+
begin
|
919
|
+
NewtonBisection.new { |x| probability(x) - p }.solve(nil, 1_000_000)
|
920
|
+
rescue
|
921
|
+
0 / 0.0
|
922
|
+
end
|
923
|
+
end
|
924
|
+
end
|
925
|
+
end
|
926
|
+
|
927
|
+
STD_NORMAL_DISTRIBUTION = NormalDistribution.new
|
928
|
+
|
929
|
+
# This class is used to compute the Chi-Square Distribution.
|
930
|
+
class ChiSquareDistribution
|
931
|
+
include Functions
|
932
|
+
|
933
|
+
# Creates a ChiSquareDistribution for +df+ degrees of freedom.
|
934
|
+
def initialize(df)
|
935
|
+
@df = df
|
936
|
+
@df_half = @df / 2.0
|
937
|
+
end
|
938
|
+
|
939
|
+
attr_reader :df
|
940
|
+
|
941
|
+
# Returns the cumulative probability (p-value) of the ChiSquareDistribution
|
942
|
+
# for the value +x+.
|
943
|
+
def probability(x)
|
944
|
+
if x < 0
|
945
|
+
0.0
|
946
|
+
else
|
947
|
+
gammaP_regularized(x / 2, @df_half)
|
948
|
+
end
|
949
|
+
end
|
950
|
+
|
951
|
+
# Returns the inverse cumulative probability value of the
|
952
|
+
# NormalDistribution for the probability +p+.
|
953
|
+
def inverse_probability(p)
|
954
|
+
case
|
955
|
+
when p <= 0, p >= 1
|
956
|
+
0.0
|
957
|
+
else
|
958
|
+
begin
|
959
|
+
bisect = NewtonBisection.new { |x| probability(x) - p }
|
960
|
+
range = bisect.bracket 0.5..10
|
961
|
+
bisect.solve(range, 1_000_000)
|
962
|
+
rescue
|
963
|
+
0 / 0.0
|
964
|
+
end
|
965
|
+
end
|
966
|
+
end
|
967
|
+
end
|
968
|
+
|
969
|
+
# This class computes a linear regression for the given image and domain data
|
970
|
+
# sets.
|
971
|
+
class LinearRegression
|
972
|
+
def initialize(image, domain = (0...image.size).to_a)
|
973
|
+
image.size != domain.size and raise ArgumentError,
|
974
|
+
"image and domain have unequal sizes"
|
975
|
+
@image, @domain = image, domain
|
976
|
+
compute
|
977
|
+
end
|
978
|
+
|
979
|
+
# The image data as an array.
|
980
|
+
attr_reader :image
|
981
|
+
|
982
|
+
# The domain data as an array.
|
983
|
+
attr_reader :domain
|
984
|
+
|
985
|
+
# The slope of the line.
|
986
|
+
attr_reader :a
|
987
|
+
|
988
|
+
# The offset of the line.
|
989
|
+
attr_reader :b
|
990
|
+
|
991
|
+
# Return true if the slope of the underlying data (not the sample data
|
992
|
+
# passed into the constructor of this LinearRegression instance) is likely
|
993
|
+
# (with alpha level _alpha_) to be zero.
|
994
|
+
def slope_zero?(alpha = 0.05)
|
995
|
+
df = @image.size - 2
|
996
|
+
return true if df <= 0 # not enough values to check
|
997
|
+
t = tvalue(alpha)
|
998
|
+
td = TDistribution.new df
|
999
|
+
t.abs <= td.inverse_probability(1 - alpha.abs / 2.0).abs
|
1000
|
+
end
|
1001
|
+
|
1002
|
+
# Returns the residues of this linear regression in relation to the given
|
1003
|
+
# domain and image.
|
1004
|
+
def residues
|
1005
|
+
result = []
|
1006
|
+
@domain.zip(@image) do |x, y|
|
1007
|
+
result << y - (@a * x + @b)
|
1008
|
+
end
|
1009
|
+
result
|
1010
|
+
end
|
1011
|
+
|
1012
|
+
private
|
1013
|
+
|
1014
|
+
def compute
|
1015
|
+
size = @image.size
|
1016
|
+
sum_xx = sum_xy = sum_x = sum_y = 0.0
|
1017
|
+
@domain.zip(@image) do |x, y|
|
1018
|
+
x += 1
|
1019
|
+
sum_xx += x ** 2
|
1020
|
+
sum_xy += x * y
|
1021
|
+
sum_x += x
|
1022
|
+
sum_y += y
|
1023
|
+
end
|
1024
|
+
@a = (size * sum_xy - sum_x * sum_y) / (size * sum_xx - sum_x ** 2)
|
1025
|
+
@b = (sum_y - @a * sum_x) / size
|
1026
|
+
self
|
1027
|
+
end
|
1028
|
+
|
1029
|
+
def tvalue(alpha = 0.05)
|
1030
|
+
df = @image.size - 2
|
1031
|
+
return 0.0 if df <= 0
|
1032
|
+
sse_y = 0.0
|
1033
|
+
@domain.zip(@image) do |x, y|
|
1034
|
+
f_x = a * x + b
|
1035
|
+
sse_y += (y - f_x) ** 2
|
1036
|
+
end
|
1037
|
+
mean = @image.inject(0.0) { |s, y| s + y } / @image.size
|
1038
|
+
sse_x = @domain.inject(0.0) { |s, x| s + (x - mean) ** 2 }
|
1039
|
+
t = a / (Math.sqrt(sse_y / df) / Math.sqrt(sse_x))
|
1040
|
+
t.nan? ? 0.0 : t
|
1041
|
+
end
|
1042
|
+
end
|
1043
|
+
|
1044
|
+
# This class is used to analyse the time measurements and compute their
|
1045
|
+
# statistics.
|
1046
|
+
class Analysis
|
1047
|
+
def initialize(measurements)
|
1048
|
+
@measurements = measurements
|
1049
|
+
@measurements.freeze
|
1050
|
+
end
|
1051
|
+
|
1052
|
+
# Returns the array of measurements.
|
1053
|
+
attr_reader :measurements
|
1054
|
+
|
1055
|
+
# Returns the number of measurements, on which the analysis is based.
|
1056
|
+
def size
|
1057
|
+
@measurements.size
|
1058
|
+
end
|
1059
|
+
|
1060
|
+
# Returns the variance of the measurements.
|
1061
|
+
def variance
|
1062
|
+
@variance ||= sum_of_squares / size
|
1063
|
+
end
|
1064
|
+
|
1065
|
+
# Returns the sample_variance of the measurements.
|
1066
|
+
def sample_variance
|
1067
|
+
@sample_variance ||= size > 1 ? sum_of_squares / (size - 1.0) : 0.0
|
1068
|
+
end
|
1069
|
+
|
1070
|
+
# Returns the sum of squares (the sum of the squared deviations) of the
|
1071
|
+
# measurements.
|
1072
|
+
def sum_of_squares
|
1073
|
+
@sum_of_squares ||= @measurements.inject(0.0) { |s, t| s + (t - arithmetic_mean) ** 2 }
|
1074
|
+
end
|
1075
|
+
|
1076
|
+
# Returns the standard deviation of the measurements.
|
1077
|
+
def standard_deviation
|
1078
|
+
@sample_deviation ||= Math.sqrt(variance)
|
1079
|
+
end
|
1080
|
+
|
1081
|
+
# Returns the standard deviation of the measurements in percentage of the
|
1082
|
+
# arithmetic mean.
|
1083
|
+
def standard_deviation_percentage
|
1084
|
+
@standard_deviation_percentage ||= 100.0 * standard_deviation / arithmetic_mean
|
1085
|
+
end
|
1086
|
+
|
1087
|
+
# Returns the sample standard deviation of the measurements.
|
1088
|
+
def sample_standard_deviation
|
1089
|
+
@sample_standard_deviation ||= Math.sqrt(sample_variance)
|
1090
|
+
end
|
1091
|
+
|
1092
|
+
# Returns the sample standard deviation of the measurements in percentage
|
1093
|
+
# of the arithmetic mean.
|
1094
|
+
def sample_standard_deviation_percentage
|
1095
|
+
@sample_standard_deviation_percentage ||= 100.0 * sample_standard_deviation / arithmetic_mean
|
1096
|
+
end
|
1097
|
+
|
1098
|
+
# Returns the sum of all measurements.
|
1099
|
+
def sum
|
1100
|
+
@sum ||= @measurements.inject(0.0) { |s, t| s + t }
|
1101
|
+
end
|
1102
|
+
|
1103
|
+
# Returns the arithmetic mean of the measurements.
|
1104
|
+
def arithmetic_mean
|
1105
|
+
@arithmetic_mean ||= sum / size
|
1106
|
+
end
|
1107
|
+
|
1108
|
+
alias mean arithmetic_mean
|
1109
|
+
|
1110
|
+
# Returns the harmonic mean of the measurements. If any of the measurements
|
1111
|
+
# is less than or equal to 0.0, this method returns NaN.
|
1112
|
+
def harmonic_mean
|
1113
|
+
@harmonic_mean ||= (
|
1114
|
+
sum = @measurements.inject(0.0) { |s, t|
|
1115
|
+
if t > 0
|
1116
|
+
s + 1.0 / t
|
1117
|
+
else
|
1118
|
+
break nil
|
1119
|
+
end
|
1120
|
+
}
|
1121
|
+
sum ? size / sum : 0 / 0.0
|
1122
|
+
)
|
1123
|
+
end
|
1124
|
+
|
1125
|
+
# Returns the geometric mean of the measurements. If any of the
|
1126
|
+
# measurements is less than 0.0, this method returns NaN.
|
1127
|
+
def geometric_mean
|
1128
|
+
@geometric_mean ||= (
|
1129
|
+
sum = @measurements.inject(0.0) { |s, t|
|
1130
|
+
case
|
1131
|
+
when t > 0
|
1132
|
+
s + Math.log(t)
|
1133
|
+
when t == 0
|
1134
|
+
break :null
|
1135
|
+
else
|
1136
|
+
break nil
|
1137
|
+
end
|
1138
|
+
}
|
1139
|
+
case sum
|
1140
|
+
when :null
|
1141
|
+
0.0
|
1142
|
+
when Float
|
1143
|
+
Math.exp(sum / size)
|
1144
|
+
else
|
1145
|
+
0 / 0.0
|
1146
|
+
end
|
1147
|
+
)
|
1148
|
+
end
|
1149
|
+
|
1150
|
+
# Returns the minimum of the measurements.
|
1151
|
+
def min
|
1152
|
+
@min ||= @measurements.min
|
1153
|
+
end
|
1154
|
+
|
1155
|
+
# Returns the maximum of the measurements.
|
1156
|
+
def max
|
1157
|
+
@max ||= @measurements.max
|
1158
|
+
end
|
1159
|
+
|
1160
|
+
# Returns the +p+-percentile of the measurements.
|
1161
|
+
# There are many methods to compute the percentile, this method uses the
|
1162
|
+
# the weighted average at x_(n + 1)p, which allows p to be in 0...100
|
1163
|
+
# (excluding the 100).
|
1164
|
+
def percentile(p = 50)
|
1165
|
+
(0...100).include?(p) or
|
1166
|
+
raise ArgumentError, "p = #{p}, but has to be in (0...100)"
|
1167
|
+
p /= 100.0
|
1168
|
+
@sorted ||= @measurements.sort
|
1169
|
+
r = p * (@sorted.size + 1)
|
1170
|
+
r_i = r.to_i
|
1171
|
+
r_f = r - r_i
|
1172
|
+
if r_i >= 1
|
1173
|
+
result = @sorted[r_i - 1]
|
1174
|
+
if r_i < @sorted.size
|
1175
|
+
result += r_f * (@sorted[r_i] - @sorted[r_i - 1])
|
1176
|
+
end
|
1177
|
+
else
|
1178
|
+
result = @sorted[0]
|
1179
|
+
end
|
1180
|
+
result
|
1181
|
+
end
|
1182
|
+
|
1183
|
+
alias median percentile
|
1184
|
+
|
1185
|
+
# Use an approximation of the Welch-Satterthwaite equation to compute the
|
1186
|
+
# degrees of freedom for Welch's t-test.
|
1187
|
+
def compute_welch_df(other)
|
1188
|
+
(sample_variance / size + other.sample_variance / other.size) ** 2 / (
|
1189
|
+
(sample_variance ** 2 / (size ** 2 * (size - 1))) +
|
1190
|
+
(other.sample_variance ** 2 / (other.size ** 2 * (other.size - 1))))
|
1191
|
+
end
|
1192
|
+
|
1193
|
+
# Returns the t value of the Welch's t-test between this Analysis
|
1194
|
+
# instance and the +other+.
|
1195
|
+
def t_welch(other)
|
1196
|
+
signal = arithmetic_mean - other.arithmetic_mean
|
1197
|
+
noise = Math.sqrt(sample_variance / size +
|
1198
|
+
other.sample_variance / other.size)
|
1199
|
+
signal / noise
|
1200
|
+
rescue Errno::EDOM
|
1201
|
+
0.0
|
1202
|
+
end
|
1203
|
+
|
1204
|
+
# Returns an estimation of the common standard deviation of the
|
1205
|
+
# measurements of this and +other+.
|
1206
|
+
def common_standard_deviation(other)
|
1207
|
+
Math.sqrt(common_variance(other))
|
1208
|
+
end
|
1209
|
+
|
1210
|
+
# Returns an estimation of the common variance of the measurements of this
|
1211
|
+
# and +other+.
|
1212
|
+
def common_variance(other)
|
1213
|
+
(size - 1) * sample_variance + (other.size - 1) * other.sample_variance /
|
1214
|
+
(size + other.size - 2)
|
1215
|
+
end
|
1216
|
+
|
1217
|
+
# Compute the # degrees of freedom for Student's t-test.
|
1218
|
+
def compute_student_df(other)
|
1219
|
+
size + other.size - 2
|
1220
|
+
end
|
1221
|
+
|
1222
|
+
# Returns the t value of the Student's t-test between this Analysis
|
1223
|
+
# instance and the +other+.
|
1224
|
+
def t_student(other)
|
1225
|
+
signal = arithmetic_mean - other.arithmetic_mean
|
1226
|
+
noise = common_standard_deviation(other) *
|
1227
|
+
Math.sqrt(size ** -1 + size ** -1)
|
1228
|
+
rescue Errno::EDOM
|
1229
|
+
0.0
|
1230
|
+
end
|
1231
|
+
|
1232
|
+
# Compute a sample size, that will more likely yield a mean difference
|
1233
|
+
# between this instance's measurements and those of +other+. Use +alpha+
|
1234
|
+
# and +beta+ as levels for the first- and second-order errors.
|
1235
|
+
def suggested_sample_size(other, alpha = 0.05, beta = 0.05)
|
1236
|
+
alpha, beta = alpha.abs, beta.abs
|
1237
|
+
signal = arithmetic_mean - other.arithmetic_mean
|
1238
|
+
df = size + other.size - 2
|
1239
|
+
pooled_variance_estimate = (sum_of_squares + other.sum_of_squares) / df
|
1240
|
+
td = TDistribution.new df
|
1241
|
+
(((td.inverse_probability(alpha) + td.inverse_probability(beta)) *
|
1242
|
+
Math.sqrt(pooled_variance_estimate)) / signal) ** 2
|
1243
|
+
end
|
1244
|
+
|
1245
|
+
# Return true, if the Analysis instance covers the +other+, that is their
|
1246
|
+
# arithmetic mean value is most likely to be equal for the +alpha+ error
|
1247
|
+
# level.
|
1248
|
+
def cover?(other, alpha = 0.05)
|
1249
|
+
t = t_welch(other)
|
1250
|
+
td = TDistribution.new(compute_welch_df(other))
|
1251
|
+
t.abs < td.inverse_probability(1 - alpha.abs / 2.0)
|
1252
|
+
end
|
1253
|
+
|
1254
|
+
# Return the confidence interval for the arithmetic mean with alpha level +alpha+ of
|
1255
|
+
# the measurements of this Analysis instance as a Range object.
|
1256
|
+
def confidence_interval(alpha = 0.05)
|
1257
|
+
td = TDistribution.new(size - 1)
|
1258
|
+
t = td.inverse_probability(alpha / 2).abs
|
1259
|
+
delta = t * sample_standard_deviation / Math.sqrt(size)
|
1260
|
+
(arithmetic_mean - delta)..(arithmetic_mean + delta)
|
1261
|
+
end
|
1262
|
+
|
1263
|
+
# Returns the array of autovariances (of length size - 1).
|
1264
|
+
def autovariance
|
1265
|
+
Array.new(size - 1) do |k|
|
1266
|
+
s = 0.0
|
1267
|
+
0.upto(size - k - 1) do |i|
|
1268
|
+
s += (@measurements[i] - arithmetic_mean) * (@measurements[i + k] - arithmetic_mean)
|
1269
|
+
end
|
1270
|
+
s / size
|
1271
|
+
end
|
1272
|
+
end
|
1273
|
+
|
1274
|
+
# Returns the array of autocorrelation values c_k / c_0 (of length size -
|
1275
|
+
# 1).
|
1276
|
+
def autocorrelation
|
1277
|
+
c = autovariance
|
1278
|
+
Array.new(c.size) { |k| c[k] / c[0] }
|
1279
|
+
end
|
1280
|
+
|
1281
|
+
# Returns the d-value for the Durbin-Watson statistic. The value is d << 2
|
1282
|
+
# for positive, d >> 2 for negative and d around 2 for no autocorrelation.
|
1283
|
+
def durbin_watson_statistic
|
1284
|
+
e = linear_regression.residues
|
1285
|
+
e.size <= 1 and return 2.0
|
1286
|
+
(1...e.size).inject(0.0) { |s, i| s + (e[i] - e[i - 1]) ** 2 } /
|
1287
|
+
e.inject(0.0) { |s, x| s + x ** 2 }
|
1288
|
+
end
|
1289
|
+
|
1290
|
+
# Returns the q value of the Ljung-Box statistic for the number of lags
|
1291
|
+
# +lags+. A higher value might indicate autocorrelation in the measurements of
|
1292
|
+
# this Analysis instance. This method returns nil if there weren't enough
|
1293
|
+
# (at least lags) lags available.
|
1294
|
+
def ljung_box_statistic(lags = 20)
|
1295
|
+
r = autocorrelation
|
1296
|
+
lags >= r.size and return
|
1297
|
+
n = size
|
1298
|
+
n * (n + 2) * (1..lags).inject(0.0) { |s, i| s + r[i] ** 2 / (n - i) }
|
1299
|
+
end
|
1300
|
+
|
1301
|
+
# This method tries to detect autocorrelation with the Ljung-Box
|
1302
|
+
# statistic. If enough lags can be considered it returns a hash with
|
1303
|
+
# results, otherwise nil is returned. The keys are
|
1304
|
+
# :lags: the number of lags,
|
1305
|
+
# :alpha_level: the alpha level for the test,
|
1306
|
+
# :q: the value of the ljung_box_statistic,
|
1307
|
+
# :p: the p-value computed, if p is higher than alpha no correlation was detected,
|
1308
|
+
# :detected: true if a correlation was found.
|
1309
|
+
def detect_autocorrelation(lags = 20, alpha_level = 0.05)
|
1310
|
+
if q = ljung_box_statistic(lags)
|
1311
|
+
p = ChiSquareDistribution.new(lags).probability(q)
|
1312
|
+
return {
|
1313
|
+
:lags => lags,
|
1314
|
+
:alpha_level => alpha_level,
|
1315
|
+
:q => q,
|
1316
|
+
:p => p,
|
1317
|
+
:detected => p >= 1 - alpha_level,
|
1318
|
+
}
|
1319
|
+
end
|
1320
|
+
end
|
1321
|
+
|
1322
|
+
# Return a result hash with the number of :very_low, :low, :high, and
|
1323
|
+
# :very_high outliers, determined by the box plotting algorithm run with
|
1324
|
+
# :median and :iqr parameters. If no outliers were found or the iqr is
|
1325
|
+
# less than epsilon, nil is returned.
|
1326
|
+
def detect_outliers(factor = 3.0, epsilon = 1E-5)
|
1327
|
+
half_factor = factor / 2.0
|
1328
|
+
quartile1 = percentile(25)
|
1329
|
+
quartile3 = percentile(75)
|
1330
|
+
iqr = quartile3 - quartile1
|
1331
|
+
iqr < epsilon and return
|
1332
|
+
result = @measurements.inject(Hash.new(0)) do |h, t|
|
1333
|
+
extreme =
|
1334
|
+
case t
|
1335
|
+
when -Infinity..(quartile1 - factor * iqr)
|
1336
|
+
:very_low
|
1337
|
+
when (quartile1 - factor * iqr)..(quartile1 - half_factor * iqr)
|
1338
|
+
:low
|
1339
|
+
when (quartile1 + half_factor * iqr)..(quartile3 + factor * iqr)
|
1340
|
+
:high
|
1341
|
+
when (quartile3 + factor * iqr)..Infinity
|
1342
|
+
:very_high
|
1343
|
+
end and h[extreme] += 1
|
1344
|
+
h
|
1345
|
+
end
|
1346
|
+
unless result.empty?
|
1347
|
+
result[:median] = median
|
1348
|
+
result[:iqr] = iqr
|
1349
|
+
result[:factor] = factor
|
1350
|
+
result
|
1351
|
+
end
|
1352
|
+
end
|
1353
|
+
|
1354
|
+
# Returns the LinearRegression object for the equation a * x + b which
|
1355
|
+
# represents the line computed by the linear regression algorithm.
|
1356
|
+
def linear_regression
|
1357
|
+
@linear_regression ||= LinearRegression.new @measurements
|
1358
|
+
end
|
1359
|
+
|
1360
|
+
# Returns a Histogram instance with +bins+ as the number of bins for this
|
1361
|
+
# analysis' measurements.
|
1362
|
+
def histogram(bins)
|
1363
|
+
Histogram.new(self, bins)
|
1364
|
+
end
|
1365
|
+
end
|
1366
|
+
|
1367
|
+
CaseMethod = Struct.new(:name, :case, :clock)
|
1368
|
+
|
1369
|
+
# This class' instance represents a method to be benchmarked.
|
1370
|
+
class CaseMethod
|
1371
|
+
# Return the short name of this CaseMethod instance, that is without the
|
1372
|
+
# "benchmark_" prefix, e. g. "foo".
|
1373
|
+
def short_name
|
1374
|
+
@short_name ||= name.sub(/\Abenchmark_/, '')
|
1375
|
+
end
|
1376
|
+
|
1377
|
+
# The comment for this method.
|
1378
|
+
attr_accessor :comment
|
1379
|
+
|
1380
|
+
# Returns the long_name of this CaseMethod of the form Foo#bar.
|
1381
|
+
def long_name
|
1382
|
+
result = "#{self.case}##{short_name}"
|
1383
|
+
result = "#{result} (#{comment})" if comment
|
1384
|
+
result
|
1385
|
+
end
|
1386
|
+
|
1387
|
+
# Return the setup_name, e. g. "setup_foo".
|
1388
|
+
def setup_name
|
1389
|
+
'setup_' + short_name
|
1390
|
+
end
|
1391
|
+
|
1392
|
+
# Return the before_name, e. g. "before_foo".
|
1393
|
+
def before_name
|
1394
|
+
'before_' + short_name
|
1395
|
+
end
|
1396
|
+
|
1397
|
+
# Return the after_name, e. g. "after_foo".
|
1398
|
+
def after_name
|
1399
|
+
'after_' + short_name
|
1400
|
+
end
|
1401
|
+
|
1402
|
+
# Return the teardown_name, e. g. "teardown_foo".
|
1403
|
+
def teardown_name
|
1404
|
+
'teardown_' + short_name
|
1405
|
+
end
|
1406
|
+
|
1407
|
+
# Return true if this CaseMethod#clock covers other.clock.
|
1408
|
+
def cover?(other)
|
1409
|
+
clock.cover?(other.clock)
|
1410
|
+
end
|
1411
|
+
|
1412
|
+
# Call before method of this CaseMethod before benchmarking it.
|
1413
|
+
def before_run
|
1414
|
+
if self.case.respond_to? before_name
|
1415
|
+
$DEBUG and warn "Calling #{before_name}."
|
1416
|
+
self.case.__send__(before_name)
|
1417
|
+
end
|
1418
|
+
end
|
1419
|
+
|
1420
|
+
# Call after method of this CaseMethod after benchmarking it.
|
1421
|
+
def after_run
|
1422
|
+
if self.case.respond_to? after_name
|
1423
|
+
$DEBUG and warn "Calling #{after_name}."
|
1424
|
+
self.case.__send__(after_name)
|
1425
|
+
end
|
1426
|
+
end
|
1427
|
+
|
1428
|
+
# Return the file name for +type+ with +suffix+ (if any) for this clock.
|
1429
|
+
def file_path(type = nil, suffix = '.dat')
|
1430
|
+
name = self.case.class.benchmark_name.dup
|
1431
|
+
name << '#' << short_name
|
1432
|
+
type and name << '-' << type
|
1433
|
+
name << suffix
|
1434
|
+
File.expand_path(name, self.case.class.output_dir)
|
1435
|
+
end
|
1436
|
+
|
1437
|
+
# Load the data of file +fp+ into this clock.
|
1438
|
+
def load(fp = file_path)
|
1439
|
+
self.clock = self.case.class.clock.new self
|
1440
|
+
$DEBUG and warn "Loading '#{fp}' into clock."
|
1441
|
+
File.open(fp, 'r') do |f|
|
1442
|
+
f.each do |line|
|
1443
|
+
line.chomp!
|
1444
|
+
line =~ /^\s*#/ and next
|
1445
|
+
clock << line.split(/\t/)
|
1446
|
+
end
|
1447
|
+
end
|
1448
|
+
self
|
1449
|
+
end
|
1450
|
+
end
|
1451
|
+
|
1452
|
+
module CommonConstants
|
1453
|
+
extend DSLKit::Constant
|
1454
|
+
|
1455
|
+
constant :yes, true
|
1456
|
+
|
1457
|
+
constant :no, false
|
1458
|
+
end
|
1459
|
+
|
1460
|
+
# This is the base class of all Benchmarking Cases.
|
1461
|
+
class Case
|
1462
|
+
|
1463
|
+
# All subclasses of Case are extended with this module.
|
1464
|
+
module CaseExtension
|
1465
|
+
def inherited(klass)
|
1466
|
+
Case.cases << klass
|
1467
|
+
end
|
1468
|
+
|
1469
|
+
extend DSLKit::DSLAccessor
|
1470
|
+
extend DSLKit::Constant
|
1471
|
+
|
1472
|
+
include CommonConstants
|
1473
|
+
|
1474
|
+
dsl_accessor :benchmark_name do name end
|
1475
|
+
|
1476
|
+
dsl_accessor :clock, Clock
|
1477
|
+
|
1478
|
+
constant :real
|
1479
|
+
|
1480
|
+
constant :total
|
1481
|
+
|
1482
|
+
constant :user
|
1483
|
+
|
1484
|
+
constant :system
|
1485
|
+
|
1486
|
+
dsl_accessor :compare_time, :real
|
1487
|
+
|
1488
|
+
dsl_accessor :warmup, false
|
1489
|
+
|
1490
|
+
dsl_accessor :batch_size, 1
|
1491
|
+
|
1492
|
+
class TruncateData
|
1493
|
+
extend DSLKit::DSLAccessor
|
1494
|
+
extend DSLKit::Constant
|
1495
|
+
|
1496
|
+
include CommonConstants
|
1497
|
+
include ModuleFunctions
|
1498
|
+
|
1499
|
+
dsl_accessor :alpha_level, 0.05
|
1500
|
+
|
1501
|
+
dsl_accessor :window_size, 10
|
1502
|
+
|
1503
|
+
dsl_accessor :slope_angle, ModuleFunctions.angle(0.1)
|
1504
|
+
|
1505
|
+
dsl_accessor :enabled, false
|
1506
|
+
|
1507
|
+
def initialize(enable, &block)
|
1508
|
+
if block
|
1509
|
+
enable.nil? or raise ArgumentError, "block form doesn't take an argument"
|
1510
|
+
instance_eval(&block)
|
1511
|
+
enabled true
|
1512
|
+
else
|
1513
|
+
enabled enable.nil? ? false : enable
|
1514
|
+
end
|
1515
|
+
end
|
1516
|
+
end
|
1517
|
+
|
1518
|
+
def truncate_data(enable = nil, &block)
|
1519
|
+
@truncate_data ||= TruncateData.new(enable, &block)
|
1520
|
+
end
|
1521
|
+
|
1522
|
+
constant :aggressive, :aggressive
|
1523
|
+
|
1524
|
+
class Covering
|
1525
|
+
extend DSLKit::DSLAccessor
|
1526
|
+
|
1527
|
+
dsl_accessor :alpha_level, 0.05
|
1528
|
+
|
1529
|
+
dsl_accessor :beta_level, 0.05
|
1530
|
+
|
1531
|
+
def initialize(&block)
|
1532
|
+
block and instance_eval(&block)
|
1533
|
+
end
|
1534
|
+
end
|
1535
|
+
|
1536
|
+
def covering(&block)
|
1537
|
+
@covering = Covering.new(&block)
|
1538
|
+
end
|
1539
|
+
|
1540
|
+
dsl_accessor :data_file, false
|
1541
|
+
|
1542
|
+
dsl_accessor :output_dir, Dir.pwd
|
1543
|
+
|
1544
|
+
class Histogram
|
1545
|
+
extend DSLKit::DSLAccessor
|
1546
|
+
extend DSLKit::Constant
|
1547
|
+
|
1548
|
+
include CommonConstants
|
1549
|
+
|
1550
|
+
dsl_accessor :bins, 10
|
1551
|
+
|
1552
|
+
dsl_accessor :enabled, false
|
1553
|
+
|
1554
|
+
dsl_accessor :file, false
|
1555
|
+
|
1556
|
+
def initialize(enable = nil, &block)
|
1557
|
+
if block
|
1558
|
+
enable.nil? or raise ArgumentError, "block form doesn't take an argument"
|
1559
|
+
instance_eval(&block)
|
1560
|
+
enabled true
|
1561
|
+
else
|
1562
|
+
enabled enable.nil? ? true : enable
|
1563
|
+
end
|
1564
|
+
end
|
1565
|
+
end
|
1566
|
+
|
1567
|
+
def histogram(enable = nil, &block)
|
1568
|
+
@histogram ||= Histogram.new(enable, &block)
|
1569
|
+
end
|
1570
|
+
|
1571
|
+
dsl_accessor :detect_outliers, true
|
1572
|
+
|
1573
|
+
dsl_accessor :outliers_factor, 3.0
|
1574
|
+
|
1575
|
+
class Autocorrelation
|
1576
|
+
extend DSLKit::DSLAccessor
|
1577
|
+
extend DSLKit::Constant
|
1578
|
+
|
1579
|
+
include CommonConstants
|
1580
|
+
|
1581
|
+
dsl_accessor :n_limit, 50
|
1582
|
+
|
1583
|
+
dsl_accessor :alpha_level, 0.05
|
1584
|
+
|
1585
|
+
dsl_accessor :max_lags, 20
|
1586
|
+
|
1587
|
+
dsl_accessor :enabled, false
|
1588
|
+
|
1589
|
+
dsl_accessor :file, true
|
1590
|
+
|
1591
|
+
def initialize(enable = nil, &block)
|
1592
|
+
if block
|
1593
|
+
enable.nil? or raise ArgumentError, "block form doesn't take an argument"
|
1594
|
+
instance_eval(&block)
|
1595
|
+
enabled true
|
1596
|
+
else
|
1597
|
+
enabled enable.nil? ? false : enable
|
1598
|
+
end
|
1599
|
+
end
|
1600
|
+
end
|
1601
|
+
|
1602
|
+
def autocorrelation(enable = nil, &block)
|
1603
|
+
@autocorrelation ||= Autocorrelation.new(enable, &block)
|
1604
|
+
end
|
1605
|
+
|
1606
|
+
dsl_accessor :linear_regression, true
|
1607
|
+
|
1608
|
+
include ModuleFunctions
|
1609
|
+
end
|
1610
|
+
|
1611
|
+
class << self
|
1612
|
+
def inherited(klass)
|
1613
|
+
klass.extend CaseExtension
|
1614
|
+
end
|
1615
|
+
|
1616
|
+
extend DSLKit::DSLAccessor
|
1617
|
+
|
1618
|
+
dsl_reader :cases, []
|
1619
|
+
|
1620
|
+
dsl_accessor :output, STDOUT
|
1621
|
+
|
1622
|
+
def output_filename(name)
|
1623
|
+
path = File.expand_path(name, output_dir)
|
1624
|
+
output File.new(path, 'a+')
|
1625
|
+
end
|
1626
|
+
|
1627
|
+
# Returns the total number of run counts +run_count+.
|
1628
|
+
def run_count
|
1629
|
+
cases.inject(0) { |s, c| s + c.run_count }
|
1630
|
+
end
|
1631
|
+
|
1632
|
+
dsl_accessor :autorun, true
|
1633
|
+
|
1634
|
+
# Iterate over all subclasses of class Case.
|
1635
|
+
def each(&block)
|
1636
|
+
cases.each(&block)
|
1637
|
+
end
|
1638
|
+
|
1639
|
+
# Run all subclasses' instances, that is all Bullshit cases, unless they
|
1640
|
+
# already have run.
|
1641
|
+
def run_all
|
1642
|
+
each do |bc_class|
|
1643
|
+
bc_class.run_count == 0 and bc_class.run
|
1644
|
+
end
|
1645
|
+
end
|
1646
|
+
|
1647
|
+
# Autorun all subclasses' instances, that is all Bullshit cases. If its
|
1648
|
+
# autorun dsl_accessor is false or it has already run, don't run the
|
1649
|
+
# case.
|
1650
|
+
def autorun_all
|
1651
|
+
each do |bc_class|
|
1652
|
+
bc_class.autorun and bc_class.run_count == 0 and bc_class.run
|
1653
|
+
end
|
1654
|
+
end
|
1655
|
+
end
|
1656
|
+
|
1657
|
+
# Returns a Case instance, that is used to run benchmark methods and
|
1658
|
+
# measure their running time.
|
1659
|
+
def initialize
|
1660
|
+
@clocks = []
|
1661
|
+
@comparison = Comparison.new
|
1662
|
+
@comparison.output self.class.output
|
1663
|
+
end
|
1664
|
+
|
1665
|
+
# Return the name of the benchmark case as a string.
|
1666
|
+
def to_s
|
1667
|
+
self.class.benchmark_name
|
1668
|
+
end
|
1669
|
+
|
1670
|
+
# Return all benchmark methods of this Case instance lexicographically
|
1671
|
+
# sorted.
|
1672
|
+
def self.sorted_bmethods
|
1673
|
+
instance_methods.map { |x| x.to_s }.grep(/\Abenchmark_/).sort
|
1674
|
+
end
|
1675
|
+
|
1676
|
+
# Return all benchmark methods of this Case instance in a random order.
|
1677
|
+
def bmethods
|
1678
|
+
unless @bmethods
|
1679
|
+
@bmethods = self.class.sorted_bmethods.sort_by do
|
1680
|
+
rand
|
1681
|
+
end
|
1682
|
+
@bmethods.map! { |n| CaseMethod.new(n, self) }
|
1683
|
+
end
|
1684
|
+
@bmethods
|
1685
|
+
end
|
1686
|
+
|
1687
|
+
# Return the CaseMethod instance for +method_name+ or nil, if there isn't
|
1688
|
+
# any method of this name.
|
1689
|
+
def [](method_name)
|
1690
|
+
method_name = "benchmark_#{method_name}"
|
1691
|
+
bmethods.find { |bm| bm.name == method_name }
|
1692
|
+
end
|
1693
|
+
|
1694
|
+
# Return the length of the longest_name of all these methods' names.
|
1695
|
+
def longest_name
|
1696
|
+
bmethods.empty? and return 0
|
1697
|
+
bmethods.map { |x| x.short_name.size }.max
|
1698
|
+
end
|
1699
|
+
|
1700
|
+
# Run benchmark case once and output results.
|
1701
|
+
def run_once
|
1702
|
+
self.class.run_count(self.class.run_count + 1)
|
1703
|
+
self.class.output.puts Time.now.strftime(' %FT%T %Z ').center(COLUMNS, '=')
|
1704
|
+
self.class.output.puts "Benchmarking on #{RUBY_DESCRIPTION}."
|
1705
|
+
self.class.output.puts self.class.message
|
1706
|
+
self.class.output.puts '=' * COLUMNS, ''
|
1707
|
+
@clocks.clear
|
1708
|
+
if self.class.warmup == :aggressive
|
1709
|
+
self.class.output.puts "Aggressively run all benchmarks for warmup first.", ''
|
1710
|
+
bmethods.each do |bc_method|
|
1711
|
+
GC.start
|
1712
|
+
clock = run_method bc_method
|
1713
|
+
self.class.output.puts evaluation(clock)
|
1714
|
+
GC.start
|
1715
|
+
end
|
1716
|
+
self.class.output.puts "Aggressive warmup done.", '', '=' * COLUMNS, ''
|
1717
|
+
end
|
1718
|
+
first = true
|
1719
|
+
bmethods.each do |bc_method|
|
1720
|
+
if first
|
1721
|
+
first = false
|
1722
|
+
else
|
1723
|
+
self.class.output.puts '-' * COLUMNS, ''
|
1724
|
+
end
|
1725
|
+
if self.class.warmup
|
1726
|
+
self.class.output.puts "This first run is only for warmup."
|
1727
|
+
GC.start
|
1728
|
+
clock = run_method bc_method
|
1729
|
+
self.class.output.puts evaluation(clock)
|
1730
|
+
GC.start
|
1731
|
+
end
|
1732
|
+
clock = run_method(bc_method)
|
1733
|
+
if self.class.truncate_data.enabled
|
1734
|
+
message = ''
|
1735
|
+
offset = clock.find_truncation_offset
|
1736
|
+
if clock.case.data_file
|
1737
|
+
slopes_file_path = clock.file_path 'slopes'
|
1738
|
+
message << "Writing slopes data file '#{slopes_file_path}'.\n"
|
1739
|
+
File.open(slopes_file_path, 'w') do |slopes_file|
|
1740
|
+
slopes_file.puts %w[#scatter slope] * "\t"
|
1741
|
+
slopes_file.puts clock.slopes.map { |s| s * "\t" }
|
1742
|
+
end
|
1743
|
+
end
|
1744
|
+
case offset
|
1745
|
+
when 0
|
1746
|
+
message << "No initial data truncated.\n =>"\
|
1747
|
+
" System may have been in a steady state from the beginning."
|
1748
|
+
when clock.repeat
|
1749
|
+
message << "After truncating measurements no data would have"\
|
1750
|
+
" remained.\n => No steady state could be detected."
|
1751
|
+
else
|
1752
|
+
if clock.case.data_file
|
1753
|
+
data_file_path = clock.file_path 'untruncated'
|
1754
|
+
message << "Writing untruncated measurement data file '#{data_file_path}'.\n"
|
1755
|
+
File.open(data_file_path, 'w') do |data_file|
|
1756
|
+
data_file.puts clock.class.to_a * "\t"
|
1757
|
+
data_file.puts clock.to_a.map { |times| times * "\t" }
|
1758
|
+
end
|
1759
|
+
end
|
1760
|
+
remaining = clock.repeat - offset
|
1761
|
+
offset_percentage = 100 * offset.to_f / clock.repeat
|
1762
|
+
message << sprintf("Truncated initial %u measurements: "\
|
1763
|
+
"%u -> %u (-%0.2f%%).\n", offset, clock.repeat, remaining,
|
1764
|
+
offset_percentage)
|
1765
|
+
clock.truncate_data(offset)
|
1766
|
+
end
|
1767
|
+
self.class.output.puts evaluation(clock), message
|
1768
|
+
else
|
1769
|
+
self.class.output.puts evaluation(clock)
|
1770
|
+
end
|
1771
|
+
@clocks << clock
|
1772
|
+
@comparison.benchmark(self, bc_method.short_name, :run => false)
|
1773
|
+
end
|
1774
|
+
@clocks
|
1775
|
+
end
|
1776
|
+
|
1777
|
+
# The clock instances, that were used during a run of this benchmark case.
|
1778
|
+
attr_reader :clocks
|
1779
|
+
|
1780
|
+
# Setup, run all benchmark cases (warmup and the real run) and output
|
1781
|
+
# results, run method speed comparisons, and teardown.
|
1782
|
+
def run(comparison = true)
|
1783
|
+
old_sync, self.class.output.sync = self.class.output.sync, true
|
1784
|
+
$DEBUG and warn "Calling setup."
|
1785
|
+
setup
|
1786
|
+
run_once
|
1787
|
+
comparison and @comparison.display
|
1788
|
+
self
|
1789
|
+
rescue => e
|
1790
|
+
warn "Caught #{e.class}: #{e}\n\n#{e.backtrace.map { |x| "\t#{x}\n" }}"
|
1791
|
+
ensure
|
1792
|
+
$DEBUG and warn "Calling teardown."
|
1793
|
+
teardown
|
1794
|
+
@clocks and write_files
|
1795
|
+
self.class.output.sync = old_sync
|
1796
|
+
end
|
1797
|
+
|
1798
|
+
# Creates an instance of this class and run it.
|
1799
|
+
def self.run
|
1800
|
+
new.run
|
1801
|
+
end
|
1802
|
+
|
1803
|
+
# Write all output files after run.
|
1804
|
+
def write_files
|
1805
|
+
for clock in @clocks
|
1806
|
+
if clock.case.data_file data_file_path = clock.file_path
|
1807
|
+
self.class.output.puts "Writing measurement data file '#{data_file_path}'."
|
1808
|
+
File.open(data_file_path, 'w') do |data_file|
|
1809
|
+
data_file.puts clock.class.to_a * "\t"
|
1810
|
+
data_file.puts clock.to_a.map { |times| times * "\t" }
|
1811
|
+
end
|
1812
|
+
end
|
1813
|
+
if clock.case.histogram.enabled and clock.case.histogram.file
|
1814
|
+
histogram_file_path = clock.file_path 'histogram'
|
1815
|
+
self.class.output.puts "Writing histogram file '#{histogram_file_path}'."
|
1816
|
+
File.open(histogram_file_path, 'w') do |data_file|
|
1817
|
+
data_file.puts %w[#binleft frequency binright] * "\t"
|
1818
|
+
data_file.puts clock.histogram(clock.case.compare_time).to_a.map { |times| times * "\t" }
|
1819
|
+
end
|
1820
|
+
end
|
1821
|
+
if clock.case.autocorrelation.enabled and clock.case.autocorrelation.file
|
1822
|
+
ac_plot_file_path = clock.file_path 'autocorrelation'
|
1823
|
+
self.class.output.puts "Writing autocorrelation plot file '#{ac_plot_file_path}'."
|
1824
|
+
File.open(ac_plot_file_path, 'w') do |data_file|
|
1825
|
+
data_file.puts %w[#lag autocorrelation] * "\t"
|
1826
|
+
data_file.puts clock.autocorrelation_plot(clock.case.compare_time).to_a.map { |ac| ac * "\t" }
|
1827
|
+
end
|
1828
|
+
end
|
1829
|
+
end
|
1830
|
+
end
|
1831
|
+
|
1832
|
+
# Output before +bc_method+ is run.
|
1833
|
+
def pre_run(bc_method)
|
1834
|
+
setup_name = bc_method.setup_name
|
1835
|
+
if respond_to? setup_name
|
1836
|
+
$DEBUG and warn "Calling #{setup_name}."
|
1837
|
+
__send__(setup_name)
|
1838
|
+
end
|
1839
|
+
self.class.output.puts "#{bc_method.long_name}:"
|
1840
|
+
end
|
1841
|
+
|
1842
|
+
# Run only pre_run and post_run methods. Yield to the block, if one was
|
1843
|
+
# given.
|
1844
|
+
def run_method(bc_method)
|
1845
|
+
pre_run bc_method
|
1846
|
+
clock = self.class.clock.__send__(self.class.clock_method, bc_method) do
|
1847
|
+
__send__(bc_method.name)
|
1848
|
+
end
|
1849
|
+
bc_method.clock = clock
|
1850
|
+
post_run bc_method
|
1851
|
+
clock
|
1852
|
+
end
|
1853
|
+
|
1854
|
+
# This method has to be implemented in subclasses, it should return the
|
1855
|
+
# evaluation results of the benchmarks as a string.
|
1856
|
+
def evaluation(clock)
|
1857
|
+
raise NotImplementedError, "has to be implemented in subclasses"
|
1858
|
+
end
|
1859
|
+
|
1860
|
+
# Output after +bc_method+ is run.
|
1861
|
+
def post_run(bc_method)
|
1862
|
+
teardown_name = bc_method.teardown_name
|
1863
|
+
if respond_to? teardown_name
|
1864
|
+
$DEBUG and warn "Calling #{teardown_name}."
|
1865
|
+
__send__(bc_method.teardown_name)
|
1866
|
+
end
|
1867
|
+
end
|
1868
|
+
|
1869
|
+
# General setup for all the benchmark methods.
|
1870
|
+
def setup
|
1871
|
+
end
|
1872
|
+
|
1873
|
+
# General teardown for all the benchmark methods.
|
1874
|
+
def teardown
|
1875
|
+
end
|
1876
|
+
end
|
1877
|
+
|
1878
|
+
# This module contains methods, that can be used in the evaluation method of
|
1879
|
+
# benchmark cases.
|
1880
|
+
module EvaluationModule
|
1881
|
+
def statistics_table(clock)
|
1882
|
+
result = ' ' * NAME_COLUMN_SIZE << ('%17s ' * 4) % times << "\n"
|
1883
|
+
result << evaluation_line('sum', times.map { |t| clock.__send__(t) })
|
1884
|
+
result << evaluation_line('min', times.map { |t| clock.min(t) })
|
1885
|
+
result << evaluation_line('std-', times.map { |t| clock.arithmetic_mean(t) - clock.sample_standard_deviation(t) })
|
1886
|
+
result << evaluation_line('mean', times.map { |t| clock.arithmetic_mean(t) })
|
1887
|
+
result << evaluation_line('std+', times.map { |t| clock.arithmetic_mean(t) + clock.sample_standard_deviation(t) })
|
1888
|
+
result << evaluation_line('max', times.map { |t| clock.max(t) })
|
1889
|
+
result << evaluation_line('std', times.map { |t| clock.sample_standard_deviation(t) })
|
1890
|
+
result << evaluation_line('std%', times.map { |t| clock.sample_standard_deviation_percentage(t) })
|
1891
|
+
result << evaluation_line('harm', times.map { |t| clock.harmonic_mean(t) })
|
1892
|
+
result << evaluation_line('geo', times.map { |t| clock.geometric_mean(t) })
|
1893
|
+
result << evaluation_line('q1', times.map { |t| clock.percentile(t, 25) })
|
1894
|
+
result << evaluation_line('med', times.map { |t| clock.median(t) })
|
1895
|
+
result << evaluation_line('q3', times.map { |t| clock.percentile(t, 75) })
|
1896
|
+
result << ' ' * NAME_COLUMN_SIZE << "%17u %17.5f %17.9f\n" % [ clock.repeat, clock.calls_mean, clock.call_time_mean ]
|
1897
|
+
result << ' ' * NAME_COLUMN_SIZE << "%17s %17s %17s\n" % %w[calls calls/sec secs/call]
|
1898
|
+
end
|
1899
|
+
|
1900
|
+
def histogram(clock)
|
1901
|
+
result = "\n"
|
1902
|
+
if clock.case.histogram.enabled
|
1903
|
+
clock.histogram(clock.case.compare_time).display(result, 50)
|
1904
|
+
end
|
1905
|
+
result
|
1906
|
+
end
|
1907
|
+
|
1908
|
+
def detect_outliers(clock)
|
1909
|
+
result = ''
|
1910
|
+
if clock.case.detect_outliers and
|
1911
|
+
outliers = clock.detect_outliers(clock.case.compare_time)
|
1912
|
+
then
|
1913
|
+
result << "\nOutliers detected with box plot algo "\
|
1914
|
+
"(median=%.5f, iqr=%.5f, factor=%.2f):\n" % outliers.values_at(:median, :iqr, :factor)
|
1915
|
+
result << outliers.select { |n, |
|
1916
|
+
[ :very_low, :low, :high, :very_high ].include?(n)
|
1917
|
+
}.map { |n, v| "#{n}=#{v}" } * ' '
|
1918
|
+
result << "\n"
|
1919
|
+
else
|
1920
|
+
result << "\nNo outliers detected with box plot algo.\n"
|
1921
|
+
end
|
1922
|
+
result
|
1923
|
+
end
|
1924
|
+
|
1925
|
+
def detect_autocorrelation(clock)
|
1926
|
+
result = ''
|
1927
|
+
clock.case.autocorrelation.enabled or return result
|
1928
|
+
if r = clock.detect_autocorrelation(clock.case.compare_time)
|
1929
|
+
result << "\nLjung-Box statistics: q=%.5f (alpha=#{r[:alpha_level]},"\
|
1930
|
+
" df=#{r[:lags]}).\n" % r[:q]
|
1931
|
+
if r[:detected]
|
1932
|
+
result << "%.5f >= %.5f => Autocorrelation was detected.\n" %
|
1933
|
+
[ r[:p], 1 - r[:alpha_level] ]
|
1934
|
+
else
|
1935
|
+
result << "%.5f < %.5f => No autocorrelation was detected.\n" %
|
1936
|
+
[ r[:p], 1 - r[:alpha_level] ]
|
1937
|
+
end
|
1938
|
+
else
|
1939
|
+
result << "\nDidn't have enough lags to compute Ljung-Box statistics.\n"
|
1940
|
+
end
|
1941
|
+
result
|
1942
|
+
end
|
1943
|
+
|
1944
|
+
private
|
1945
|
+
|
1946
|
+
def evaluation_line(name, values)
|
1947
|
+
name.ljust(NAME_COLUMN_SIZE) << ('%17.9f ' * 4) % values << "\n"
|
1948
|
+
end
|
1949
|
+
|
1950
|
+
def times
|
1951
|
+
self.class.clock.times
|
1952
|
+
end
|
1953
|
+
end
|
1954
|
+
|
1955
|
+
# This is a Benchmarking Case that uses a time limit.
|
1956
|
+
class TimeCase < Case
|
1957
|
+
include EvaluationModule
|
1958
|
+
|
1959
|
+
class << self
|
1960
|
+
extend DSLKit::DSLAccessor
|
1961
|
+
extend DSLKit::Constant
|
1962
|
+
|
1963
|
+
constant :clock_method, :time
|
1964
|
+
|
1965
|
+
dsl_accessor :duration
|
1966
|
+
|
1967
|
+
dsl_accessor :run_count, 0
|
1968
|
+
|
1969
|
+
def message
|
1970
|
+
"Running '#{self}' for a duration of #{duration} secs/method"\
|
1971
|
+
" (compare_time=#{compare_time}):"
|
1972
|
+
end
|
1973
|
+
|
1974
|
+
dsl_accessor(:output) { ::Bullshit::Case.output }
|
1975
|
+
end
|
1976
|
+
|
1977
|
+
# Returns the evaluation for +bullshit_case+ with the results of the
|
1978
|
+
# benchmarking as a String.
|
1979
|
+
def evaluation(clock)
|
1980
|
+
clock.repeat == 0 and
|
1981
|
+
raise BullshitException, "no measurements were gauged"
|
1982
|
+
result = ''
|
1983
|
+
result << statistics_table(clock)
|
1984
|
+
result << histogram(clock)
|
1985
|
+
result << detect_outliers(clock)
|
1986
|
+
result << detect_autocorrelation(clock)
|
1987
|
+
result << "\n"
|
1988
|
+
end
|
1989
|
+
|
1990
|
+
# Check if duration has been set. If yes call Case#run, otherwise raise a
|
1991
|
+
# BullshitException exception.
|
1992
|
+
def run(*)
|
1993
|
+
self.class.duration or
|
1994
|
+
raise BullshitException, 'duration has to be set'
|
1995
|
+
super
|
1996
|
+
end
|
1997
|
+
end
|
1998
|
+
|
1999
|
+
# This is a Benchmarking Case that uses a repetition limit.
|
2000
|
+
class RepeatCase < Case
|
2001
|
+
include EvaluationModule
|
2002
|
+
|
2003
|
+
class << self
|
2004
|
+
extend DSLKit::DSLAccessor
|
2005
|
+
extend DSLKit::Constant
|
2006
|
+
|
2007
|
+
constant :clock_method, :repeat
|
2008
|
+
|
2009
|
+
dsl_accessor :iterations
|
2010
|
+
|
2011
|
+
dsl_accessor :run_count, 0
|
2012
|
+
|
2013
|
+
def message
|
2014
|
+
"Running '#{self}' for #{iterations} iterations/method"\
|
2015
|
+
" (compare_time=#{compare_time})"
|
2016
|
+
end
|
2017
|
+
|
2018
|
+
dsl_accessor(:output) { ::Bullshit::Case.output }
|
2019
|
+
end
|
2020
|
+
|
2021
|
+
# Returns the evaluation for +bullshit_case+ with the results of the
|
2022
|
+
# benchmarking as a String.
|
2023
|
+
def evaluation(clock)
|
2024
|
+
clock.repeat == 0 and
|
2025
|
+
raise BullshitException, "no measurements were gauged"
|
2026
|
+
result = ''
|
2027
|
+
result << statistics_table(clock)
|
2028
|
+
result << histogram(clock)
|
2029
|
+
result << detect_outliers(clock)
|
2030
|
+
result << detect_autocorrelation(clock)
|
2031
|
+
result << "\n"
|
2032
|
+
end
|
2033
|
+
|
2034
|
+
# vn heck if iterations has been set. If yes call Case#run, otherwise raise a
|
2035
|
+
# BullshitException exception.
|
2036
|
+
def run(*)
|
2037
|
+
self.class.iterations or
|
2038
|
+
raise BullshitException, 'iterations have to be set'
|
2039
|
+
super
|
2040
|
+
end
|
2041
|
+
end
|
2042
|
+
|
2043
|
+
# A range case is a benchmark case, where each iteration depends on a
|
2044
|
+
# different argument, which might cause a non-constant running time. The
|
2045
|
+
# evaluation of the case doesn't check for constancy and steady-state
|
2046
|
+
# properties.
|
2047
|
+
class RangeCase < Case
|
2048
|
+
include EvaluationModule
|
2049
|
+
|
2050
|
+
class << self
|
2051
|
+
extend DSLKit::DSLAccessor
|
2052
|
+
extend DSLKit::Constant
|
2053
|
+
|
2054
|
+
constant :clock_method, :scale_range
|
2055
|
+
|
2056
|
+
dsl_accessor :range
|
2057
|
+
|
2058
|
+
dsl_accessor :scatter, 1
|
2059
|
+
|
2060
|
+
dsl_accessor :run_count, 0
|
2061
|
+
|
2062
|
+
def message
|
2063
|
+
"Running '#{self}' for range #{range.inspect}"\
|
2064
|
+
" (compare_time=#{compare_time})"
|
2065
|
+
end
|
2066
|
+
|
2067
|
+
dsl_accessor(:output) { ::Bullshit::Case.output }
|
2068
|
+
|
2069
|
+
attr_accessor :args
|
2070
|
+
end
|
2071
|
+
|
2072
|
+
def args
|
2073
|
+
self.class.args
|
2074
|
+
end
|
2075
|
+
private :args
|
2076
|
+
|
2077
|
+
# Returns the evaluation for +bullshit_case+ with the results of the
|
2078
|
+
# benchmarking as a String.
|
2079
|
+
def evaluation(clock)
|
2080
|
+
clock.repeat == 0 and
|
2081
|
+
raise BullshitException, "no measurements were gauged"
|
2082
|
+
result = ''
|
2083
|
+
result << statistics_table(clock)
|
2084
|
+
result << "\n"
|
2085
|
+
end
|
2086
|
+
|
2087
|
+
# Check if iterations has been set. If yes call Case#run, otherwise raise a
|
2088
|
+
# BullshitException exception.
|
2089
|
+
def run(*)
|
2090
|
+
self.class.range or
|
2091
|
+
raise BullshitException,
|
2092
|
+
'range has to be set to an enumerable of arguments'
|
2093
|
+
super
|
2094
|
+
end
|
2095
|
+
end
|
2096
|
+
|
2097
|
+
# A Comparison instance compares the benchmark results of different
|
2098
|
+
# case methods and outputs the results.
|
2099
|
+
class Comparison
|
2100
|
+
extend DSLKit::Constant
|
2101
|
+
extend DSLKit::DSLAccessor
|
2102
|
+
|
2103
|
+
include CommonConstants
|
2104
|
+
|
2105
|
+
dsl_accessor :output, STDOUT
|
2106
|
+
|
2107
|
+
dsl_accessor :output_dir, Dir.pwd
|
2108
|
+
|
2109
|
+
# Output results to the file named +name+.
|
2110
|
+
def output_filename(name)
|
2111
|
+
path = File.expand_path(name, output_dir)
|
2112
|
+
output File.new(path, 'a+')
|
2113
|
+
end
|
2114
|
+
|
2115
|
+
# Return a comparison object configured by +block+.
|
2116
|
+
def initialize(&block)
|
2117
|
+
@cases = {}
|
2118
|
+
@benchmark_methods = []
|
2119
|
+
block and instance_eval(&block)
|
2120
|
+
end
|
2121
|
+
|
2122
|
+
# Benchmark case method +method+ of +bc_class+. Options are:
|
2123
|
+
# * :run to not run this +bc_class+ if set to false (defaults to true),
|
2124
|
+
# * :load to load the data of a previous run for this +method+, if set to
|
2125
|
+
# true. If the true value is a file path string, load the data from the
|
2126
|
+
# given file at the path.
|
2127
|
+
def benchmark(bc_class, method, opts = {})
|
2128
|
+
opts = { :run => true, :combine => true }.merge opts
|
2129
|
+
if Case === bc_class
|
2130
|
+
bullshit_case, bc_class = bc_class, bullshit_case.class
|
2131
|
+
@cases[bc_class] ||= []
|
2132
|
+
if opts[:combine]
|
2133
|
+
if @cases[bc_class].empty?
|
2134
|
+
@cases[bc_class] << bullshit_case
|
2135
|
+
else
|
2136
|
+
bullshit_case = @cases[bc_class].first
|
2137
|
+
end
|
2138
|
+
else
|
2139
|
+
@cases[bc_class] << bullshit_case
|
2140
|
+
end
|
2141
|
+
else
|
2142
|
+
@cases[bc_class] ||= []
|
2143
|
+
if opts[:combine]
|
2144
|
+
unless bullshit_case = @cases[bc_class].first
|
2145
|
+
bullshit_case = bc_class.new
|
2146
|
+
@cases[bc_class] << bullshit_case
|
2147
|
+
end
|
2148
|
+
else
|
2149
|
+
bullshit_case = bc_class.new
|
2150
|
+
@cases[bc_class] << bullshit_case
|
2151
|
+
end
|
2152
|
+
end
|
2153
|
+
bc_method = bullshit_case[method] or raise BullshitException,
|
2154
|
+
"unknown benchmark method #{bc_class}##{method}"
|
2155
|
+
if comment = opts[:comment]
|
2156
|
+
bc_method.comment = comment
|
2157
|
+
end
|
2158
|
+
@benchmark_methods << bc_method
|
2159
|
+
if file_path = opts[:load]
|
2160
|
+
if file_path != true
|
2161
|
+
bc_method.load file_path
|
2162
|
+
else
|
2163
|
+
bc_method.load
|
2164
|
+
end
|
2165
|
+
else
|
2166
|
+
opts[:run] and bullshit_case.run false
|
2167
|
+
end
|
2168
|
+
nil
|
2169
|
+
end
|
2170
|
+
|
2171
|
+
# Return all benchmark methods for all the cached benchmark cases.
|
2172
|
+
attr_reader :benchmark_methods
|
2173
|
+
|
2174
|
+
# Return all benchmark methods ordered by the result of comparator call to
|
2175
|
+
# their clock values.
|
2176
|
+
def compare_methods(comparator)
|
2177
|
+
benchmark_methods.sort_by { |m| m.clock.__send__(comparator) }
|
2178
|
+
end
|
2179
|
+
|
2180
|
+
# Return the length of the longest name of all benchmark methods.
|
2181
|
+
def longest_name_size
|
2182
|
+
benchmark_methods.map { |m| m.long_name.size }.max
|
2183
|
+
end
|
2184
|
+
|
2185
|
+
# Returns the prefix_string for a method speed comparison in the output.
|
2186
|
+
def prefix_string(method)
|
2187
|
+
"% -#{longest_name_size}s %u repeats:" %
|
2188
|
+
[ method.long_name , method.clock.repeat ]
|
2189
|
+
end
|
2190
|
+
|
2191
|
+
# Output all speed comparisons between methods.
|
2192
|
+
def display
|
2193
|
+
output.puts Time.now.strftime(' %FT%T %Z ').center(COLUMNS, '=')
|
2194
|
+
for comparator in [ :call_time_mean, :call_time_median ]
|
2195
|
+
output.puts
|
2196
|
+
methods = compare_methods(comparator)
|
2197
|
+
methods.size < 2 and return
|
2198
|
+
max = methods.last.clock.__send__(comparator)
|
2199
|
+
output.puts "Comparing times (#{comparator}):"
|
2200
|
+
methods.each_with_index do |m, i|
|
2201
|
+
covers = []
|
2202
|
+
for x in methods
|
2203
|
+
if m != x and m.cover?(x)
|
2204
|
+
j = 0
|
2205
|
+
if methods.find { |y| j += 1; x == y }
|
2206
|
+
my_case = m.case.class
|
2207
|
+
iterations = m.clock.analysis[my_case.compare_time].suggested_sample_size(
|
2208
|
+
x.clock.analysis[my_case.compare_time], my_case.covering.alpha_level, my_case.covering.beta_level)
|
2209
|
+
if iterations.nan? or iterations.infinite?
|
2210
|
+
covers << "#{j} (?)"
|
2211
|
+
else
|
2212
|
+
min_iterations = iterations.ceil
|
2213
|
+
min_iterations >= 1E6 and min_iterations = "%f" % (1 / 0.0)
|
2214
|
+
covers << "#{j} (>=#{min_iterations})"
|
2215
|
+
end
|
2216
|
+
end
|
2217
|
+
end
|
2218
|
+
end
|
2219
|
+
covers *= ', '
|
2220
|
+
output.printf\
|
2221
|
+
"% 2u #{prefix_string(m)}\n %17.9f"\
|
2222
|
+
" (%#{::Bullshit::Clock::TIMES_MAX}s) -> %8.3fx %s\n"\
|
2223
|
+
" %17.9f\n",
|
2224
|
+
i + 1, m.clock.calls(comparator), m.case.class.compare_time,
|
2225
|
+
max / m.clock.__send__(comparator), covers, m.clock.__send__(comparator)
|
2226
|
+
end
|
2227
|
+
output.puts " %17s (%#{::Bullshit::Clock::TIMES_MAX}s) -> %8s %s\n"\
|
2228
|
+
" %17s\n"\
|
2229
|
+
% %w[calls/sec time speed covers secs/call]
|
2230
|
+
end
|
2231
|
+
output.puts '=' * COLUMNS
|
2232
|
+
end
|
2233
|
+
end
|
2234
|
+
|
2235
|
+
# Create a Comparison instance configured by +block+ and compute and display
|
2236
|
+
# the results.
|
2237
|
+
def self.compare(&block)
|
2238
|
+
Comparison.new(&block).display
|
2239
|
+
end
|
2240
|
+
|
2241
|
+
at_exit do
|
2242
|
+
Case.autorun and Case.run_all
|
2243
|
+
end
|
2244
|
+
end
|