-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbenchmark.R
247 lines (210 loc) · 10.1 KB
/
benchmark.R
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
require(digest)
require(stringi)
require(data.table)
################################################################################################
#
# 01. Loading of benchmark data sets
#
################################################################################################
# 01b. Get text from randomly selected tweets
################################################################################################
tweets <- readLines('data/tweets.txt', encoding = 'UTF-8')
# verify checksum of loaded lines
digest(paste0(tweets, collapse = '||'),
algo='sha256',
serialize=F)==
"7fa3bf921c393fe7009bc60971b2bb8396414e7602bb4f409bed78c7192c30f4"
# 01c. Get text from randomly selected blog descriptions
################################################################################################
# make sure we can read it back in
blogs <- readLines('data/blogs.txt', encoding = 'UTF-8')
# verify checksum of loaded lines
digest(paste0(blogs, collapse = '||'),
algo='sha256',
serialize=F)==
"14b3c593e543eb8b2932cf00b646ed653e336897a03c82098b725e6e1f9b7aa2"
################################################################################################
#
# 02. Define the functions used for benchmarking
#
################################################################################################
# 02a. Pre-processing functions
################################################################################################
# split.sentence
# Returns a matrix containing in column i the part of the line before the ith word (sentence)
# and the ith word (nextWord).
# The function is used in benchmark to generate and evaluate predictions for the partial lines.
split.sentence <- compiler::cmpfun(function(line) {
require(stringi)
# append a space to the sentence (to make sure we always create one result with only the
# last word missing)
sent <- paste0(line, ' ')
sep <- stri_locate_all_regex(line,
pattern = '[^\\w\'@#\u2018\u2019\u201b]+',
omit_empty=T,
case_insensitive=T)[[1]]
sapply(seq_len(nrow(sep)),
function(i) {
c(sentence=ifelse(i>1, substr(line, 1, sep[i-1,2]), ''),
nextWord=tolower(substr(line, max(sep[i-1,2]+1, 1), min(nchar(line), sep[i,1]-1)))
)
})
}, options=list(optimize=3))
# 02b. Benchmarking function
################################################################################################
# benchmark
# Evaluates the performance of a next word prediction algorithm based on the provided test data-
# set(s).
#
# Parameters
# FUN Function that produces the next word prediction. The function should take a single
# character value as first input and return a vector of character values represen-
# ting the top-3 predictions (with the 1st value being the first prediction).
# ... Additional parameters to pass to FUN.
# sent.list Named list of character vectors containing the text lines used for the benchmark.
# ext.output If TRUE, return additional details about the R environment and loaded packages
# after completing the benchmark.
benchmark <- compiler::cmpfun(function(FUN, ..., sent.list, ext.output=T) {
require(stringi)
require(digest)
require(data.table)
result <- rbindlist(lapply(names(sent.list),
function(list.name) {
sentences <- sent.list[[list.name]]
score <- 0
max.score <-0
hit.count.top3 <- 0
hit.count.top1 <- 0
total.count <- 0
time <- system.time({
for (sent in sentences) {
split <- split.sentence(sent[1])
max.score <- max.score + ncol(split)*3
total.count <- total.count + ncol(split)
rank <- mcmapply(function(i) {
min(which(FUN(split[1,i], ...)==split[2,i]),4)
}, seq_len(ncol(split)), SIMPLIFY = TRUE, mc.cores = coreCount)
# rank <-
# sapply(seq_len(ncol(split)),
# function(i) {
# min(which(FUN(split[1,i], ...) == split[2,i]),4)
# })
print(rank)
score <- score + sum(4-rank)
hit.count.top3 <- hit.count.top3 + sum(rank<4)
hit.count.top1 <- hit.count.top1 + sum(rank==1)
}
})
list('list.name' = list.name,
'line.count' = length(sentences),
'word.count' = sum(stri_count_words(sentences)),
'hash' = digest(paste0(sentences, collapse = '||'), algo='sha256', serialize=F),
'score' = score,
'max.score' = max.score,
'hit.count.top3' = hit.count.top3,
'hit.count.top1' = hit.count.top1,
'total.count' = total.count,
'total.runtime' = time[3]
)
}), use.names=T)
setkey(result, list.name)
# The overall scores are calculated weighting each data set equally (independent of the
# number of lines in each dataset).
overall.score.percent = 100 * result[,sum(score/max.score)/.N]
overall.precision.top3 = 100 * result[,sum(hit.count.top3/total.count)/.N]
overall.precision.top1 = 100 * result[,sum(hit.count.top1/total.count)/.N]
average.runtime = 1000 * result[,sum(total.runtime)/sum(total.count)]
number.of.predictions = result[,sum(total.count)]
total.mem.used = sum(unlist(lapply(ls(.GlobalEnv),
function(x) {
object.size(get(x,
envir = .GlobalEnv,
inherits = FALSE))
})))/(1024^2)
cat(sprintf(paste0('Overall top-3 score: %.2f %%\n',
'Overall top-1 precision: %.2f %%\n',
'Overall top-3 precision: %.2f %%\n',
'Average runtime: %.2f msec\n',
'Number of predictions: %d\n',
'Total memory used: %.2f MB\n'),
overall.score.percent,
overall.precision.top1,
overall.precision.top3,
average.runtime,
number.of.predictions,
total.mem.used
))
cat('\nDataset details\n')
for (p.list.name in result$list.name) {
res <- result[list(p.list.name)]
cat(sprintf(paste0(' Dataset "%s" (%d lines, %d words, hash %s)\n',
' Score: %.2f %%, Top-1 precision: %.2f %%, Top-3 precision: %.2f %%\n'
),
p.list.name,
res$line.count,
res$word.count,
res$hash,
100 * res$score/res$max.score,
100 * res$hit.count.top1/res$total.count,
100 * res$hit.count.top3/res$total.count
))
}
if (ext.output==T) {
packages <- sort(stri_replace_first_fixed(search()[stri_detect_regex(search(),
'^package:')],
'package:', ''))
cat(sprintf(paste0('\n\n%s, platform %s\n',
'Attached non-base packages: %s\n',
'Unattached non-base packages: %s'
),
sessionInfo()$R.version$version.string,
sessionInfo()$platform,
paste0(sapply(sessionInfo()$otherPkgs,
function(pkg) {
paste0(pkg$Package, ' (v', pkg$Version, ')')
}),
collapse = ', '),
paste0(sapply(sessionInfo()$loadedOnly,
function(pkg) {
paste0(pkg$Package, ' (v', pkg$Version, ')')
}),
collapse = ', ')
))
}
}, options=list(optimize =3))
################################################################################################
#
# 03. Define the wrapper function to be called by benchmark
#
################################################################################################
# As an example, we create a very simple baseline algorithm which always returns
# the three most frequent English words.
#predict.baseline <- function(x){
# stop(length(x))
# c('the', 'on', 'a')
#}
source("common.R")
tweetsModelDb <- dbInit(modelName('tweets'))
#blogsModelDb <- dbInit(modelName('blogs'))
tweetsModelEnv <- as(tweetsModelDb, "list")
predict.baseline.pre <- function(q) {
#tweetsModel <- tweetsModelDb$model
#blogsModel <- blogsModelDb$model
results <- predict.baseline.raw(q, list(tweetsModelEnv$model)) #blogsModel,
}
if (!exists('predict.baseline'))
predict.baseline <- memoise(predict.baseline.pre)
#if (!exists("memoQueryNgramProbs"))
# memoQueryNgramProbs <- memoise(queryNgramProbs)
#dbTweets <- db$tweets
#cl <- makeCluster.default()
################################################################################################
#
# 04. Perform the benchmark
#
################################################################################################
benchmark(predict.baseline,
# additional parameters to be passed to the prediction function can be inserted here
sent.list = list('tweets' = tweets,
'blogs' = blogs),
ext.output = T)