uaparserjs
July 9, 2026 · View on GitHub
uaparserjs
Parses HTTP user agent strings and returns user agent, device and OS information. This is a ‘V8’-backed package based on the ‘ua-parser’ project https://github.com/ua-parser.
Functions
The following functions are implemented:
ua_parse: Parse a vector of user agents into a data frame
Installation
install.packages("uaparserjs")
Example Usage
library(uaparserjs)
# current version
packageVersion("uaparserjs")
## [1] '0.4.0'
dplyr::glimpse(ua_parse("Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/535.2 (KHTML, like Gecko) Ubuntu/11.10 Chromium/15.0.874.106 Chrome/15.0.874.106 Safari/535.2"))
## Rows: 1
## Columns: 9
## $ userAgent <chr> "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/535.2 (KHTM…
## $ ua.family <chr> "Chromium"
## $ ua.major <chr> "15"
## $ ua.minor <chr> "0"
## $ ua.patch <chr> "874"
## $ os.family <chr> "Ubuntu"
## $ os.major <chr> "11"
## $ os.minor <chr> "10"
## $ device.family <chr> "Other"
agents <- readLines(system.file("extdat", "agents.txt", package = "uaparserjs"))
dplyr::glimpse(ua_parse(agents))
## Rows: 1,091
## Columns: 13
## $ userAgent <chr> "Mozilla/5.0 (Windows; U; en-US) AppleWebKit/531.9 (KHTM…
## $ ua.family <chr> "AdobeAIR", "Amazon Silk", "Amazon Silk", "Amazon Silk",…
## $ ua.major <chr> "2", "1", "2", "2", "2", "3", "2", "2", "2", "2", "3", "…
## $ ua.minor <chr> "5", "1", "0", "1", "2", "25", "2", "3", "3", "3", "0", …
## $ ua.patch <chr> "1", "0-80", NA, NA, NA, NA, "2", "3", "4", "5", "1", "3…
## $ os.family <chr> "Windows", "Android", "Android", "Android", "Android", "…
## $ device.family <chr> "Other", "Kindle", "Kindle Fire HD", "Kindle Fire", "Kin…
## $ device.brand <chr> NA, "Amazon", "Amazon", "Amazon", "Amazon", "Amazon", "H…
## $ device.model <chr> NA, "Kindle", "Kindle Fire HD 7\"", "Kindle Fire", "Kind…
## $ os.major <chr> NA, NA, NA, NA, NA, "4", "2", "2", "2", "2", "3", "4", "…
## $ os.minor <chr> NA, NA, NA, NA, NA, "0", "2", "3", "3", "3", "0", "0", "…
## $ os.patch <chr> NA, NA, NA, NA, NA, "3", "2", "3", "4", "5", "1", "3", "…
## $ os.patchMinor <chr> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
Performance
Time for parsing small numbers of UA strings with useNA defaulting to FALSE and list cache
set.seed(100)
#batch_100 <- sample(agents, 100)
assign("cache", list(), envir=uaparserjs:::.pkgenv)
microbenchmark::microbenchmark(
ua_parse(agents)
)
## Unit: seconds
## expr min lq mean median uq max neval
## ua_parse(agents) 1.230242 1.274419 1.306185 1.299452 1.332416 1.522362 100
Time for parsing small numbers of UA strings with list cache and useNA=TRUE
set.seed(100)
#batch_100 <- sample(agents, 100)
assign("cache", list(), envir=uaparserjs:::.pkgenv)
microbenchmark::microbenchmark(
ua_parse(agents,useNA=TRUE)
)
## Unit: milliseconds
## expr min lq mean median uq
## ua_parse(agents, useNA = TRUE) 31.55939 36.4024 44.03564 38.06212 39.90679
## max neval
## 582.4312 100
Time for parsing small numbers of UA strings with environment cache and useNA=TRUE
set.seed(100)
#batch_100 <- sample(agents, 100)
assign("cache", new.env(), envir=uaparserjs:::.pkgenv)
microbenchmark::microbenchmark(
ua_parse(agents,useNA=TRUE)
)
## Unit: milliseconds
## expr min lq mean median uq
## ua_parse(agents, useNA = TRUE) 27.53631 33.13031 40.71838 35.76616 37.97613
## max neval
## 535.2129 100
Effect of useNA
This addresses relative performance with and without the useNA flag. This test exercises the plumbing that handles unpacking the parameter list and compilation of the results with effectively no cache involvement. This test minimises the effect of caching by using the same UA string over and over. This test does not call the underlying Javascript parser repeatedly, values are passed to the JavaScript parser one at a time and parse results are cached for each user agent.
Raw data
class(results2) = "data.frame"
results2
## expr
## 1 uaparserjs::ua_parse(rep(workingUA, i), useNA = f)
## 2 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 2048688, 50, TRUE)
## 3 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 3093938, 100, TRUE)
## 4 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 11911549, 500, TRUE)
## 5 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 23629774, 1000, TRUE)
## 6 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 131288542, 5000, TRUE)
## 7 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 319359199, 10000, TRUE)
## 8 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 1448103817, 50000, TRUE)
## 9 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 14191446, 10, FALSE)
## 10 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 54778869, 50, FALSE)
## 11 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 111332221, 100, FALSE)
## 12 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 538215923, 500, FALSE)
## 13 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 1077447407, 1000, FALSE)
## 14 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 5369082177, 5000, FALSE)
## 15 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 10456872749, 10000, FALSE)
## 16 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 53705485643, 50000, FALSE)
## time UAcount useNA timesec
## 1 2277664 10 TRUE 0.002277664
## 2 2048688 50 TRUE 0.002048688
## 3 3093938 100 TRUE 0.003093938
## 4 11911549 500 TRUE 0.011911549
## 5 23629774 1000 TRUE 0.023629774
## 6 131288542 5000 TRUE 0.131288542
## 7 319359199 10000 TRUE 0.319359199
## 8 1448103817 50000 TRUE 1.448103817
## 9 14191446 10 FALSE 0.014191446
## 10 54778869 50 FALSE 0.054778869
## 11 111332221 100 FALSE 0.111332221
## 12 538215923 500 FALSE 0.538215923
## 13 1077447407 1000 FALSE 1.077447407
## 14 5369082177 5000 FALSE 5.369082177
## 15 10456872749 10000 FALSE 10.456872749
## 16 53705485643 50000 FALSE 53.705485643
Effect of cache misses with large cache sizes
This examines the worst-case effect of cache misses with large cache
sizes. The cache is populated with a number of unique strings, the
number matching the test case size.
Every measured call experiences that number of cache misses before
finding its value in the cache, for example with 1000 User Agent strings
passed in 1000 unique cache entries are created before the test and the
test finds the last cache entry.
The test is coded to force use of a list() because list will be removed in the next release (see test results below).
results = c()
for(f in naFlags)
{
for(i in counts)
{
unloadNamespace("uaparserjs")
assign("cache", list(), envir=uaparserjs:::.pkgenv)
for(j in 1:i)
{
# fake the right number of UA strings for this request cycle
uaparserjs::ua_parse(paste(j, workingUA),useNA = TRUE)
}
bmResult = microbenchmark::microbenchmark(uaparserjs::ua_parse(rep(workingUA,i),useNA = f),times = 1)
bmResult$UAcount = i
bmResult$useNA = f
results = c(results, list(bmResult))
}
}
results2 = do.call("rbind", results)
results2$timesec = results2$time/1000000000
library(ggplot2)
ggplot(results2, aes(UAcount)) +
aes(x=UAcount, y=timesec, colour=useNA) +
geom_line() +
labs(
title = "Elapsed Time for varying numbers of user agent strings per request",
subtitle = "100% cache misses, cache size = request size",
x = "Number of User Agent Strings",
y = "Elased Time (seconds)",
color = "useNA parameter values"
)
Raw Data
class(results2) = "data.frame"
results2
## expr
## 1 uaparserjs::ua_parse(rep(workingUA, i), useNA = f)
## 2 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 3330069, 50, TRUE)
## 3 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 4165148, 100, TRUE)
## 4 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 17082272, 500, TRUE)
## 5 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 32123521, 1000, TRUE)
## 6 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 276723239, 5000, TRUE)
## 7 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 1011251300, 10000, TRUE)
## 8 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 62306937133, 50000, TRUE)
## 9 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 12241732, 10, FALSE)
## 10 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 51719883, 50, FALSE)
## 11 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 105408103, 100, FALSE)
## 12 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 520955656, 500, FALSE)
## 13 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 1115157118, 1000, FALSE)
## 14 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 7709131441, 5000, FALSE)
## 15 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 21918423943, 10000, FALSE)
## 16 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 434529588060, 50000, FALSE)
## time UAcount useNA timesec
## 1 2221922 10 TRUE 0.002221922
## 2 3330069 50 TRUE 0.003330069
## 3 4165148 100 TRUE 0.004165148
## 4 17082272 500 TRUE 0.017082272
## 5 32123521 1000 TRUE 0.032123521
## 6 276723239 5000 TRUE 0.276723239
## 7 1011251300 10000 TRUE 1.011251300
## 8 62306937133 50000 TRUE 62.306937133
## 9 12241732 10 FALSE 0.012241732
## 10 51719883 50 FALSE 0.051719883
## 11 105408103 100 FALSE 0.105408103
## 12 520955656 500 FALSE 0.520955656
## 13 1115157118 1000 FALSE 1.115157118
## 14 7709131441 5000 FALSE 7.709131441
## 15 21918423943 10000 FALSE 21.918423943
## 16 434529588060 50000 FALSE 434.529588060
Effect of cache misses with large cache sizes using an environment instead of list as cache.
Using an environment, the only native hashmap structure in base R, worst case performance is significantly better than using a list. How much this matters depends very much on the distribution of UAs in the input data, but when dealing with two to four million access log records its worth recovering every possible second.
results = c()
library(uaparserjs)
for(f in naFlags)
{
for(i in counts)
{
assign("cache", new.env(), envir=uaparserjs:::.pkgenv)
for(j in 1:i)
{
# fake the right number of UA strings for this request cycle
uaparserjs::ua_parse(paste(j, workingUA),useNA = TRUE)
}
bmResult = microbenchmark::microbenchmark(uaparserjs::ua_parse(rep(workingUA,i),useNA = f),times = 1)
bmResult$UAcount = i
bmResult$useNA = f
results = c(results, list(bmResult))
}
}
results3 = do.call("rbind", results)
results3$timesec = results3$time/1000000000
library(ggplot2)
ggplot(results3, aes(UAcount)) +
aes(x=UAcount, y=timesec, colour=useNA) +
geom_line() +
labs(
title = "Elapsed Time for varying numbers of user agent strings per request",
subtitle = "100% cache misses, cache size = request size",
x = "Number of User Agent Strings",
y = "Elased Time (seconds)",
color = "useNA parameter values"
)
Raw Data
class(results3) = "data.frame"
results3
## expr
## 1 uaparserjs::ua_parse(rep(workingUA, i), useNA = f)
## 2 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 3612107, 50, TRUE)
## 3 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 4212055, 100, TRUE)
## 4 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 13765687, 500, TRUE)
## 5 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 25481705, 1000, TRUE)
## 6 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 128063436, 5000, TRUE)
## 7 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 258059169, 10000, TRUE)
## 8 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 1394473442, 50000, TRUE)
## 9 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 16555736, 10, FALSE)
## 10 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 53977570, 50, FALSE)
## 11 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 102167137, 100, FALSE)
## 12 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 532422069, 500, FALSE)
## 13 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 1062102735, 1000, FALSE)
## 14 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 5524649017, 5000, FALSE)
## 15 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 10968916653, 10000, FALSE)
## 16 uaparserjs::ua_parse(rep(workingUA, i), useNA = f).list(1, 52765686701, 50000, FALSE)
## time UAcount useNA timesec
## 1 3829436 10 TRUE 0.003829436
## 2 3612107 50 TRUE 0.003612107
## 3 4212055 100 TRUE 0.004212055
## 4 13765687 500 TRUE 0.013765687
## 5 25481705 1000 TRUE 0.025481705
## 6 128063436 5000 TRUE 0.128063436
## 7 258059169 10000 TRUE 0.258059169
## 8 1394473442 50000 TRUE 1.394473442
## 9 16555736 10 FALSE 0.016555736
## 10 53977570 50 FALSE 0.053977570
## 11 102167137 100 FALSE 0.102167137
## 12 532422069 500 FALSE 0.532422069
## 13 1062102735 1000 FALSE 1.062102735
## 14 5524649017 5000 FALSE 5.524649017
## 15 10968916653 10000 FALSE 10.968916653
## 16 52765686701 50000 FALSE 52.765686701