Safe Substitution

String Substitutions

Modifying existing strings via substitution is a common practice in programing. To this end, functions like gsub provide a method to accomplish this. Below is an example where “hey” is replaced with “ho” transforming a line from the Ramones into Santa Claus leaving on Christmas Eve.

s = "hey ho, let's go!"
gsub("hey", "ho", s)
## [1] "ho ho, let's go!"

Simultaneous Substitutions

gsub only supports one string of matching with one string of replacement. What this means is while you can match on multiple conditions, you can only provide one condition of replacement. Below we construct a regular expression which matches on “hey” or “ho” and replaces any such matches with “yo”.

s = "hey ho, let's go!"
gsub("hey|ho", "yo", s)
## [1] "yo yo, let's go!"

If you wanted to replace “hey” with “get” and “ho” with “ready” you would need two steps.

s = "hey ho, let's go!"
s_new = gsub("hey", "get", s)
s_new = gsub("ho", "ready", s_new)
## [1] "get ready, let's go!"

This sequential process however can result in undesired changes. If we want to swap where “hey” and “ho” are, we can see the process breaks down. Because each change happens in order, “hey” becomes “ho” and then every “ho” becomes “hey”, undoing the first step.

s = "hey ho, let's go!"
s_new = gsub("hey", "ho", s)
s_new = gsub("ho", "hey", s_new)
## [1] "hey hey, let's go!"


This is where the idea of mgsub comes in. mgsub is a safe, simultaneous string substitution function. We pass in a patterns to match as well as replacements and the replacements are applied simultaneously.

s = "hey ho, let's go!"
mgsub::mgsub(string = s, pattern = c("hey", "ho"), replacement = c("ho", "hey"))
## [1] "ho hey, let's go!"

Regular Expression Support

mgsub fully supports regular expressions as matching criteria as well as backreferences in the replacement. Note how the matching criteria ignores “dopachloride” for replacement but matches both “Dopazamine” and “dopastriamine” (all fake chemicals despite what the replace string claims!).

## [1] "Metazamine is not the same as dopachloride or Metastriamine, yet is still real."

Furthermore, you can pass through any options from the gsub family. In the example below you can see fixed string matching

## [1] "All my life I chased balloons and dolphins - not success!"

Safe Substitution

This is actually the most compelling feature of mgsub. Several packages implement a similar type function (also named mgsub) which do not employ safe substitution - qdap, bazar and textclean. A detailed analysis of safety can be found on my blog. Here is a quick overview of what is meant by safety:

  1. Longer matches are preferred over shorter matches for substitution first
  2. No placeholders are used so accidental string collisions don’t occur

First, a demonstration of the first form of safety. Note how we are searching for ‘they’ and ‘the’ where ‘the’ is a substring of ‘they’. If ‘the’ is matched before ‘they’, we would expect to see “ay don’t understand the value of what they seek.”, but in both cases, the replacements occur correctly.

## [1] "we don't understand a value of what we seek."
## [1] "we don't understand a value of what we seek."

We can continue to test this by using variable length regular expression matches. Note that we provide two different matching criteria, one a regular expression of length 6 but which matches a length 10 and the other a match of length 9. However, qdap only prioritizes based on the length of the regular expression, not on the actual length of the match. While this is an edge case, it an example of safety provided by mgsub.

## [1] "metazamine is a fake chemical"
## [1] "freakoute is a fake chemical"

In the second case, mgsub does not utilize placeholders and therefore guarantees no string collisions when replacing. Consider a simple example of shifting each word in the following string one spot to the left. mgsub correctly shifts each word while qdap provides two wrong sets of substitutions depending on the other arguments you provide.

## [1] "how, are you hey?"
## [1] "how, are you how?"
## [1] "hey, hey hey hey?"


mgsub pays the price of safety in performance. When only a single string is passed in to be modified, it runs about as fast as qdap. However, when multiple strings are passed in, qdap is far more elastic than mgsub.


s = c("Dopazamine is not the same as Dopachloride and is still fake.",
      "dopazamine is undergoing a review by the fake news arm of the Dopazamine government")
pattern = c("[Dd]opa(.*?mine)", "fake")
replacement = c("Meta\\1", "real")

  mgsub = mgsub::mgsub(s[1], pattern, replacement),
  qdap = qdap::mgsub(pattern, replacement, s[1], fixed = FALSE)
## Unit: microseconds
##   expr     min       lq      mean   median       uq     max neval
##  mgsub 154.401 157.5085 166.09614 163.7445 169.2980 291.607   100
##   qdap  47.920  50.9605  55.57665  55.2865  58.8185  99.124   100
  mgsub = mgsub::mgsub(s, pattern, replacement),
  qdap = qdap::mgsub(pattern, replacement, s, fixed = FALSE)
## Unit: microseconds
##   expr     min       lq      mean   median       uq     max neval
##  mgsub 319.401 326.4465 335.79102 333.4840 337.4685 434.950   100
##   qdap  54.882  60.0235  65.00009  65.9265  67.0450 120.468   100