I would like to input a string and return a regular expression that can be used to describe the string's structure. The regex will be used to find more strings of the same structure as the first.
This is intentionally ambiguous because I will certainly miss a case that someone in the SO community will catch.
Please post any and all possible ways to do this.
The trivial answer, and probably not what you want, is: return the input string (with regex special characters escaped). That is always a regular expression that matches the string.
If you wish certain structures to be identified, you have to provide more information about the kind of structures you wish to have identified. Without that information, the problem is stated in an ambiguous way and there are many possible solutions. For instance, the input string 'aba' can be described as
'aba'
'aba*'
'aba?'
'ab\w'
'\w{3}'
'(.)b\1'
etc.
Sorry for the length of this. I took the premise of this as a little challenge and came up with a proof of concept in Ruby.
I worked on the assumption that you could supply a number of strings that should match the regular expression (HITS) and a number that should fail to match (MISSES).
I based the code on a naive implementation of a genetic algorith. See the notes at the bottom for my thoughts on the success, or otherwise, of this approach.
LOOP_COUNT = 100
class Attempt
# let's try email
HITS = %w[[email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected] [email protected]]
MISSES = %w[j@j j@[email protected] j.com @domain.com nochance eric@google. [email protected]. username-at-domain-dot-com linux.org eff.org microsoft.com sjobs.apple.com www.apple.com]
# odd mixture of numbers and letters, designed to confuse
# HITS = %w[a123 a999 a600 a545 a100 b001 b847 a928 c203]
# MISSES = %w[abc def ghi jkl mno pqr stu vwx xyz h234 k987]
# consonants versus vowels
# HITS = %w[bcd cdb fgh ghf jkl klj mnp npm qrs srq tvw vwt xzb bzx]
# MISSES = %w[aei oyu oio euu uio ioe aee ooo]
# letters < 11 chars and no numbers
# HITS = %w[aaa aaaa abaa azaz monkey longstring stringlong]
# MISSES = %w[aa aa1 aa0 b9b 6zz longstringz m_m ff5 666 anotherlongstring]
MAX_SUCCESSES = HITS.size + MISSES.size
# Setup the various Regular Expression operators, etc..
RELEMENTS = %w[. ? * + ( ) \[ \] - | ^ $ \\ : @ / { }]
%w[A b B d D S s W w z Z].each do |chr|
RELEMENTS << "\\#{chr}"
end
%w[alnum alpha blank cntrl digit lower print punct space upper xdigit].each do |special|
RELEMENTS << "[:#{special}:]"
end
('a'..'z').each do |chr|
RELEMENTS << chr
end
('A'..'Z').each do |chr|
RELEMENTS << chr
end
(0..9).each do |chr|
RELEMENTS << chr.to_s
end
START_SIZE = 8
attr_accessor :operators, :successes
def initialize(ary = [])
@operators = ary
if ary.length < 1
START_SIZE.times do
@operators << random_op
end
end
@score = 0
@decay = 1
make_regexp
end
def make_regexp
begin
@regexp = Regexp.new( @operators.join("") )
rescue
# "INVALID Regexp"
@regexp = nil
@score = -1000
end
end
def random_op
RELEMENTS[rand(RELEMENTS.size)]
end
def decay
@decay -= 1
end
def test
@successes = 0
if @regexp
HITS.each do |hit|
result = (hit =~ @regexp)
if result != nil
reward
end
end
MISSES.each do |miss|
result = (miss =~ @regexp)
if result == nil
reward
end
end
end
@score = @successes
self
end
def reward
@successes += 1
end
def cross other
len = size
olen = other.size
split = rand(len)
ops = []
@operators.length.times do |num|
if num < split
ops << @operators[num]
else
ops << other.operators[num + (olen - len)]
end
end
Attempt.new ops
end
# apply a random mutation, you don't have to use all of them
def mutate
send [:flip, :add_rand, :add_first, :add_last, :sub_rand, :sub_first, :sub_last, :swap][rand(8)]
make_regexp
self
end
## mutate methods
def flip
@operators[rand(size)] = random_op
end
def add_rand
@operators.insert rand(size), random_op
end
def add_first
@operators.insert 0, random_op
end
def add_last
@operators << random_op
end
def sub_rand
@operators.delete_at rand(size)
end
def sub_first
@operators.delete_at 0
end
def sub_last
@operators.delete_at size
end
def swap
to = rand(size)
begin
from = rand(size)
end while to == from
@operators[to], @operators[from] = @operators[from], @operators[to]
end
def regexp_to_s
@operators.join("")
end
def <=> other
score <=> other.score
end
def size
@operators.length
end
def to_s
"#{regexp_to_s} #{score}"
end
def dup
Attempt.new @operators.dup
end
def score
if @score > 0
ret = case
when (size > START_SIZE * 2)
@score-20
when size > START_SIZE
@score-2
else
@score #+ START_SIZE - size
end
ret + @decay
else
@score + @decay
end
end
def == other
to_s == other.to_s
end
def stats
puts "Regexp #{@regexp.inspect}"
puts "Length #{@operators.length}"
puts "Successes #{@successes}/#{MAX_SUCCESSES}"
puts "HITS"
HITS.each do |hit|
result = (hit =~ @regexp)
if result == nil
puts "\tFAIL #{hit}"
else
puts "\tOK #{hit} #{result}"
end
end
puts "MISSES"
MISSES.each do |miss|
result = (miss =~ @regexp)
if result == nil
puts "\tOK #{miss}"
else
puts "\tFAIL #{miss} #{result}"
end
end
end
end
$stderr.reopen("/dev/null", "w") # turn off stderr to stop streams of bad rexexp messages
# find some seed attempt values
results = []
10000.times do
a = Attempt.new
a.test
if a.score > 0
# puts "#{a.regexp_to_s} #{a.score}"
results << a
end
end
results.sort!.reverse!
puts "SEED ATTEMPTS"
puts results[0..9]
old_result = nil
LOOP_COUNT.times do |i|
results = results[0..9]
results.map {|r| r.decay }
3.times do
new_results = results.map {|r| r.dup.mutate.test}
results.concat new_results
new_results = results.map {|r| r.cross( results[rand(10)] ).test }
results.concat new_results
end
new_results = []
20.times do
new_results << Attempt.new.test
end
results.concat new_results
results.sort!.reverse!
if old_result != results[0].score
old_result = results[0].score
end
puts "#{i} #{results[0]}"
end
puts "\n--------------------------------------------------"
puts "Winner! #{results[0]}"
puts "--------------------------------------------------\n"
results[0].stats
Lessons learned from playing with this code.
Overall, it appears that running shorter loops several times is most likely to produce a usable result. However, this may be due to my implementation rather than the nature of genetic algorithms.
You may get results that work but still contain parts that are gibberish.
You are going to need a pretty firm grasp of regular expressions to understand how many of the results actually achieve what they do.
Ultimately your time is probably much better spent learning Regular Expressions than trying to use this code as a shortcut. I realise that the questioner may not have had that motive and the reason I tried this was because it was an interesting idea.
There are many trade-offs in the results. The more diverse HITS and MISSES you supply, the longer it will take to produce a result and the more loops you will have to run. Having less of each will likely produce a result that is either massively specific to your supplied strings or so generic that it wouldn't be useful in a real world situation.
I have also hard-coded some assumptions, such as marking down expressions which get too long.