N-grams
N-grams is a draft programming task. It is not yet considered ready to be promoted as a complete task, for reasons that should be found in its talk page.
An N-gram is a sequence of N contiguous elements of a given text. Although N-grams refer sometimes to words or syllables, in this task we will consider only sequences of characters. The task consists in, given a text and an integer size of the desired N-grams, find all the different contiguous sequences of N characters, together with the number of times they appear in the text. For example, the 2-grams of the text "Live and let live" are:
"LI" - 2 "IV" - 2 "VE" - 2 " L" - 2 "E " - 1 " A" - 1 "AN" - 1 "ND" - 1 "D " - 1 "LE" - 1 "ET" - 1 "T " - 1
Note that space and other non-alphanumeric characters are taken into account.
- See also
Arturo
ngrams: function [s :string n :integer][
0..sub size s n | map 'i -> slice upper s i i+n-1
| tally
]
loop [2 3 4] 'n [
print ~"|n|-grams:"
loop ngrams "Live and let live" n [k v] -> print [~{"|k|"} v]
print ""
]
- Output:
2-grams: "LI" 2 "IV" 2 "VE" 2 "E " 1 " A" 1 "AN" 1 "ND" 1 "D " 1 " L" 2 "LE" 1 "ET" 1 "T " 1 3-grams: "LIV" 2 "IVE" 2 "VE " 1 "E A" 1 " AN" 1 "AND" 1 "ND " 1 "D L" 1 " LE" 1 "LET" 1 "ET " 1 "T L" 1 " LI" 1 4-grams: "LIVE" 2 "IVE " 1 "VE A" 1 "E AN" 1 " AND" 1 "AND " 1 "ND L" 1 "D LE" 1 " LET" 1 "LET " 1 "ET L" 1 "T LI" 1 " LIV" 1
Common Lisp
A hash table is used to store and retrieve the n-grams fast.
(defun n-grams (text n)
"Return a list of all the N-grams of length n in the text, together with their frequency"
(let* (res (*ht-n-grams* (make-hash-table :test 'equal)) )
(loop for i from 0 to (- (length text) n) do
(let* ((n-gram (string-upcase (subseq text i (+ i n))))
(freq (gethash n-gram *ht-n-grams*)))
(setf (gethash n-gram *ht-n-grams*) (if (null freq) 1 (1+ freq))) ))
(maphash #'(lambda (key val)
(push (cons key val) res) )
*ht-n-grams* )
(sort res #'> :key #'cdr) ))
- Output:
> (n-grams "Live and let live" 2)
(("LI" . 2) ("IV" . 2) ("VE" . 2) (" L" . 2) ("E " . 1) (" A" . 1) ("AN" . 1)
("ND" . 1) ("D " . 1) ("LE" . 1) ("ET" . 1) ("T " . 1))
Factor
USING: ascii grouping kernel math.statistics prettyprint ;
: n-grams ( str n -- assoc ) [ >upper ] dip clump histogram ;
"Live and let live" 2 n-grams .
- Output:
H{
{ "ET" 1 }
{ "IV" 2 }
{ "T " 1 }
{ " A" 1 }
{ "VE" 2 }
{ "LI" 2 }
{ "E " 1 }
{ "D " 1 }
{ " L" 2 }
{ "ND" 1 }
{ "LE" 1 }
{ "AN" 1 }
}
jq
Works with jq and gojq, that is, the C and Go implementations of jq.
# Generic "bag of words" utility:
def bow(stream):
reduce stream as $word ({}; .[($word|tostring)] += 1);
# The ngrams as a bow
def ngrams($n):
ascii_upcase as $text
| bow( range(0;$text|1+ length - $n) as $i | $text[$i:$i+$n]);
# The task
# Sort by increasing frequency, then by lexicographical order
def ngrams($text; $n):
($text|ngrams($n)) as $ngrams
| "\nAll \($n)-grams of '\($text)' and their frequencies:",
($ngrams|to_entries|sort_by(.value,.key)[] | "\(.key): \(.value)" ) ;
ngrams("Live and let live"; 2,3,4)- Output:
All 2-grams of 'Live and let live' and their frequencies: A: 1 AN: 1 D : 1 E : 1 ET: 1 LE: 1 ND: 1 T : 1 L: 2 IV: 2 LI: 2 VE: 2 All 3-grams of 'Live and let live' and their frequencies: AN: 1 LE: 1 LI: 1 AND: 1 D L: 1 E A: 1 ET : 1 LET: 1 ND : 1 T L: 1 VE : 1 IVE: 2 LIV: 2 All 4-grams of 'Live and let live' and their frequencies: AND: 1 LET: 1 LIV: 1 AND : 1 D LE: 1 E AN: 1 ET L: 1 IVE : 1 LET : 1 ND L: 1 T LI: 1 VE A: 1 LIVE: 2
Phix
A dictionary is used to find the index of already-seen n-grams, even though a simpler find() would be good enough for this task.
I have replicated most orderings found on this page, the task description order corresponds to orig/freq,
and jq is alpha/freq with high last, but there is no equivalent for the Factor or Raku orderings here ;-).
with javascript_semantics function n_grams(integer len, string txt, sequence orders) sequence ng = {}, ngc = {} integer d = new_dict() txt = upper(txt) for i=1 to length(txt)-len+1 do string tn = txt[i..i+len-1] integer ndx = getdd(tn,0,d) if ndx=0 then ng = append(ng,tn) ngc = append(ngc,1) ndx = length(ng) setd(tn,ndx,d) else ngc[ndx] += 1 end if end for destroy_dict(d) integer l = length(ng) sequence ares = columnize({ng,ngc,tagset(l)}), res = {} for c in orders do if c="original" then -- original/first found order res = append(res,ares) elsif c="orig/freq" then -- "" but higher freq first res = append(res,sort_columns(ares,{-2,3})) elsif c="alphabetic" then -- alphabetical order res = append(res,sort(deep_copy(ares))) elsif c="alpha/freq" then -- "" but higher freq first res = append(res,sort_columns(ares,{-2,1})) else ?9/0 -- (unknown ordering requested) end if end for return res end function constant src = "Live and let live", orders = {"original","orig/freq", "alphabetic","alpha/freq"} printf(1,"For \"%s\":\n",src) for l=2 to 4 do sequence res = n_grams(l,src,orders) string count = ordinal(length(res[1]),true) printf(1,"There are %s unique %d-grams:\n",{count,l}) for i,r in res do printf(1,"%12s: %s\n",{orders[i],join(r,", ",fmt:="%s %d")}) end for end for
- Output:
For "Live and let live":
There are twelve unique 2-grams:
original: LI 2, IV 2, VE 2, E 1, A 1, AN 1, ND 1, D 1, L 2, LE 1, ET 1, T 1
orig/freq: LI 2, IV 2, VE 2, L 2, E 1, A 1, AN 1, ND 1, D 1, LE 1, ET 1, T 1
alphabetic: A 1, L 2, AN 1, D 1, E 1, ET 1, IV 2, LE 1, LI 2, ND 1, T 1, VE 2
alpha/freq: L 2, IV 2, LI 2, VE 2, A 1, AN 1, D 1, E 1, ET 1, LE 1, ND 1, T 1
There are thirteen unique 3-grams:
original: LIV 2, IVE 2, VE 1, E A 1, AN 1, AND 1, ND 1, D L 1, LE 1, LET 1, ET 1, T L 1, LI 1
orig/freq: LIV 2, IVE 2, VE 1, E A 1, AN 1, AND 1, ND 1, D L 1, LE 1, LET 1, ET 1, T L 1, LI 1
alphabetic: AN 1, LE 1, LI 1, AND 1, D L 1, E A 1, ET 1, IVE 2, LET 1, LIV 2, ND 1, T L 1, VE 1
alpha/freq: IVE 2, LIV 2, AN 1, LE 1, LI 1, AND 1, D L 1, E A 1, ET 1, LET 1, ND 1, T L 1, VE 1
There are thirteen unique 4-grams:
original: LIVE 2, IVE 1, VE A 1, E AN 1, AND 1, AND 1, ND L 1, D LE 1, LET 1, LET 1, ET L 1, T LI 1, LIV 1
orig/freq: LIVE 2, IVE 1, VE A 1, E AN 1, AND 1, AND 1, ND L 1, D LE 1, LET 1, LET 1, ET L 1, T LI 1, LIV 1
alphabetic: AND 1, LET 1, LIV 1, AND 1, D LE 1, E AN 1, ET L 1, IVE 1, LET 1, LIVE 2, ND L 1, T LI 1, VE A 1
alpha/freq: LIVE 2, AND 1, LET 1, LIV 1, AND 1, D LE 1, E AN 1, ET L 1, IVE 1, LET 1, ND L 1, T LI 1, VE A 1
Python
This example generates n-grams lazily, much like the sliding_window recipe from the Python itertools docs.
from collections import Counter
from collections import deque
from itertools import islice
def n_grams(text, n):
"""Generate contiguous sequences of _n_ characters from _text_."""
it = iter(text.upper())
ngram = deque(islice(it, n), maxlen=n)
if len(ngram) == n:
yield "".join(ngram)
for ch in it:
ngram.append(ch)
yield "".join(ngram)
if __name__ == "__main__":
import pprint
example = "Live and let live"
for n in range(2, 5):
result = Counter(n_grams(example, n)).most_common()
print(
f"{len(result)} {n}-grams of {example!r}:\n",
pprint.pformat(result, compact=True),
end="\n\n",
)
- Output:
12 2-grams of 'Live and let live':
[('LI', 2), ('IV', 2), ('VE', 2), (' L', 2), ('E ', 1), (' A', 1), ('AN', 1),
('ND', 1), ('D ', 1), ('LE', 1), ('ET', 1), ('T ', 1)]
13 3-grams of 'Live and let live':
[('LIV', 2), ('IVE', 2), ('VE ', 1), ('E A', 1), (' AN', 1), ('AND', 1),
('ND ', 1), ('D L', 1), (' LE', 1), ('LET', 1), ('ET ', 1), ('T L', 1),
(' LI', 1)]
13 4-grams of 'Live and let live':
[('LIVE', 2), ('IVE ', 1), ('VE A', 1), ('E AN', 1), (' AND', 1), ('AND ', 1),
('ND L', 1), ('D LE', 1), (' LET', 1), ('LET ', 1), ('ET L', 1), ('T LI', 1),
(' LIV', 1)]
Raku
sub n-gram ($this, $N=2) { Bag.new( flat $this.uc.map: { .comb.rotor($N => -($N-1))».join } ) }
dd 'Live and let live'.&n-gram; # bi-gram
dd 'Live and let live'.&n-gram(3); # tri-gram
- Output:
("IV"=>2,"T "=>1,"VE"=>2,"E "=>1,"LE"=>1,"AN"=>1,"LI"=>2,"ND"=>1,"ET"=>1," L"=>2," A"=>1,"D "=>1).Bag
("ET "=>1,"AND"=>1,"LIV"=>2," LI"=>1,"ND "=>1," LE"=>1,"IVE"=>2,"E A"=>1,"VE "=>1,"T L"=>1,"D L"=>1,"LET"=>1," AN"=>1).Bag
Wren
import "./str" for Str
import "./maputil" for MapUtil
import "./fmt" for Fmt
var findNgrams = Fn.new { |n, text|
text = Str.upper(text)
var ngrams = {}
for (i in 0..text.count-n) {
var ngram = text[i...i+n]
MapUtil.increase(ngrams, ngram)
}
return ngrams
}
// sort by decreasing frequency, then by lexicographical order
var comparer = Fn.new { |i, j|
if (i.value != j.value) return j.value < i.value
return Str.lt(i.key, j.key)
}
var text = "Live and let live"
for (n in [2, 3, 4]) {
var ngrams = findNgrams.call(n, text)
System.print("All %(n)-grams of '%(text)' and their frequencies:")
var ng = ngrams.toList.sort(comparer).map { |me| "(\"%(me.key)\" : %(me.value))"}
Fmt.tprint("$s ", ng, 5)
System.print()
}- Output:
All 2-grams of 'Live and let live' and their frequencies:
(" L" : 2) ("IV" : 2) ("LI" : 2) ("VE" : 2) (" A" : 1)
("AN" : 1) ("D " : 1) ("E " : 1) ("ET" : 1) ("LE" : 1)
("ND" : 1) ("T " : 1)
All 3-grams of 'Live and let live' and their frequencies:
("IVE" : 2) ("LIV" : 2) (" AN" : 1) (" LE" : 1) (" LI" : 1)
("AND" : 1) ("D L" : 1) ("E A" : 1) ("ET " : 1) ("LET" : 1)
("ND " : 1) ("T L" : 1) ("VE " : 1)
All 4-grams of 'Live and let live' and their frequencies:
("LIVE" : 2) (" AND" : 1) (" LET" : 1) (" LIV" : 1) ("AND " : 1)
("D LE" : 1) ("E AN" : 1) ("ET L" : 1) ("IVE " : 1) ("LET " : 1)
("ND L" : 1) ("T LI" : 1) ("VE A" : 1)