What is the most efficient way in Python to convert a string to all lowercase stripping out all non-ascii alpha characters?

I would:

• lowercase the string
• replace all [^a-z] with ""

Like that:

def strip_string_to_lowercase():
  nonascii = re.compile('[^a-z]')
  return lambda s: nonascii.sub('', s.lower().strip())

EDIT: It turns out that the original version (below) is really slow, though some performance can be gained by converting it into a closure (above).

def strip_string_to_lowercase(s):
  return re.sub('[^a-z]', '', s.lower().strip())

My performance measurements with 100,000 iterations against the string

"A235th@#$&( er Ra{}|?>ndom"

revealed that:

• f_re_0 took 2672.000 ms (this is the original version of this answer)
• f_re_1 took 2109.000 ms (this is the closure version shown above)
• f_re_2 took 2031.000 ms (the closure version, without the redundant strip())
• f_fl_1 took 1953.000 ms (unwind’s filter/lambda version)
• f_fl_2 took 1485.000 ms (Coady’s filter version)
• f_jn_1 took 1860.000 ms (Dana’s join version)

For the sake of the test, I did not print the results.

Not especially runtime efficient, but certainly nicer on poor, tired coder eyes:

def strip_string_and_lowercase(s):
    return ''.join(c for c in s.lower() if c in string.ascii_lowercase)

>>> import string
>>> a = "O235th@#$&( er Ra{}|?<ndom"
>>> ''.join(i for i in a.lower() if i in string.ascii_lowercase)
'otheraltndom'

doing essentially the same as you.

This is a typical application of list compehension:

import string
s = "O235th@#$&( er Ra{}|?<ndom"
print ''.join(c for c in s.lower() if c in string.ascii_lowercase)

It won’t filter out “<” (html entity), as in your example, but I assume that was accidental cut and past problem.

Personally I would use a regular expression and then convert the final string to lower case. I have no idea how to write it in Python, but the basic idea is to:

1. Remove characters in string that don’t match case-insensitive regex “w“
2. Convert string to lower-case

or vice versa.

Similar to @Dana’s, but I think this sounds like a filtering job, and that should be visible in the code. Also without the need to explicitly call join():

def strip_string_to_lowercase(s):
  return filter(lambda x: x in string.ascii_lowercase, s.lower())

>>> filter(str.isalpha, "This is a Test").lower()
'thisisatest'
>>> filter(str.isalpha, "A235th@#$&( er Ra{}|?>ndom").lower()
'atherrandom'

I added the filter solutions to Brian’s code:

import string, re, timeit

# Precomputed values (for str_join_set and translate)

letter_set = frozenset(string.ascii_lowercase + string.ascii_uppercase)
tab = string.maketrans(string.ascii_lowercase + string.ascii_uppercase,
                       string.ascii_lowercase * 2)
deletions = ''.join(ch for ch in map(chr,range(256)) if ch not in letter_set)

s="A235th@#$&( er Ra{}|?>ndom"

def test_original(s):
    tmpStr = s.lower().strip()
    retStrList = []
    for x in tmpStr:
        if x in string.ascii_lowercase:
            retStrList.append(x)

    return ''.join(retStrList)


def test_regex(s):
    return re.sub('[^a-z]', '', s.lower())

def test_regex_closure(s):
  nonascii = re.compile('[^a-z]')
  def replacer(s):
    return nonascii.sub('', s.lower().strip())
  return replacer(s)


def test_str_join(s):
    return ''.join(c for c in s.lower() if c in string.ascii_lowercase)

def test_str_join_set(s):
    return ''.join(c for c in s.lower() if c in letter_set)

def test_filter_set(s):
    return filter(letter_set.__contains__, s.lower())

def test_filter_isalpha(s):
    return filter(str.isalpha, s).lower()

def test_filter_lambda(s):
    return filter(lambda x: x in string.ascii_lowercase, s.lower())

def test_translate(s):
    return string.translate(s, tab, deletions)

for test in sorted(globals()):
    if test.startswith("test_"):
        print "%30s : %s" % (test, timeit.Timer("f(s)", 
              "from __main__ import %s as f, s" % test).timeit(200000))

This gives me:

       test_filter_isalpha : 1.31981746283
        test_filter_lambda : 2.23935583992
           test_filter_set : 0.76511679557
             test_original : 2.13079176264
                test_regex : 2.44295629752
        test_regex_closure : 2.65205913042
             test_str_join : 2.25571266739
         test_str_join_set : 1.75565888961
            test_translate : 0.269259640541

It appears that isalpha is using a similar algorithm, at least in terms of O(), to the set algorithm.

Edit:
Added the filter set, and renamed the filter functions to be a little more clear.

Python 2.x translate method

Convert to lowercase and filter non-ascii non-alpha characters:

from string import ascii_letters, ascii_lowercase, maketrans

table = maketrans(ascii_letters, ascii_lowercase*2)
deletechars = ''.join(set(maketrans('','')) - set(ascii_letters))

print "A235th@#$&( er Ra{}|?>ndom".translate(table, deletechars)
# -> 'atherrandom'

Python 3 translate method

Filter non-ascii:

ascii_bytes = "A235th@#$&(٠٫٢٥ er Ra{}|?>ndom".encode('ascii', 'ignore')

Use bytes.translate() to convert to lowercase and delete non-alpha bytes:

from string import ascii_letters, ascii_lowercase

alpha, lower = [s.encode('ascii') for s in [ascii_letters, ascii_lowercase]]
table = bytes.maketrans(alpha, lower*2)           # convert to lowercase
deletebytes = bytes(set(range(256)) - set(alpha)) # delete nonalpha

print(ascii_bytes.translate(table, deletebytes))
# -> b'atherrandom'

Python 2.x:

import string
valid_chars= string.ascii_lowercase + string.ascii_uppercase

def only_lower_ascii_alpha(text):
    return filter(valid_chars.__contains__, text).lower()

Works with either str or unicode arguments.

>>> only_lower_ascii_alpha("Hello there 123456!")
'hellothere'
>>> only_lower_ascii_alpha(u"435 café")
u'caf'