Created
January 16, 2014 22:12
-
-
Save harlo/8464467 to your computer and use it in GitHub Desktop.
This is the python implementation of our label cleanser (which we ultimately ported to ruby.) In python, we use Levenshtein, but in ruby, we just JaroWinkler. The results are slightly different, so the thresholds had to be adjusted accordingly.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
from collections import namedtuple | |
from Levenshtein import ratio | |
from Levenshtein import distance | |
import re, csv, os | |
delimiter = ',' | |
quotechar = '|' | |
quoting = csv.QUOTE_MINIMAL | |
BrandInfo = namedtuple('BrandInfo', 'model_text utqg_correlate') | |
TirelineInfo = namedtuple('TirelineInfo', 'tireline_text utqg_correlate') | |
debug = True | |
def padSingleDigit(s): | |
""" | |
(I ENDED UP NOT USING THIS) | |
If the DOT id is not 2 digits, pad with a zero. | |
This should only apply to an entity with and ID of 1-9 | |
""" | |
s_o = s | |
is_single_digit = None | |
if len(s) != 2: | |
rx = r'^(\d{1})' | |
is_single_digit = re.findall(rx, s) | |
if len(is_single_digit) == 1: | |
s = str("0%s" % is_single_digit[0]) | |
if debug: | |
print s_o, s, is_single_digit | |
return s | |
def correlateLabel(s, label_set, scrutinize_suffixes=False): | |
correlated = [] | |
CorrelateObject = namedtuple("CorrelateObject", "label lev_ratio lev_dist") | |
pre_rx = r'.*\s(.{3,4})$' | |
for label in label_set: | |
r = ratio(s.lower(), label.lower()) | |
d = distance(s.lower(), label.lower()) | |
if r >= 0.91: | |
''' | |
in some cases, we have to scrutinize our data for | |
false positives. for instance, some manufacturers | |
have multiple lines of a particular product | |
(i.e. Firestone GTX and Firestone GTA) so we have to | |
sniff out these suffixes and compare them. | |
''' | |
if scrutinize_suffixes: | |
s_prefix = re.findall(pre_rx, s.lower()) | |
l_prefix = re.findall(pre_rx, label.lower()) | |
if len(s_prefix) == 1 and len(l_prefix) == 1: | |
if debug: | |
print "s_suffix: %s | l_suffix: %s (%.9f)" % (s_prefix, l_prefix, ratio(s_prefix[0], l_prefix[0])) | |
if ratio(s_prefix[0], l_prefix[0]) < 0.8: | |
continue | |
correlated.append(CorrelateObject(label, r, d)) | |
if len(correlated) > 0: | |
max_val = max(c.lev_ratio for c in correlated) | |
best_matches = [c for c in correlated if c.lev_ratio == max_val] | |
if debug: | |
if max_val >= 0.91 and max_val != 1: | |
print "%s | %s | %s " % (s.upper(), best_matches[0].label.lower(), best_matches) | |
return best_matches[0].label | |
return s | |
if __name__ == "__main__": | |
this_dir = os.path.abspath(__file__) | |
par_dir = os.path.abspath(os.path.join(this_dir, os.pardir)) | |
data_dump = os.path.join(par_dir, 'data') | |
utqg_info = None | |
utqg_brands = None | |
utqg_tirelines = None | |
new_brand_info = None | |
new_tireline_info = None | |
''' | |
open up utqg csv. | |
we want the brand column (1) and tireline column (3). | |
''' | |
UTQGInfo = namedtuple('UTQGInfo', 'brand tireline') | |
with open(os.path.join(data_dump, "UTQG_data.csv"), 'rU') as utqg_data: | |
utqg_csv = csv.reader(utqg_data, delimiter=delimiter, quotechar=quotechar) | |
utqg_info = [UTQGInfo(x[0].lower(), x[2].lower()) for x in utqg_csv] | |
''' | |
create unique sets of brand and tireline observations. | |
''' | |
utqg_brands = list(set([x.brand for x in utqg_info])) | |
utqg_tirelines = list(set([x.tireline for x in utqg_info])) | |
with open(os.path.join(data_dump, "FLAT_CMPL_TIRE_excel.csv"), 'rU') as cmpl_data: | |
cmpl_csv = csv.reader(cmpl_data, delimiter=delimiter, quotechar=quotechar) | |
new_brand_info = [BrandInfo(x[3], correlateLabel(x[3], utqg_brands)) for x in cmpl_csv] | |
new_tireline_info = [TirelineInfo(x[4], correlateLabel(x[4], utqg_tirelines, scrutinize_suffixes=True)) for x in cmpl_csv] | |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment