我是一個python n00b,我想就如何改進算法來改進此方法的性能來計算兩個名稱的Jaro-Winkler距離提出一些建議。winkler的Python性能改進請求
def winklerCompareP(str1, str2):
"""Return approximate string comparator measure (between 0.0 and 1.0)
USAGE:
score = winkler(str1, str2)
ARGUMENTS:
str1 The first string
str2 The second string
DESCRIPTION:
As described in 'An Application of the Fellegi-Sunter Model of
Record Linkage to the 1990 U.S. Decennial Census' by William E. Winkler
and Yves Thibaudeau.
Based on the 'jaro' string comparator, but modifies it according to whether
the first few characters are the same or not.
"""
# Quick check if the strings are the same - - - - - - - - - - - - - - - - - -
#
jaro_winkler_marker_char = chr(1)
if (str1 == str2):
return 1.0
len1 = len(str1)
len2 = len(str2)
halflen = max(len1,len2)/2 - 1
ass1 = '' # Characters assigned in str1
ass2 = '' # Characters assigned in str2
#ass1 = ''
#ass2 = ''
workstr1 = str1
workstr2 = str2
common1 = 0 # Number of common characters
common2 = 0
#print "'len1', str1[i], start, end, index, ass1, workstr2, common1"
# Analyse the first string - - - - - - - - - - - - - - - - - - - - - - - - -
#
for i in range(len1):
start = max(0,i-halflen)
end = min(i+halflen+1,len2)
index = workstr2.find(str1[i],start,end)
#print 'len1', str1[i], start, end, index, ass1, workstr2, common1
if (index > -1): # Found common character
common1 += 1
#ass1 += str1[i]
ass1 = ass1 + str1[i]
workstr2 = workstr2[:index]+jaro_winkler_marker_char+workstr2[index+1:]
#print "str1 analyse result", ass1, common1
#print "str1 analyse result", ass1, common1
# Analyse the second string - - - - - - - - - - - - - - - - - - - - - - - - -
#
for i in range(len2):
start = max(0,i-halflen)
end = min(i+halflen+1,len1)
index = workstr1.find(str2[i],start,end)
#print 'len2', str2[i], start, end, index, ass1, workstr1, common2
if (index > -1): # Found common character
common2 += 1
#ass2 += str2[i]
ass2 = ass2 + str2[i]
workstr1 = workstr1[:index]+jaro_winkler_marker_char+workstr1[index+1:]
if (common1 != common2):
print('Winkler: Wrong common values for strings "%s" and "%s"' % \
(str1, str2) + ', common1: %i, common2: %i' % (common1, common2) + \
', common should be the same.')
common1 = float(common1+common2)/2.0 ##### This is just a fix #####
if (common1 == 0):
return 0.0
# Compute number of transpositions - - - - - - - - - - - - - - - - - - - - -
#
transposition = 0
for i in range(len(ass1)):
if (ass1[i] != ass2[i]):
transposition += 1
transposition = transposition/2.0
# Now compute how many characters are common at beginning - - - - - - - - - -
#
minlen = min(len1,len2)
for same in range(minlen+1):
if (str1[:same] != str2[:same]):
break
same -= 1
if (same > 4):
same = 4
common1 = float(common1)
w = 1./3.*(common1/float(len1) + common1/float(len2) + (common1-transposition)/common1)
wn = w + same*0.1 * (1.0 - w)
return wn
示例輸出
ZIMMERMANN ARMIENTO 0.814583333
ZIMMERMANN ZIMMERMANN 1
ZIMMERMANN CANNONS 0.766666667
CANNONS AKKER 0.8
CANNONS ALDERSON 0.845833333
CANNONS ALLANBY 0.833333333
如果您有幾個例子可以解決......幾個名稱對和正確的分數,這將有所幫助。否則,它很難知道我們的編輯沒有打破功能。 – JudoWill 2010-04-30 02:40:02
感謝您編輯問題。我會得到這些例子。 – Martlark 2010-04-30 04:11:04
我沒有得到與您的示例相同的輸出(除非字符串完全相同),但w對於我從Jaro-Winkler距離維基百科頁面嘗試的示例是正確的。 – 2010-04-30 05:05:03