用最大熵模型進行字標註中文分詞(Python實現)
同前面的那篇文章一樣(參見:最大熵模型進行中文分詞),本文運用字標註法進行中文分詞,分別使用4-tag和6-tag對語料進行字標註,觀察分詞效果。前面的文章中使用了模型工具包中自帶的一個樣例進行4-tag中文分詞,但由於其選取的特徵是針對英文詞性標註開發的,故準確率和召回率較低(召回率為83.7%,準確率為84.1%)。 PS:為什麼用作詞性標註的特徵也可以用來做分詞呢?這是因為最大熵模型處理問題的時候和具體的實際問題是無關的,模型處理任何問題都會轉化為對序列資料的標註問題。因此序列化標註的問題都可以通過最大熵模型解決,只不過針對不同的問題,特徵的選取也不太一樣。特徵模板的的優劣會影響到結果的正確性。
第一部分 最大熵模型工具包安裝說明
最大熵模型方面使用開源的張樂博士的最大熵模型工具包(Maximum Entropy Modeling Toolkit for Python and C++)。使用的中文語料資源是SIGHAN提供的backoff 2005語料,目前封閉測試最好的結果是4-tag+CFR標註分詞,在北大語料庫上可以在準確率,召回率以及F值上達到92%以上的效果,在微軟語料庫上可以到達96%以上的效果。以下我們將轉入這篇文章的主題,基於最大熵模型的字標註中文分詞。
下載安裝和使用張樂博士的最大熵模型工具包,本文使用的是其在github上的程式碼:
1.進入到程式碼主目錄maxent-master後,正常按照“configure & make & (sudo) make install就可以完成C++庫的安裝。
注意:(1)gcc編譯器最好是4.7版本以上,我試過在4.4.3上面是不成功的,升級到4.7之後就可以了,具體請參閱:升級Ubuntu中g++和gcc的版本)。
(2)gcc版本沒問題了,如果報出 ./configure錯誤,請參閱:Linux下./configure錯誤詳解。
2.再進入到子目錄python下,安裝python包:python setup.py build & (sudo) python setup.py install,這個python庫是通過強大的
注意:中間如果報出:python.h 沒有各個檔案或目錄 的錯誤,請參閱:解決python.h 沒有那個檔案或目錄 的方法。
關於這個最大熵模型工具包詳情及背景,推薦看官方manual文件,寫得非常詳細。
第二部分 4-tag和6-tag
1.字標註。
什麼是字標註呢?先看一個句子:我是一名程式設計師。將所有字分為4類,S表示單字,B表示詞首,M表示詞中,E表示詞尾。
如果我們知道上述句子中每個字的類別,即:
我/S 是/S 一/B 名/E 程/B 序/M 員/E 。/S
那麼我們就可以知道這個句子的分詞結果:我 是 一名 程式設計師 。
從這裡可以看出,分詞問題轉化成了一個分類問題,即對每個字分類。我們知道,機器學習方法可以很好地處理分類問題。 所以接下來,我們需要用機器學習的方法來解決分詞。
2. 4-tag && 6-tag
如果把字的型別分為四類(4-tag),就是:S(單字),B(詞首),M(詞中),E(詞尾)。
如果把字的型別分成六類(6-tag),就是:S(單字),B(詞首),C(第二個字),D(第三個字),M(詞中),E(詞尾)。
例如:如果【中華人民共和國】作為一個詞的話,按照4-tag,描述為:中/B 華/M 人/M 民/M 共/M 和/M 國/E。如果按照6-tag,描述為:中/B 華/C 人/D 民/M 共/M 和/M 國/E。
從這裡可以看出,分詞問題轉化成了一個分類問題,即對每個字分類。我們知道,機器學習方法可以很好地處理分類問題。 所以接下來,我們需要用機器學習的方法來解決分詞。本文將分別用4-tag和6-tag對訓練語料和測試語料進行標註,利用訓練語料求得對應的模型,得到測試語料的分詞結果。
第三部分 機器學習方法機器學習處理問題的基本思想是,首先對標註好的資料進行訓練,得到模型,再根據模型對新資料進行預測。這樣我們將問題轉化為三個子問題:
(1)標註資料是什麼?
(2)模型是什麼?
(3)如何使用模型進行預測?
具體到以字分詞問題,我們簡單地回答上述三個問題。
(1)標註好的資料就是:字x在情境A下類別為a;字x在情境B下類別為b;字x在情境C下類別為c……
(2)模型就是:一些公式。這些公式是對資料的一種描述,其中包含了標註好的資料的資訊,以及一些未知的引數,對模型的訓練就是採用機器學習的方法,賦予未知的引數一些合適的值,這些值使得公式的值達到最優。
(3)如何預測:給定字x,情境Z,我們得到模型後,就可以根據 x,Z,模型,得到在這種情況下x取各個類別的概率是多少,由此可以預測出x的類別。
在上述回答中,有幾個地方是沒有定義的,例如,什麼叫情境。在以字分詞中,情境就是對一個字環境的描述,例如在句子我是程式設計師中,如果我們將一個字的情境定義為:這個字前面的字和這個字後面的字。那麼“程”這個字的情境為:“是”,以及“序”。所以,所謂標註好的資料就是:{類別:情境}的集合,我們舉例說明。
將C-1定義為字前的字,C0定義為當前字,C1定義為字後的字,那麼如果我們有一句分好詞的句子:我 是 一名 程式設計師 。
首先我們需要對每個字打個標籤,表示類別:我/S 是/S 一/B 名/E 程/B 序/M 員/E 。/S
然後就可以轉化為標註資料了,例如“程”字,可以得到一條標註資料為: B C-1=名 C0=程 C1=序。這條標註資料意思為當遇到C-1=名 C0=程 C1=序這種情境時,類別為B。
所以,由一堆分好詞的資料,就可以得到許多條這樣的標註資料。得到標註資料後,就可以使用機器學習包進行訓練,訓練後可以得到相應的模型,預測時匯入模型,然後給出相應情境(即C-1=? C0=? C1=?),就可以知道在此情境下各個類別的概率,在此可以簡單地 認識哪個類別概率大,結果就是哪個類別,當然,也可以設計其它的方式。
在整個過程中,情境,機器學習方法,預測的方法都是可變的,但是以字分詞大概的思想應該都是這樣的。
第四部分 分詞實踐
1.語料來源
(1)訓練資料:icwb2-data/training/pku_ training.utf8
(2)測試資料:icwb2-data/testing/pku_ test.utf8
(3)正確分詞結果:icwb2-data/gold/pku_ test_ gold.utf8
(4)評分工具:icwb2-data/script/socre
2.分詞過程資料示例
注:下面展示的僅僅是資料示例,並非完整的資料
(1)訓練語料
(2)訓練語料的字標註結果
(3)訓練語料的特徵
其中,Ci 表示與當前字偏移為i的字
(4)測試語料
(5)測試語料的特徵
(6)測試語料的字標註結果
(7)測試語料的分詞結果
3.計算準確率和召回率以及F值
有了這個字標註分詞結果,我們就可以利用backoff2005的測試指令碼來測一下這次分詞的效果了:
./icwb2-data/scripts/score ./icwb2-data/gold/pku_training_words.utf8 ./icwb2-data/gold/pku_test_gold.utf8 pku_result.utf8 > pku_maxent.score
其中,pku_result.utf8 是你的結果檔案,pku_maxent.score是你指定的存放評分結果的檔名
第五部分 分詞結果
下面是測試結果:
類別 | 迭代次數 | 真實詞數 | 得到詞數 | 召回率 | 準確率 | F值 |
4-tag | 100 | 104372 | 102810 | 89.40% | 90.70% | 0.9 |
4-tag | 150 | 104372 | 102806 | 89.40% | 90.80% | 0.901 |
4-tag | 200 | 104372 | 102808 | 89.50% | 90.80% | 0.901 |
6-tag | 100 | 104372 | 103246 | 91.00% | 92.00% | 0.915 |
6-tag | 150 | 104372 | 103240 | 91.00% | 92.00% | 0.915 |
6-tag | 200 | 104372 | 103237 | 91.10% | 92.10% | 0.915 |
6-tag | 300 | 104372 | 103212 | 91.10% | 92.10% | 0.916 |
從結果來看,6-tag的結果要優於4-tag的結果。最高值為:召回率91.1%,準確率92.1,F值為0.916。
第六部分 原始碼
#! /usr/bin/env python
# -*- coding: utf-8 -*-
# 基於最大熵模型及字標註的分詞工具4-tag && 6-tag
import codecs
import sys
from maxent import MaxentModel
# 標註訓練集6-tag
def tag6_training_set(training_file, tag_training_set_file):
fin = codecs.open(training_file, 'r', 'utf-8')
contents = fin.read()
contents = contents.replace(u'\r', u'')
contents = contents.replace(u'\n', u'')
words = contents.split(' ')
print len(words)
tag_words_list = []
i = 0
for word in words:
i += 1
if(i % 100 == 0):
tag_words_list.append(u'\r')
if(len(word) == 0):
continue
if(len(word) == 1):
tag_word = word + '/S'
elif (len(word) == 2):
tag_word = word[0] + '/B' + word[1] + '/E'
elif (len(word) == 3):
tag_word = word[0] + '/B' + word[1] + '/C' + word[2] + '/E'
elif (len(word) == 4):
tag_word = word[0] + '/B' + word[1] + '/C' + word[2] + '/D' + word[3] + '/E'
else:
tag_word = word[0] + '/B' + word[1] + '/C' + word[2] + '/D'
mid_words = word[3:-1]
for mid_word in mid_words:
tag_word += (mid_word + '/M')
tag_word += (word[-1] + '/E')
tag_words_list.append(tag_word)
tag_words = ''.join(tag_words_list)
fout = codecs.open(tag_training_set_file, 'w', 'utf-8')
fout.write(tag_words)
fout.close()
return (words, tag_words_list)
# 標註訓練集4-tag
def tag4_training_set(training_file, tag_training_set_file):
fin = codecs.open(training_file, 'r', 'utf-8')
contents = fin.read()
contents = contents.replace(u'\r', u'')
contents = contents.replace(u'\n', u'')
words = contents.split(' ')
print len(words)
tag_words_list = []
i = 0
for word in words:
i += 1
if(i % 100 == 0):
tag_words_list.append(u'\r')
if(len(word) == 0):
continue
if(len(word) == 1):
tag_word = word + '/S'
elif (len(word) == 2):
tag_word = word[0] + '/B' + word[1] + '/E'
else:
tag_word = word[0] + '/B' + word[1] + '/C' + word[2] + '/D'
mid_words = word[1:-1]
for mid_word in mid_words:
tag_word += (mid_word + '/M')
tag_word += (word[-1] + '/E')
tag_words_list.append(tag_word)
tag_words = ''.join(tag_words_list)
fout = codecs.open(tag_training_set_file, 'w', 'utf-8')
fout.write(tag_words)
fout.close()
return (words, tag_words_list)
def get_near_char(contents, i, times):
words_len = len(contents) / times;
if (i < 0 or i > words_len - 1):
return '_'
else:
return contents[i*times]
def get_near_tag(contents, i ,times):
words_len = len(contents) / times;
if (i < 0 or i > words_len - 1):
return '_'
else:
return contents[i*times*2]
def isPu(char):
punctuation = [u',', u'。', u'?', u'!', u';', u'--', u'、', u'——', u'(', u')', u'《', u'》', u':', u'“', u'”', u'’', u'‘']
if char in punctuation:
return '1'
else:
return '0'
def get_class(char):
zh_num = [u'零', u'○', u'一', u'二', u'三', u'四', u'五', u'六', u'七', u'八', u'九', u'十', u'百', u'千', u'萬']
ar_num = [u'0', u'1', u'2', u'3', u'4', u'5', u'6', u'7', u'8', u'9', u'.', u'0',u'1',u'2',u'3',u'4',u'5',u'6',u'7',u'8',u'9']
date = [u'日', u'年', u'月']
letter = ['a', 'b', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'g', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z']
if char in zh_num or char in ar_num:
return '1'
elif char in date:
return '2'
elif char in letter:
return '3'
else:
return '4'
# 獲取訓練集特徵
def get_event(tag_file_path, event_file_path):
f = codecs.open(tag_file_path, 'r', 'utf-8')
contents = f.read()
contents = contents.replace(u'\r', u'')
contents = contents.replace(u'\n', u'')
words_len = len(contents)/3
event_list = []
index = range(0, words_len)
for i in index:
pre_char = get_near_char(contents, i-1, 3)
pre_pre_char = get_near_char(contents, i-2, 3)
cur_char = get_near_char(contents, i, 3)
next_char = get_near_char(contents, i+1, 3)
next_next_char = get_near_char(contents, i+2, 3)
event_list.append(
contents[i*3+2] + ' '
+ 'C-2='+pre_pre_char + ' ' + 'C-1='+pre_char + ' '
+ ' ' + 'C0=' + cur_char + ' '
+ 'C1=' + next_char + ' ' + 'C2=' + next_next_char + ' '
+ 'C-2=' + pre_pre_char + 'C-1=' + pre_char + ' '
+ 'C-1=' + pre_char + 'C0=' + cur_char + ' '
+ 'C0=' + cur_char + 'C1=' + next_char + ' '
+ 'C1=' + next_char + 'C2=' + next_next_char + ' '
+ 'C-1=' + pre_char + 'C1=' + next_char + ' '
+ 'C-2=' + pre_pre_char + 'C-1=' + pre_char + 'C0=' + cur_char + ' '
+ 'C-1=' + pre_char + 'C0=' + cur_char + 'C1=' + next_char + ' '
+ 'C0=' + cur_char + 'C1=' + next_char + 'C2=' + next_next_char + ' '
+ 'Pu=' + isPu(cur_char) + ' '
+ 'Tc-2=' + get_class(pre_pre_char) + 'Tc-1=' + get_class(pre_char)
+ 'Tc0=' + get_class(cur_char) + 'Tc1=' + get_class(next_char)
+ 'Tc2=' + get_class(next_next_char) + ' '
+ '\r')
# events = ''.join(event_list)
fout = codecs.open(event_file_path, 'w', 'utf-8')
for event in event_list:
fout.write(event)
fout.close()
return event_list
# 測試集生成特徵
def get_feature(test_file_path, feature_file_path):
f = codecs.open(test_file_path, 'r', 'utf-8')
contents = f.read()
contents_list = contents.split('\r\n')
contents_list.remove('')
contents_list.remove('')
fout = codecs.open(feature_file_path, 'w', 'utf-8')
for line in contents_list:
words_len = len(line)
feature_list = []
index = range(0, words_len)
for i in index:
pre_char = get_near_char(line, i-1, 1)
pre_pre_char = get_near_char(line, i-2, 1)
cur_char = get_near_char(line, i, 1)
next_char = get_near_char(line, i+1, 1)
next_next_char = get_near_char(line, i+2, 1)
feature_list.append(
'C-2=' + pre_pre_char + ' ' + 'C-1=' + pre_char + ' '
+ 'C0=' + cur_char + ' '
+ 'C1=' + next_char + ' ' + 'C2=' + next_next_char + ' '
+ 'C-2=' + pre_pre_char + 'C-1=' + pre_char + ' '
+ 'C-1=' + pre_char + 'C0=' + cur_char + ' '
+ 'C0=' + cur_char + 'C1=' + next_char + ' '
+ 'C1=' + next_char + 'C2=' + next_next_char + ' '
+ 'C-1=' + pre_char + 'C1=' + next_char + ' '
+ 'C-2=' + pre_pre_char + 'C-1=' + pre_char + 'C0=' + cur_char + ' '
+ 'C-1=' + pre_char + 'C0=' + cur_char + 'C1=' + next_char + ' '
+ 'C0=' + cur_char + 'C1=' + next_char + 'C2=' + next_next_char + ' '
+ 'Pu=' + isPu(cur_char) + ' '
+ 'Tc-2=' + get_class(pre_pre_char) + 'Tc-1=' + get_class(pre_char)
+ 'Tc0=' + get_class(cur_char) + 'Tc1=' + get_class(next_char)
+ 'Tc2=' + get_class(next_next_char) + ' '
+ '\r')
for item in feature_list:
fout.write(item)
fout.write('split\r\n')
fout.close()
return feature_list
#
def split_by_blank(line):
line_list = []
line_len = len(line)
i = 0
while i < line_len:
line_list.append(line[i])
i += 2
return line_list
# 訓練模型
def training(feature_file_path, trained_model_file, times):
m = MaxentModel()
fin = codecs.open(feature_file_path, 'r', 'utf-8')
all_list = []
m.begin_add_event()
for line in fin:
line = line.rstrip()
line_list = line.split(' ')
str_list = []
for item in line_list:
str_list.append(item.encode('utf-8'))
all_list.append(str_list)
m.add_event(str_list[1:], str_list[0], 1)
m.end_add_event()
print 'begin training'
m.train(times, "lbfgs")
print 'end training'
m.save(trained_model_file)
return all_list
#
def max_prob(label_prob_list):
max_prob = 0
max_prob_label = ''
for label_prob in label_prob_list:
if label_prob[1] > max_prob:
max_prob = label_prob[1]
max_prob_label = label_prob[0]
return max_prob_label
# 標註測試集
def tag_test(test_feature_file, trained_model_file, tag_test_set_file):
fin = codecs.open(test_feature_file, 'r', 'utf-8')
fout = codecs.open(tag_test_set_file, 'w', 'utf-8')
m = MaxentModel()
m.load(trained_model_file)
contents = fin.read()
feature_list = contents.split('\r')
feature_list.remove('\n')
for feature in feature_list:
if (feature == 'split'):
fout.write('\n\n\n')
continue
str_feature = []
u_feature = feature.split(' ')
for item in u_feature:
str_feature.append(item.encode('utf-8'))
label_prob_list = m.eval_all(str_feature)
label = max_prob(label_prob_list)
try:
new_tag = str_feature[2].split('=')[1] + '/' + label
except IndexError:
print str_feature
fout.write(new_tag.decode('utf-8'))
pre_tag = label
return feature_list
# 獲取最終結果6-tag
def tag6_to_words(tag_training_set_file, result_file):
fin = codecs.open(tag_training_set_file, 'r', 'utf-8')
fout = codecs.open(result_file, 'w', 'utf-8')
contents = fin.read()
words_len = len(contents) / 3
result = []
i = 0
while (i < words_len):
cur_word_label = contents[i*3+2]
cur_word = contents[i*3]
if (cur_word_label == 'S'):
result.append(cur_word + ' ')
elif (cur_word_label == 'B'):
result.append(cur_word)
elif (cur_word_label == 'C'):
result.append(cur_word)
elif (cur_word_label == 'D'):
result.append(cur_word)
elif (cur_word_label == 'M'):
result.append(cur_word)
elif (cur_word_label == 'E'):
result.append(cur_word + ' ')
else:
result.append(cur_word)
i += 1
fout.write(''.join(result))
# 獲取最終結果4-tag
def tag4_to_words(tag_training_set_file, result_file):
fin = codecs.open(tag_training_set_file, 'r', 'utf-8')
fout = codecs.open(result_file, 'w', 'utf-8')
contents = fin.read()
words_len = len(contents) / 3
result = []
i = 0
while (i < words_len):
cur_word_label = contents[i*3+2]
cur_word = contents[i*3]
if (cur_word_label == 'S'):
result.append(cur_word + ' ')
elif (cur_word_label == 'B'):
result.append(cur_word)
elif (cur_word_label == 'M'):
result.append(cur_word)
elif (cur_word_label == 'E'):
result.append(cur_word + ' ')
else:
result.append(cur_word)
i += 1
fout.write(''.join(result))
#
def main():
args = sys.argv[1:]
if len(args) < 3:
print 'Usage: python ' + sys.argv[0] + ' training_file test_file result_file'
exit(-1)
training_file = args[0]
test_file = args[1]
result_file = args[2]
# 標註訓練集
tag_training_set_file = training_file + ".tag"
#tag4_training_set(training_file, tag_training_set_file)
tag6_training_set(training_file, tag_training_set_file)
print 'tag train set succeed'
# 獲取訓練集特徵
feature_file_path = training_file + ".feature"
get_event(tag_training_set_file, feature_file_path)
print 'get training set feature succeed'
# 測試集生成特徵
test_feature_file = test_file + ".feature"
get_feature(test_file, test_feature_file)
print 'get test set features succeed'
# 訓練模型
times = [10]
for time in times:
trained_model_file = training_file + '.' + str(time) + ".model"
training(feature_file_path, trained_model_file, time)
print 'training model succeed: ' + str(time)
# 標註測試集
tag_test_set_file = test_file + ".tag"
tag_test(test_feature_file, trained_model_file, tag_test_set_file)
print 'tag test set succeed'
# 獲取最終結果
#tag4_to_words(tag_test_set_file, result_file + '.' + str(time))
tag6_to_words(tag_test_set_file, result_file + '.' + str(time))
print 'get final result succeed ' + result_file + '.' + str(time)
'''
def main():
args = sys.argv[1:]
if len(args) < 3:
print 'Usage: python ' + sys.argv[0] + ' trainen_model_file test_file result_file'
exit(-1)
trained_model_file = args[0]
test_file = args[1]
result_file = args[2]
# 測試集生成特徵
test_feature_file = test_file + ".feature"
get_feature(test_file, test_feature_file)
print 'get test set features succeed'
# 訓練模型
times = [10]
for time in times:
#trained_model_file = training_file + '.' + str(time) + ".model"
#training(feature_file_path, trained_model_file, time)
#print 'training model succeed: ' + str(time)
# 標註測試集
tag_test_set_file = test_file + ".tag"
tag_test(test_feature_file, trained_model_file, tag_test_set_file)
print 'tag test set succeed'
# 獲取最終結果
tag_to_words(tag_test_set_file, result_file + '.' + str(time))
print 'get final result succeed ' + result_file + '.' + str(time)
'''
if __name__ == "__main__":
main()
參考資料:
本文測試過程主要參考了下面的文章:
【1】使用Python,字標註及最大熵法進行中文分詞http://blog.csdn.net/on_1y/article/details/9769919
相關推薦
用最大熵模型進行字標註中文分詞(Python實現)
同前面的那篇文章一樣(參見:最大熵模型進行中文分詞),本文運用字標註法進行中文分詞,分別使用4-tag和6-tag對語料進行字標註,觀察分詞效果。前面的文章中使用了模型工具包中自帶的一個樣例進行4-tag中文分詞,但由於其選取的特徵是針對英文詞性標註開發
用條件隨機場CRF進行字標註中文分詞(Python實現)
本文運用字標註法進行中文分詞,使用4-tag對語料進行字標註,觀察分詞效果。模型方面選用開源的條件隨機場工具包“CRF++: Yet Another CRF toolkit”進行分詞。 本文使用的中文語料資源是SIGHAN提供的backof
最大熵模型進行中文分詞
本文運用字標註法進行中文分詞,最大熵模型方面使用開源的張樂博士的最大熵模型工具包(Maximum Entropy Modeling Toolkit for Python and C++)。使用的中文語料資源是SIGHAN提供的backoff 2005語料,
最大熵模型
定性 全部 投資 情況 進行 算法 出了 信息 簡單 我們不要把雞蛋都放在一個籃子裏面講得就是最大熵原理,從投資的角度來看這就是風險最小原則。從信息論的角度來說,就是保留了最大的不確定性,也就是讓熵達到了最大。最大熵院裏指出,對一個隨機事件的概率分布進行預測的時候,我
通俗理解最大熵模型
log logs ima 最大熵 ges es2017 最大熵模型 blog image 通俗理解最大熵模型
淺談最大熵模型中的特徵
最近在看到自然語言處理中的條件隨機場模型時,發現了裡面涉及到了最大熵模型,這才知道最大熵模型自己還是一知半解,於是在知乎上查閱了很多資料,發現特別受用,飲水思源,我將自己整理的一些資料寫下來供大家參考 僅僅對輸入抽取特徵。即特徵函式為 對輸入和輸出同時抽取特徵。即特徵函式為
斯坦福大學-自然語言處理入門 筆記 第十一課 最大熵模型與判別模型(2)
一、最大熵模型 1、模型介紹 基本思想:我們希望資料是均勻分佈的,除非我們有其他的限制條件讓給我們相信資料不是均勻分佈的。均勻分佈代表高熵(high entropy)。所以,最大熵模型的基本思想就是我們要找的分佈是滿足我們限制條件下,同時熵最高的分佈。 熵:表示分佈的不
斯坦福大學-自然語言處理入門 筆記 第八課 最大熵模型與判別模型
一、生成模型與判別模型 1、引言 到目前為止,我們使用的是生成模型(generative model),但是在實際使用中我們也在大量使用判別模型(discriminative model),主要是因為它有如下的優點: 準確性很高 更容易包含很多和
【統計學習方法-李航-筆記總結】六、邏輯斯諦迴歸和最大熵模型
本文是李航老師《統計學習方法》第六章的筆記,歡迎大佬巨佬們交流。 主要參考部落格: http://www.cnblogs.com/YongSun/p/4767100.html https://blog.csdn.net/tina_ttl/article/details/53519391
統計學習---邏輯斯蒂迴歸與最大熵模型
邏輯斯蒂迴歸和最大熵模型 邏輯斯蒂分佈 邏輯斯蒂迴歸模型 將權值向量和輸入向量加以擴充後的邏輯斯蒂模型為 模型引數估計 極大似然估計法 最大熵模型 最大熵原理:在所有可能的概率模型中,熵最大的模型是最好的模型。通常用約
【機器學習】最大熵模型原理小結
最大熵模型(maximum entropy model, MaxEnt)也是很典型的分類演算法了,它和邏輯迴歸類似,都是屬於對數線性分類模型。在損失函式優化的過程中,使用了和支援向量機類似的凸優化技術。而對熵的使用,讓我們想起了決策樹演算法中的ID3和C4.5演算法。理解了最
最大熵模型(MaxEnt)解析
給出了最大熵模型的一般形式(其中的f為特徵函式,後面我們還會講到): 而文獻【5】中我們從另外一種不同的角度也得出了多元邏輯迴歸的一般形式: 可見,儘管採用的方法不同,二者最終是殊途同歸、萬法歸宗了。 所以我們說無論是多元邏輯迴歸,還是最大熵模型,又或者是Sof
一些對最大熵模型的理解
一、最大熵原理 概念:對於隨機變數X,其概率分佈為P(X),一般在約束條件下會有無數P(X)存在。最大熵原理就是在所有符合約束條件的P(X)中,熵最大的模型即為最優模型。 二、最大熵模型 最大熵模型,就是基於最大熵原理的分類模型。李航《統計學習方法》中對最大熵模型的描述
最大熵模型中的數學推導
最大熵模型中的數學推導 0 引言 寫完SVM之後,一直想繼續寫機器學習的系列,無奈一直時間不穩定且對各個模型演算法的理
NLP --- 最大熵模型的解法(GIS演算法、IIS演算法)
上一節中我們詳細的介紹了什麼是最大熵模型,也推匯出了最大熵模型的目標公式,但是沒給出如何求解的問題,本節將詳細講解GIS演算法求解最大熵模型的過程,這裡先把上一節的推匯出的公式拿過來: 上面第一個式子是說我們要尋找的P要滿足k個約束條件,下式說是在滿足的約束的情況下,找到是熵值最大的那
NLP --- 最大熵模型的引入
前幾節我們詳細的闡述了什麼是HMM,同時給出了HMM的三個問題,也給出瞭解決這三個問題的方法最後給出了HMM的簡單的應用。其中為了解決第三個問題我們引入了EM演算法,這個演算法有點麻煩,但是不難理解,而解決第一個和第二個問題時使用的演算法基本上都是基於動態規劃的,這裡需要大家首先對動態規劃演算法
《統計學習方法(李航)》邏輯斯蒂迴歸與最大熵模型學習筆記
作者:jliang https://blog.csdn.net/jliang3 1.重點歸納 1)線性迴歸 (1)是確定兩種或以上變數間相互依賴的定量關係的一種統計分析方法。 (2)模型:y=wx+b (3)誤差函式: (4)常見求解方法 最小
深入解析最大熵模型
不要把雞蛋放到一個籃子裡 理解了這句話其實已經理解了最大熵模型的精髓了,不過這句話還是有點含蓄,下面講一下我的理解,歡迎交流。 “不要把雞蛋放到一個籃子裡”,這樣可以降低風險。 為啥不放到一個籃子裡就可以降低風險啊?如果有人告訴你就算世界毀滅這個籃子也不會破也不會摔倒地上,那麼就永遠
十、最大熵模型與EM演算法
一、最大熵模型 lnx<=x−1lnx<=x−1 證明:f(x)=x−1−lnx,x>0f(x)=x−1−lnx,x>0,求導是凸函式,在x=1處取得極值 1、熵 熵是資訊的度量,與資訊量成反比。
最大熵模型及其python實現
剛開始學習最大熵模型的時候,自以為書中的推導都看明白了。等到自己實現時才發現問題多多。因此,這篇部落格將把重點放在python程式的解讀上,為什麼說是解讀呢,因為這個程式不是我寫的(輕點噴~~),這個程式參考了網上的一篇部落格,地址:http://blog.cs