improve POS tagging

jieba/__init__.py

@@ -82,7 +82,7 @@ def calc(sentence,DAG,idx,route):
 		candidates = [ ( FREQ.get(sentence[idx:x+1],min_freq) * route[x+1][0],x ) for x in DAG[idx] ]
 		route[idx] = max(candidates)
 
-def __cut_DAG(sentence):
+def get_DAG(sentence):
 	N = len(sentence)
 	i,j=0,0
 	p = trie
@@ -107,11 +107,15 @@ def __cut_DAG(sentence):
 	for i in xrange(len(sentence)):
 		if not i in DAG:
 			DAG[i] =[i]
-	#pprint.pprint(DAG)
+	return DAG
+
+def __cut_DAG(sentence):
+	DAG = get_DAG(sentence)
 	route ={}
 	calc(sentence,DAG,0,route=route)
 	x = 0
 	buf =u''
+	N = len(sentence)
 	while x<N:
 		y = route[x][1]+1
 		l_word = sentence[x:y]

jieba/posseg/__init__.py

@@ -1,6 +1,8 @@
 import re
 import os
 import viterbi
+import jieba
+
 
 def load_model(f_name):
 	_curpath=os.path.normpath( os.path.join( os.getcwd(), os.path.dirname(__file__) )  )
@@ -19,9 +21,32 @@ prob_trans = load_model("prob_trans.py")
 prob_emit = load_model("prob_emit.py")
 char_state_tab = load_model("char_state_tab.py")
 
+class pair(object):
+	def __init__(self,word,flag):
+		self.word = word
+		self.flag = flag
+
+	def __unicode__(self):
+		return self.word+u"/"+self.flag
+
+	def __repr__(self):
+		return self.__unicode__()
 
-def __cut(sentence):
-	prob, pos_list =  viterbi.viterbi(sentence,char_state_tab, prob_start, prob_trans, prob_emit)
+	def encode(self,arg):
+		return self.__unicode__().encode(arg)
+
+def __cut(sentence,tags_limited=False):
+	limit_tags = None
+	if tags_limited:
+		limit_tags = []
+		if len(sentence)==1:
+			limit_tags = ['S']
+		else:
+			limit_tags.append('B')
+			for i in xrange(len(sentence)-2):
+				limit_tags.append('M')
+			limit_tags.append('E')
+	prob, pos_list =  viterbi.viterbi(sentence,char_state_tab, prob_start, prob_trans, prob_emit,limit_tags)
 	begin, next = 0,0
 
 	for i,char in enumerate(sentence):
@@ -29,13 +54,48 @@ def __cut(sentence):
 		if pos=='B':
 			begin = i
 		elif pos=='E':
-			yield sentence[begin:i+1]+"/"+pos_list[i][1]
+			yield pair(sentence[begin:i+1], pos_list[i][1])
 			next = i+1
 		elif pos=='S':
-			yield char+"/"+pos_list[i][1]
+			yield pair(char,pos_list[i][1])
 			next = i+1
 	if next<len(sentence):
-		yield sentence[next:]+"/"+pos_list[next][1]
+		yield pair(sentence[next:], pos_list[next][1] )
+
+def __cut_DAG(sentence):
+	DAG = jieba.get_DAG(sentence)
+	route ={}
+	jieba.calc(sentence,DAG,0,route=route)
+	x = 0
+	buf =u''
+	N = len(sentence)
+	while x<N:
+		y = route[x][1]+1
+		l_word = sentence[x:y]
+		if y-x==1:
+			buf+= l_word
+		else:
+			if len(buf)>0:
+				if len(buf)==1:
+					yield list(__cut(buf))[0]
+					buf=u''
+				else:
+					regognized = __cut(buf)
+					for t in regognized:
+						yield t
+					buf=u''
+			for w in __cut(l_word,tags_limited=True):
+				yield w
+		x =y
+
+	if len(buf)>0:
+		if len(buf)==1:
+			yield list(__cut(buf))[0]
+		else:
+			regognized = __cut(buf)
+			for t in regognized:
+				yield t
+
 
 def cut(sentence):
 	if not ( type(sentence) is unicode):
@@ -48,10 +108,15 @@ def cut(sentence):
 
 	for blk in blocks:
 		if re_han.match(blk):
-				for word in __cut(blk):
+				for word in __cut_DAG(blk):
 					yield word
 		else:
 			tmp = re_skip.split(blk)
 			for x in tmp:
 				if x!="":
-					yield x
+					if re.match(ur"[0-9]+",x):
+						yield pair(x,'m')
+					elif re.match(ur"[a-zA-Z+#]+",x):
+						yield pair(x,'eng')
+					else:
+						yield pair(x,'x')

jieba/posseg/viterbi.py

@@ -5,7 +5,7 @@ def get_top_states(t_state_v,K=4):
 	topK= sorted(items,key=operator.itemgetter(1),reverse=True)[:K]
 	return [x[0] for x in topK]
 
-def viterbi(obs, states, start_p, trans_p, emit_p):
+def viterbi(obs, states, start_p, trans_p, emit_p,limit_tags):
 	V = [{}] #tabular
 	mem_path = [{}]
 	all_states = trans_p.keys()
@@ -15,19 +15,24 @@ def viterbi(obs, states, start_p, trans_p, emit_p):
 	for t in range(1,len(obs)):
 		V.append({})
 		mem_path.append({})
-		prev_states =[ x for x in mem_path[t-1].keys() if len(trans_p[x])>0 ]
-		#print get_top_states(V[t-1])
 		prev_states = get_top_states(V[t-1])
+		prev_states =[ x for x in mem_path[t-1].keys() if len(trans_p[x])>0 ]
+		tmp = prev_states
+		if limit_tags:
+			prev_states = [x for x in prev_states if x[0]==limit_tags[t-1]]
+		if len(prev_states)==0:
+			prev_states = tmp
 		prev_states_expect_next = set( (y  for x in prev_states for y in trans_p[x].keys() ) )
 		obs_states = states.get(obs[t],all_states)
 		obs_states = set(obs_states) &  set(prev_states_expect_next)
+		if limit_tags:
+			obs_states = [x for x in obs_states if x[0]==limit_tags[t]]
 		if len(obs_states)==0: obs_states = all_states
 		for y in obs_states:
 			(prob,state ) = max([(V[t-1][y0] * trans_p[y0].get(y,0) * emit_p[y].get(obs[t],0) ,y0) for y0 in prev_states])
 			V[t][y] =prob
 			mem_path[t][y] = state
 
-
 	last = [(V[-1][y], y) for y in mem_path[-1].keys() ]
 	#if len(last)==0:
 		#print obs

setup.py

@@ -7,5 +7,5 @@ setup(name='jieba',
       url='http://github.com/fxsjy',  
       packages=['jieba'],  
       package_dir={'jieba':'jieba'},
-      package_data={'jieba':['*.*','finalseg/*']}
+      package_data={'jieba':['*.*','finalseg/*','analyse/*','posseg/*']}
 )  

test/test.txt

@@ -0,0 +1,2 @@
+西三旗硅谷先锋小区半地下室出租,便宜可合租硅谷
+工信处女干事每月经过下属科室都要亲口交代24口交换机等技术性器件的安装工作

test/test_file.py

@@ -0,0 +1,20 @@
+import urllib2
+import sys,time
+import sys
+sys.path.append("../")
+import jieba
+
+url = sys.argv[1]
+content = open(url,"rb").read()
+t1 = time.time()
+words = list(jieba.cut(content))
+
+t2 = time.time()
+tm_cost = t2-t1
+
+log_f = open("1.log","wb")
+for w in words:
+	print >> log_f, w.encode("gbk"), "/" ,
+
+print 'speed' , len(content)/tm_cost, " bytes/second"
+

test/test_pos_file.py

@@ -0,0 +1,20 @@
+import urllib2
+import sys,time
+import sys
+sys.path.append("../")
+import jieba.posseg as pseg
+
+url = sys.argv[1]
+content = open(url,"rb").read()
+t1 = time.time()
+words = list(pseg.cut(content))
+
+t2 = time.time()
+tm_cost = t2-t1
+
+log_f = open("1.log","wb")
+for w in words:
+	print >> log_f, w.encode("gbk"), "/" ,
+
+print 'speed' , len(content)/tm_cost, " bytes/second"
+