improve speed of tagging

jieba/posseg/__init__.py

@@ -11,9 +11,12 @@ def load_model(f_name):
 	if f_name.endswith(".py"):
 		return eval(open(prob_p_path,"rb").read())
 	else:
-		result = set()
+		result = {}
 		for line in open(prob_p_path,"rb"):
-			result.add(line.strip().decode('utf-8'))
+			line = line.strip()
+			if line=="":continue
+			word, tag = line.split(' ')
+			result[word.decode('utf-8')]=tag
 		return result
 
 
@@ -21,6 +24,7 @@ prob_start = load_model("prob_start.py")
 prob_trans = load_model("prob_trans.py")
 prob_emit = load_model("prob_emit.py")
 char_state_tab = load_model("char_state_tab.py")
+word_tag_tab = load_model("tags.txt")
 
 class pair(object):
 	def __init__(self,word,flag):
@@ -39,18 +43,8 @@ class pair(object):
 	def encode(self,arg):
 		return self.__unicode__().encode(arg)
 
-def __cut(sentence,tags_limited=False):
+def __cut(sentence):
-	limit_tags = None
+	prob, pos_list =  viterbi.viterbi(sentence,char_state_tab, prob_start, prob_trans, prob_emit)
-	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):
@@ -81,20 +75,19 @@ def __cut_DAG(sentence):
 		else:
 			if len(buf)>0:
 				if len(buf)==1:
-					yield list(__cut(buf))[0]
+					yield pair(buf,word_tag_tab.get(buf,'x'))
 					buf=u''
 				else:
 					regognized = __cut(buf)
 					for t in regognized:
 						yield t
 					buf=u''
-			for w in __cut(l_word,tags_limited=True):
+			yield pair(l_word,word_tag_tab.get(l_word,'x'))
-				yield w
 		x =y
 
 	if len(buf)>0:
 		if len(buf)==1:
-			yield list(__cut(buf))[0]
+			yield pair(buf,word_tag_tab.get(buf,'x'))
 		else:
 			regognized = __cut(buf)
 			for t in regognized:
@@ -108,6 +101,7 @@ def cut(sentence):
 		except:
 			sentence = sentence.decode('gbk','ignore')
 	re_han, re_skip = re.compile(ur"([\u4E00-\u9FA5]+)"), re.compile(ur"[^a-zA-Z0-9+#\n%]")
+	re_eng,re_num = re.compile(ur"[a-zA-Z+#]+"), re.compile(ur"[0-9]+")
 	blocks = re_han.split(sentence)
 
 	for blk in blocks:
@@ -118,9 +112,9 @@ def cut(sentence):
 			tmp = re_skip.split(blk)
 			for x in tmp:
 				if x!="":
-					if re.match(ur"[0-9]+",x):
+					if re_num.match(x):
 						yield pair(x,'m')
-					elif re.match(ur"[a-zA-Z+#]+",x):
+					elif re_eng.match(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,limit_tags):
+def viterbi(obs, states, start_p, trans_p, emit_p):
 	V = [{}] #tabular
 	mem_path = [{}]
 	all_states = trans_p.keys()
@@ -17,16 +17,11 @@ def viterbi(obs, states, start_p, trans_p, emit_p,limit_tags):
 		mem_path.append({})
 		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])