-
Notifications
You must be signed in to change notification settings - Fork 0
/
parser.py
297 lines (255 loc) · 7.96 KB
/
parser.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
'''
Author: Touhidul Alam
'''
import sys
from collections import deque
from reader import Corpus
import numpy as np
from model import FeatureMapper
from model import Model
import pickle
'''
State class initialize initial stack, buffer, arcs, left-dependency and right-dependency
'''
class State:
def __init__(self, stack, buffer):
self.stack = deque(stack)
self.buffer = deque(buffer)
self.arcs = np.empty(len(self.buffer)+len(self.stack),dtype=np.int32)
self.arcs.fill(-1)
self.ld = np.empty(len(self.buffer)+len(self.stack),dtype=np.int32)
self.rd = np.empty(len(self.buffer)+len(self.stack),dtype=np.int32)
self.ld.fill(-1)
self.rd.fill(-1)
def get_dependency(self, head):
dep_list = []
for i in range(0,len(self.arcs)):
if self.arcs[i]==head:
dep_list.append(i)
return dep_list
'''
Instance class saves the transition label with its corresponding features
'''
class Instance:
def __init__(self, label, featureVector):
self.label = label
self.featureVector = featureVector
'''
Parser class does oracle parsing during training and predicting and making a parser
'''
class Parser:
def __init__(self, state, sentence, feature_map):
self.sentence = sentence
self.state = state
self.map = feature_map
self.data = []
'''Oracle parsing done after extracting feature in each transition '''
def oracle(self):
train_data = []
while self.state.buffer:
feature_list = np.asarray(self.map.feature_template(self.state, self.sentence))
if self.state.stack and self.should_left_arc():
self.left_arc()
instance = Instance(0, feature_list)
elif self.state.stack and self.should_right_arc():
self.right_arc()
instance = Instance(1, feature_list)
elif self.state.stack and self.should_reduce():
self.reduce()
instance = Instance(2, feature_list)
else:
self.shift()
instance = Instance(3, feature_list)
train_data.append(instance)
self.data = train_data
return self.data
'''Parse functions predict each transitions'''
def parse(self, loaded_model):
weights = loaded_model.weights
while self.state.buffer:
feature_list = self.map.feature_template(self.state, self.sentence)
scores = np.zeros((4,))
for index in feature_list:
for i in range(0,4):
scores[i] += weights[i][index]
'''np.argsort, sort array, - means with high-to-low'''
predicted = np.argsort(-scores)
for item in predicted:
if item==0 and self.state.stack and self.can_left_arc():
self.left_arc()
break
elif item==1 and self.state.stack:
self.right_arc()
break
elif item==2 and self.state.stack and self.can_reduce():
self.reduce()
break
elif item==3 and self.state.stack:
self.shift()
break
''' Add to left-neighbour if headless word found, except first element'''
for i in range(0, len(self.state.arcs)):
if i==0:
continue
else:
if self.state.arcs[i]==-1:
self.state.arcs[i] = i-1
return self.state.arcs
def print_transition(self, pred):
print(pred, self.state.stack, self.state.buffer, self.state.arcs)
'''
should_XX() , checks the oracle condition before executing
'''
def should_left_arc(self):
result = False
stack_top = self.state.stack[-1]
buff_front = self.state.buffer[0]
if (buff_front, stack_top) in self.sentence.gold_arcs:
return True
return result
def should_right_arc(self):
result = False
stack_top = self.state.stack[-1]
buff_front = self.state.buffer[0]
if (stack_top, buff_front) in self.sentence.gold_arcs:
return True
return result
def should_reduce(self):
result = False
stack_top = self.state.stack[-1]
buff_front = self.state.buffer[0]
if self.has_head(stack_top) and self.has_all_children(stack_top):
return True
return result
'''
can_XX() checks the precondition of a transition before parsing
'''
def can_left_arc(self):
result = False
stack_top = self.state.stack[-1]
if stack_top!=0 and self.state.arcs[stack_top]==-1:
return True
return result
def can_right_arc(self):
pass
def can_reduce(self):
result = False
stack_top = self.state.stack[-1]
if self.state.arcs[stack_top]!= -1:
return True
return result
def can_shift(self):
result = False
if len(self.state.buffer)>=1 or self.state.stack:
return True
return result
'''hasHead and hasAllChildren condition before checking reduce'''
def has_head(self, stack_top):
count=0
if self.state.arcs[stack_top]!=-1:
count+=1
if count<1:
return False
else:
return True
def has_all_children(self, stack_top):
for gold_arcs in self.sentence.gold_arcs:
head = gold_arcs[0]
dep = gold_arcs[1]
if head==stack_top:
if self.state.arcs[dep] != stack_top:
return False
return True
'''
do_XX() execute the transition
'''
def left_arc(self):
last_stack = self.state.stack.pop()
self.state.arcs[last_stack] = self.state.buffer[0]
self.state.ld[self.state.buffer[0]] = min(self.state.get_dependency(self.state.buffer[0]))
self.state.rd[self.state.buffer[0]] = max(self.state.get_dependency(self.state.buffer[0]))
def right_arc(self):
last_stack = self.state.stack[-1]
self.state.arcs[self.state.buffer[0]] = last_stack
self.state.ld[last_stack] = min(self.state.get_dependency(last_stack))
self.state.rd[last_stack] = max(self.state.get_dependency(last_stack))
self.state.stack.append(self.state.buffer[0])
self.state.buffer.popleft()
def reduce(self):
self.state.stack.pop()
def shift(self):
last_buffer = self.state.buffer.popleft()
self.state.stack.append(last_buffer)
def load_model(language):
f = open('model_'+language, 'rb')
model = pickle.load(f)
f.close()
return model
if __name__ == "__main__":
'''
Arc-eager Transition-based parser with Averaged-perceptron
USAGE: python3 parser.py [train/test/dev] [en/de] dataset_file_location
'''
model_type = sys.argv[1]
language = sys.argv[2]
file_path = sys.argv[3]
init_stack = np.arange(1)
feature_map = FeatureMapper()
'''
TRAIN----
'''
if model_type=='train':
sentences = Corpus(file_path, model_type, language).sentences
train_data = []
for sentence in sentences:
init_buffer = np.arange(1, len(sentence.form))
state = State(init_stack, init_buffer)
parser_data = Parser(state, sentence, feature_map).oracle()
train_data.append(parser_data)
train_data = np.concatenate(train_data).flatten().tolist()
feature_map.frozen = True
weights = np.zeros((4, feature_map.id), dtype=np.float32)
print(weights.shape)
model = Model(feature_map, weights)
model.train(train_data)
model.save_model(model, language)
'''
DEV----
'''
if model_type=='dev':
sentences = Corpus(file_path, model_type, language).sentences
loaded_model = load_model(language)
i=0
j=0
for sentence in sentences:
init_buffer = np.arange(1, len(sentence.form))
state = State(init_stack, init_buffer)
arcs = Parser(state, sentence, loaded_model.map).parse(loaded_model)
'''checking each word wise arc comparisong, first word is root, which is ommitted'''
for gold_arc, cur_arc in zip(sentence.head[1:], arcs[1:]):
if cur_arc==gold_arc:
j+=1
i+=1
print("Dev Accuracy: ",str(j/i))
'''
TEST----
'''
if model_type=='test':
sentences = Corpus(file_path, model_type, language).sentences
loaded_model = load_model(language)
outfile = open('pred_'+language+'.conll06','w')
i=0
j=0
for sentence in sentences:
init_buffer = np.arange(1, len(sentence.form))
state = State(init_stack, init_buffer)
arcs = Parser(state, sentence, loaded_model.map).parse(loaded_model)
for arc in range(1,len(arcs)):
''' In german, we have extra morph line to be added'''
if language=='en':
outfile.write(str(arc)+"\t"+sentence.form[arc]+"\t"+sentence.lemma[arc]+"\t"+sentence.pos[arc]+"\t"+"_"+"\t"+"_"+"\t"+str(arcs[arc])+"\t"+"_"+"\t"+"_"+"\t"+"_")
else:
outfile.write(str(arc)+"\t"+sentence.form[arc]+"\t"+sentence.lemma[arc]+"\t"+sentence.pos[arc]+"\t"+"_"+"\t"+sentence.morph[arc]+"\t"+str(arcs[arc])+"\t"+"_"+"\t"+"_"+"\t"+"_")
outfile.write('\n')
outfile.write('\n')
outfile.close()