-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
219 lines (193 loc) · 9.25 KB
/
model.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
import numpy as np
import pickle
import random
'''
FeatureMapper takes the unique feature and add it to the feature dictionary
'''
class FeatureMapper():
def __init__(self):
self.feature_map = {}
self.id = 1
self.frozen = False
def feature_template(self, state, sentence):
feature_list = np.empty((0), int)
stack = state.stack
buffer = state.buffer
ld = state.ld
rd = state.rd
form = sentence.form
pos = sentence.pos
lemma = sentence.lemma
morph = sentence.morph
head = state.arcs
if stack:
s0 = stack[-1]
''' For stack[0] element'''
feature_list = np.append(feature_list, self.get_feature("s0_form="+form[s0]))
feature_list = np.append(feature_list, self.get_feature("s0_pos="+pos[s0]))
feature_list = np.append(feature_list, self.get_feature("s0_lemma="+lemma[s0]))
feature_list = np.append(feature_list, self.get_feature("s0_form_pos="+form[s0]+pos[s0]))
feature_list = np.append(feature_list, self.get_feature("s0_lemma_pos="+lemma[s0]+pos[s0]))
'''For arc-eager, taking head, ld, rd of stack[0] elements'''
if head[s0]>=0:
feature_list = np.append(feature_list, self.get_feature("head_s0_form="+form[head[s0]]))
feature_list = np.append(feature_list, self.get_feature("head_s0_pos="+pos[head[s0]]))
if ld[s0]>=0:
feature_list = np.append(feature_list, self.get_feature("ld_s0_form="+form[ld[s0]]))
feature_list = np.append(feature_list, self.get_feature("ld_s0_pos="+pos[ld[s0]]))
if rd[s0]>=0:
feature_list = np.append(feature_list, self.get_feature("rd_s0_form="+form[rd[s0]]))
feature_list = np.append(feature_list, self.get_feature("rd_s0_pos="+pos[rd[s0]]))
if len(stack)>1:
''' For stack[1] element '''
s1 = stack[-2]
feature_list = np.append(feature_list, self.get_feature("s1_form="+form[s1]))
feature_list = np.append(feature_list, self.get_feature("s1_pos="+pos[s1]))
feature_list = np.append(feature_list, self.get_feature("s1_form_pos="+form[s1]+pos[s1]))
if morph:
''' For German language only Morphological feature of s[0]'''
feature_list = np.append(feature_list, self.get_feature("s0_morph="+morph[s0]))
if buffer:
b0 = buffer[0]
''' For buffer[0] element '''
feature_list = np.append(feature_list, self.get_feature("b0_form="+form[b0]))
feature_list = np.append(feature_list, self.get_feature("b0_pos="+pos[b0]))
feature_list = np.append(feature_list, self.get_feature("b0_lemma="+lemma[b0]))
feature_list = np.append(feature_list, self.get_feature("b0_form_pos="+form[b0]+pos[b0]))
feature_list = np.append(feature_list, self.get_feature("b0_lemma_pos="+lemma[b0]+pos[b0]))
''' head, ld and rd for buffer[0] element '''
if head[b0]>=0:
feature_list = np.append(feature_list, self.get_feature("head_b0_form="+form[head[b0]]))
feature_list = np.append(feature_list, self.get_feature("head_b0_pos="+pos[head[b0]]))
if ld[b0]>=0:
feature_list = np.append(feature_list, self.get_feature("ld_b0_form="+form[ld[b0]]))
feature_list = np.append(feature_list, self.get_feature("ld_b0_pos="+pos[ld[b0]]))
if rd[b0]>=0:
feature_list = np.append(feature_list, self.get_feature("rd_b0_form="+form[rd[b0]]))
feature_list = np.append(feature_list, self.get_feature("rd_b0_pos="+pos[rd[b0]]))
if len(buffer)>1:
b1 = buffer[1]
''' buffer[1] features '''
feature_list = np.append(feature_list, self.get_feature("b1_form="+form[b1]))
feature_list = np.append(feature_list, self.get_feature("b1_pos="+pos[b1]))
feature_list = np.append(feature_list, self.get_feature("b1_form_pos="+form[b1]+pos[b1]))
feature_list = np.append(feature_list, self.get_feature("b0_form+b1_form="+form[b0]+form[b1]))
feature_list = np.append(feature_list, self.get_feature("b0_pos+b1_pos="+pos[b0]+pos[b1]))
if stack:
feature_list = np.append(feature_list, self.get_feature("b0_pos+b1_pos+s0_pos="+pos[b0]+pos[b1]+pos[stack[-1]]))
if len(buffer)>2:
''' buffer[2] features '''
b2 = buffer[2]
feature_list = np.append(feature_list, self.get_feature("b2_pos="+pos[b2]))
feature_list = np.append(feature_list, self.get_feature("b2_form="+form[b2]))
feature_list = np.append(feature_list, self.get_feature("b2_form_pos="+form[b2]+pos[b2]))
feature_list = np.append(feature_list, self.get_feature("b0_pos+b1_pos+b2_pos="+pos[b0]+pos[b1]+pos[b2]))
if len(buffer)>3:
''' buffer[3] POS only '''
b3 = buffer[3]
feature_list = np.append(feature_list, self.get_feature("b3_pos="+pos[b3]))
if morph:
''' Morphological feature for buffer[0] for German language '''
feature_list = np.append(feature_list, self.get_feature("b0_morph="+morph[b0]))
feature_list = np.append(feature_list, self.get_feature("b0_pos_morph="+pos[b0]+morph[b0]))
if stack and buffer:
s0 = stack[-1]
b0 = buffer[0]
''' Both stack and buffer '''
feature_list = np.append(feature_list, self.get_feature("s0_form_pos+b0_form_pos="+form[s0]+pos[s0]+form[b0]+pos[b0]))
feature_list = np.append(feature_list, self.get_feature("s0_form_pos+b0_form="+form[s0]+pos[s0]+form[b0]))
feature_list = np.append(feature_list, self.get_feature("s0_form+b0_form_pos="+form[s0]+form[b0]+pos[b0]))
feature_list = np.append(feature_list, self.get_feature("s0_form_pos+b0_pos="+form[s0]+pos[s0]+pos[b0]))
feature_list = np.append(feature_list, self.get_feature("s0_pos+b0_form_pos="+pos[s0]+form[b0]+pos[b0]))
feature_list = np.append(feature_list, self.get_feature("s0_form+b0_form="+form[s0]+form[b0]))
feature_list = np.append(feature_list, self.get_feature("s0_pos+b0_pos="+pos[s0]+pos[b0]))
feature_list = np.append(feature_list, self.get_feature("s0_lemma+b0_lemma="+lemma[s0]+lemma[b0]))
if morph:
feature_list = np.append(feature_list, self.get_feature("s0_b0_morph="+morph[s0]+morph[b0]))
feature_list = np.append(feature_list, self.get_feature("s0_b0_pos_morph="+pos[s0]+pos[b0]+morph[s0]+morph[b0]))
''' Distance between stack[0] and buffer[0] '''
distance = str(b0 - s0)
if head[s0]>=0:
feature_list = np.append(feature_list, self.get_feature("s0_pos+b0_pos+hd_s0_pos="+pos[s0]+pos[b0]+pos[head[s0]]))
if ld[b0]>=0:
feature_list = np.append(feature_list, self.get_feature("s0_pos+b0_pos+ld_s0_pos="+pos[s0]+pos[b0]+pos[ld[s0]]))
feature_list = np.append(feature_list, self.get_feature("s0_pos+b0_pos+ld_b0_pos="+pos[s0]+pos[b0]+pos[ld[b0]]))
if rd[b0]>=0:
feature_list = np.append(feature_list, self.get_feature("s0_pos+b0_pos+rd_s0_pos="+pos[s0]+pos[b0]+pos[rd[s0]]))
feature_list = np.append(feature_list, self.get_feature("s0_pos+b0_pos+rd_b0_pos="+pos[s0]+pos[b0]+pos[rd[b0]]))
''' combined distance of s0 and b0 features'''
feature_list = np.append(feature_list, self.get_feature("s0_lemma+b0_lemma+dist="+lemma[s0]+lemma[b0]+distance))
feature_list = np.append(feature_list, self.get_feature("s0_form+dist="+form[s0]+distance))
feature_list = np.append(feature_list, self.get_feature("s0_pos+dist="+pos[s0]+distance))
feature_list = np.append(feature_list, self.get_feature("b0_form+dist="+form[b0]+distance))
feature_list = np.append(feature_list, self.get_feature("b0_pos+dist="+pos[b0]+distance))
feature_list = np.append(feature_list, self.get_feature("s0_form+b0_form+dist="+form[s0]+form[b0]+distance))
feature_list = np.append(feature_list, self.get_feature("s0_pos+b0_pos+dist="+pos[s0]+pos[b0]+distance))
return feature_list
def get_feature(self, feature):
if self.frozen:
if feature not in self.feature_map:
return 0
else:
return self.feature_map[feature]
else:
if feature not in self.feature_map:
self.feature_map[feature] = self.id
self.id += 1
return self.feature_map[feature]
'''
Model saves the FeatureMapper and weights
'''
class Model:
def __init__(self, feature_map, weights):
pass
self.map = feature_map
self.weights = weights
def save_model(self, model, language):
f = open('model_'+language, 'wb')
pickle.dump(model, f, -1)
f.close()
''' this function runs an averaged perceptron '''
def train(self, train_data):
print("In trainer...")
u= np.zeros(self.weights.shape, dtype=np.float32)
q=0
for epoch in range(0,15):
correct=0
print("epoch: ",epoch+1)
j=0
#Shuffling between each epochs
random.shuffle(train_data, random.random)
for data in train_data:
q += 1
j += 1
scores = np.zeros((4,))
feature_vector = np.ones((len(data.featureVector),), dtype = np.float32)
for index in data.featureVector:
for i in range(0,4):
scores[i] += self.weights[i][index]
predicted = np.argmax(scores)
if predicted!= data.label:
for index in data.featureVector:
self.weights[data.label][index]+=1
self.weights[predicted][index]-=1
u[data.label][index]+=q
u[predicted][index]-=q
if predicted==data.label:
correct+=1
if j%5000==0:
print("States",j,": ", (correct/j))
print("Accuracy: ", (correct/len(train_data)))
self.weights -= u * (1/q)
#3epochs - Average - English
#Accuracy: 0.9722499306538014
#Testing Accuracy: 0.8738860688099267
#3epochs - Normal Perceptron - English
#Accuracy: 0.9709464539174227
#Testing Accuracy: 0.8332393307012597
#15epochs - Average - English
#Accuracy: 0.9923746224592055
#Testing Accuracy: 0.8739988719684151
#15epochs - Average - German
#Accuracy: 0.9964475086687264
#Testing Accuracy: 0.8832525031289111