-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataset2.py
272 lines (236 loc) · 13.4 KB
/
dataset2.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
# coding:utf-8
import math
import torch
from torch.utils import data
from dataset_utils import load_file, load_json_file
import numpy as np
from transformers import BertTokenizer
# bert_tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
class DialogSet(data.Dataset):
def __init__(self, data_path, tokenize_fn, vocab):
data = load_file(data_path)
self.instance = []
for index in range(len(data)):
src, tgt, filter_knowledge = data[index]
src_id = tokenize_fn(vocab, src, 1, 1)
tgt_id = tokenize_fn(vocab, tgt, 1, 1)
filter_knowledge_id = tokenize_fn(vocab, filter_knowledge, 1, 2)
self.instance.append((src_id, tgt_id, filter_knowledge_id, tgt))
def __len__(self):
return len(self.instance)
def __getitem__(self, index):
src_id, tgt_id, filter_knowledge_id, tgt = self.instance[index]
return src_id, tgt_id, filter_knowledge_id, tgt
def cal_max_len(self, ids, curdepth, maxdepth):
"""calculate max sequence length"""
assert curdepth <= maxdepth
if isinstance(ids[0], list):
res = max([self.cal_max_len(k, curdepth + 1, maxdepth) for k in ids])
else:
res = len(ids)
return res
def collate_fn(self, batch):
sv, tv, kv, tstr = zip(*batch) # sv and tv: [batch, seq]; kv: [batch, know_num, know_seq]
sv = [s[1:-1] for s in sv]
kv = [[kk[1:-1] for kk in k] for k in kv]
pad_sv = []
pad_tv = []
pad_kv = []
sv_max_len = max([self.cal_max_len(s, 1, 1) for s in sv])
tv_max_len = max([self.cal_max_len(t, 1, 1) for t in tv])
kn_max_len = max([self.cal_max_len(k, 1, 2) for k in kv])
kn_max_num = max([len(k) for k in kv])
for i in range(len(sv)):
tmp_sv = [0] * sv_max_len
tmp_tv = [0] * tv_max_len
tmp_kv = [[0 for i in range(kn_max_len)] for j in range(kn_max_num)] # [know_num, know_seq]
sv_len = len(sv[i])
tmp_sv[:sv_len] = map(int, sv[i])
tv_len = len(tv[i])
tmp_tv[:tv_len] = map(int, tv[i])
kv_num = len(kv[i])
for j in range(kv_num):
kn_len = len(kv[i][j])
tmp_kv[j][:kn_len] = map(int, kv[i][j])
pad_sv.append(tmp_sv)
pad_tv.append(tmp_tv)
pad_kv.append(tmp_kv) # [batch, know_num, know_seq]
sv_len = [len(s) for s in sv]
tv_len = [len(t) for t in tv]
kv_len = [
[len(kt) for kt in k] + [0 for _ in range(kn_max_num - len(k))] for k in kv
] # [batch, know_num]
return (
np.asarray(pad_sv).reshape((-1, sv_max_len)),
np.asarray(sv_len),
np.asarray(pad_tv).reshape((-1, tv_max_len)),
np.asarray(tv_len),
np.asarray(pad_kv).reshape((-1, kn_max_num, kn_max_len)),
np.asarray(kv_len).reshape((-1, kn_max_num)),
tstr,
)
def GetDataloader(self, batch_size, shuffle, num_workers):
data_loader = data.DataLoader(
self.instance,
batch_size=batch_size,
shuffle=shuffle,
num_workers=num_workers,
collate_fn=self.collate_fn,
)
return data_loader
class AMRDialogSetNew(data.Dataset):
def __init__(self, data_path, tokenize_fn, word_vocab, concept_vocab, relation_vocab, lower=True, use_bert=False):
self.instance = []
self.max_tok_len = 512 if use_bert else 1000
with open(data_path + 'source.txt', 'r', encoding='utf-8') as srcf:
with open(data_path + 'sourceDrop.concept', 'r', encoding='utf-8') as conf:
with open(data_path + 'sourceDrop.path', 'r', encoding='utf-8') as con_relf:
with open(data_path + 'target.txt', 'r', encoding='utf-8') as tgtf:
# with open(data_path + '.mask', 'r', encoding='utf-8') as maskf:
# with open(data_path + '.rel', 'r', encoding='utf-8') as word_relf:
for src_tok, src_concept, con_relation_raw, tgt_tok in zip(srcf, conf, con_relf, tgtf):
src_tok = src_tok.replace(' [SEP]', '')
con_relation_lst_raw = con_relation_raw.strip().split(" ")
seg_len = int(math.sqrt(len(con_relation_lst_raw)))
assert seg_len * seg_len == len(con_relation_lst_raw)
if lower:
con_relation_lst = [
' '.join(con_relation_lst_raw[i: i + seg_len]).lower()
for i in range(0, len(con_relation_lst_raw), seg_len)
]
else:
con_relation_lst = [
' '.join(con_relation_lst_raw[i: i + seg_len])
for i in range(0, len(con_relation_lst_raw), seg_len)
]
# word_mask_lst = [itm.split(' ') for itm in word_rel_mask.split('\t')]
# word_rel_lst = [itm for itm in word_rel.split('\t')]
# print(src_relation_lst)
if lower:
if use_bert:
# print('using bert ...')
src_id = bert_tokenizer.convert_tokens_to_ids(['[CLS]'] + src_tok.strip().lower().split())
else:
src_tok = src_tok.lower().strip()
src_id = tokenize_fn(word_vocab, src_tok, 1, 1, dtype="word")
concept_id = tokenize_fn(concept_vocab, src_concept.strip().lower(), 1, 1, dtype="concept")
tgt_id = tokenize_fn(word_vocab, tgt_tok.strip().lower(), 1, 1, dtype="word")
else:
if use_bert:
src_id = bert_tokenizer.convert_tokens_to_ids(src_tok.strip().lower().split())
else:
src_id = tokenize_fn(word_vocab, src_tok.strip(), 1, 1, dtype="word")
concept_id = tokenize_fn(concept_vocab, src_concept.strip(), 1, 1, dtype="concept")
tgt_id = tokenize_fn(word_vocab, tgt_tok.strip(), 1, 1, dtype="word")
con_rel_id = tokenize_fn(relation_vocab, con_relation_lst, 1, 2, dtype="relation")
# word_mask_id = [[int(iitm) for iitm in itm] for itm in word_mask_lst]
# word_rel_id = tokenize_fn(word_rel_vocab, word_rel_lst, 1, 2, dtype="relation")
# if len(con_rel_id) == len(concept_id) and len(con_rel_id[0]) == len(concept_id):
assert len(con_rel_id) == len(concept_id) and len(con_rel_id[0]) == len(concept_id), "concept_len: {}, relation_size:{} x {}".format(len(concept_id), len(con_rel_id), len(con_rel_id[0]))
# assert len(src_id) - 1 == len(word_mask_id[0]) and len(src_id) - 1 == len(word_rel_id[0]), "word_len: {}, mask_size:{} x {}, rel_size: {} x {}\nword_tok:{}\nword_ids:{}".format(len(src_id), len(word_mask_id), len(word_mask_id[0]), len(word_rel_id), len(word_rel_id[0]), src_tok, ' '.join([str(itm) for itm in src_id]))
# filter_knowledge_id = tokenize_fn(vocab, filter_knowledge, 1, 2)
self.instance.append((src_id, concept_id, con_rel_id, tgt_id, tgt_tok))
def __len__(self):
return len(self.instance)
def __getitem__(self, index):
src_id, concept_id, con_rel_id, tgt_id, tgt_tok, word_mask, word_rel_id = self.instance[index]
return src_id, concept_id, con_rel_id, tgt_id, tgt_tok, word_mask, word_rel_id
def cal_max_len(self, ids, curdepth, maxdepth):
"""calculate max sequence length"""
assert curdepth <= maxdepth
if isinstance(ids[0], list):
res = max([self.cal_max_len(k, curdepth + 1, maxdepth) for k in ids])
else:
res = len(ids)
return res
def collate_fn(self, batch, ):
sv, cv, rv, tv, tstr = zip(*batch) # sv, cv, rv and tv: [batch, seq];
sv = [s[:-1] for s in sv] # remove eos
pad_sv, pad_cv, pad_rv, pad_tv = [], [], [], []
sv_max_len = min(max([self.cal_max_len(s, 1, 1) for s in sv]), self.max_tok_len)
cv_max_len = max([self.cal_max_len(c, 1, 1) for c in cv])
rv_max_len = max([self.cal_max_len(r, 1, 2) for r in rv])
tv_max_len = max([self.cal_max_len(t, 1, 1) for t in tv])
# mask_max_len = min(max([self.cal_max_len(m, 1, 2) for m in mask]), self.max_tok_len)
# wr_max_len = min(max([self.cal_max_len(w, 1, 2) for w in wr]), self.max_tok_len)
# kn_max_len = max([self.cal_max_len(k, 1, 2) for k in kv])
# kn_max_num = max([len(k) for k in kv])
assert rv_max_len == cv_max_len, "Error, rv_max_len should be equal with cv_max_len!!"
# assert sv_max_len == mask_max_len and sv_max_len == wr_max_len, "Error, sv_max_len should be equal with mask_max_len and wr_max_len!!"
# print('sv_max_len', sv_max_len)
for i in range(len(sv)):
tmp_sv = [0] * sv_max_len
tmp_cv = [0] * cv_max_len
tmp_tv = [0] * tv_max_len
tmp_rv = [
[0 for i in range(rv_max_len)] for j in range(rv_max_len)
] # [rv_max_len, rv_max_len]
# tmp_mask = [
# [0 for i in range(mask_max_len)] for j in range(mask_max_len)
# ]
# tmp_wr = [
# [0 for i in range(wr_max_len)] for j in range(wr_max_len)
# ]
ith_sv_len = min(len(sv[i]), self.max_tok_len)
# print('sv: ori_len: {}, sv_len: {}'.format(len(sv[i]), ith_sv_len))
# tmp_sv[:ith_sv_len] = map(int, sv[i][:self.max_tok_len]) #
tmp_sv[:ith_sv_len] = map(int, sv[i][-self.max_tok_len:]) #
ith_cv_len = len(cv[i])
tmp_cv[:ith_cv_len] = map(int, cv[i])
ith_tv_len = len(tv[i])
tmp_tv[:ith_tv_len] = map(int, tv[i])
ith_rv_len = len(rv[i]) # rv_len
for j in range(ith_rv_len):
rv_len_j = len(rv[i][j])
tmp_rv[j][:rv_len_j] = map(int, rv[i][j])
# ith_mask_len = min(len(mask[i]), self.max_tok_len) # rv_len
# for j in range(ith_mask_len):
# mask_len_j = min(len(mask[i][j]), self.max_tok_len)
# # tmp_mask[j][:mask_len_j] = map(int, mask[i][j][:self.max_tok_len])
# tmp_mask[j][:mask_len_j] = map(int, mask[i][j][-self.max_tok_len:])
# ith_wr_len = min(len(wr[i]), self.max_tok_len) # rv_len
# for j in range(ith_wr_len):
# wr_len_j = min(len(wr[i][j]), self.max_tok_len)
# # tmp_wr[j][:wr_len_j] = map(int, wr[i][j][:self.max_tok_len])
# tmp_wr[j][:wr_len_j] = map(int, wr[i][j][-self.max_tok_len:])
# kv_num = len(kv[i])
# for j in range(kv_num):
# kn_len = len(kv[i][j])
# tmp_kv[j][:kn_len] = map(int, kv[i][j])
pad_sv.append(tmp_sv) # [batch, sv_max_len]
pad_cv.append(tmp_cv)
pad_rv.append(tmp_rv) # [batch, len_c, len_c]
pad_tv.append(tmp_tv)
# pad_mask.append(tmp_mask)
# pad_wr.append(tmp_wr)
# pad_kv.append(tmp_kv) # [batch, know_num, know_seq]
sv_len = [min(len(s), self.max_tok_len) for s in sv]
cv_len = [len(c) for c in cv]
tv_len = [len(t) for t in tv]
# kv_len = [
# [len(kt) for kt in k] + [0 for _ in range(kn_max_num - len(k))] for k in kv
# ] # [batch, know_num]
# print(pad_sv)
# tmp_sv_ = np.asarray(pad_sv)
# print('tmp_sv', tmp_sv_.shape)
return (
torch.tensor(pad_sv, dtype=torch.long).view((-1, sv_max_len)),
torch.tensor(sv_len, dtype=torch.long),
torch.tensor(pad_cv, dtype=torch.long).view((-1, cv_max_len)),
torch.tensor(cv_len, dtype=torch.long),
torch.tensor(pad_rv, dtype=torch.long).view((-1, rv_max_len, rv_max_len)),
torch.tensor(pad_tv, dtype=torch.long).view((-1, tv_max_len)),
torch.tensor(tv_len, dtype=torch.long),
# torch.tensor(pad_mask, dtype=torch.long).view((-1, mask_max_len, mask_max_len)),
# torch.tensor(pad_wr, dtype=torch.long).view((-1, wr_max_len, wr_max_len)),
tstr,
)
def GetDataloader(self, batch_size, shuffle, num_workers):
data_loader = data.DataLoader(
self.instance,
batch_size=batch_size,
shuffle=shuffle,
num_workers=num_workers,
collate_fn=self.collate_fn,
)
return data_loader