-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
167 lines (142 loc) · 7.67 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
# !/usr/bin/env python3
"""
==== No Bugs in code, just some Random Unexpected FEATURES ====
┌─────────────────────────────────────────────────────────────┐
│┌───┬───┬───┬───┬───┬───┬───┬───┬───┬───┬───┬───┬───┬───┬───┐│
││Esc│!1 │@2 │#3 │$4 │%5 │^6 │&7 │*8 │(9 │)0 │_- │+= │|\ │`~ ││
│├───┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴───┤│
││ Tab │ Q │ W │ E │ R │ T │ Y │ U │ I │ O │ P │{[ │}] │ BS ││
│├─────┴┬──┴┬──┴┬──┴┬──┴┬──┴┬──┴┬──┴┬──┴┬──┴┬──┴┬──┴┬──┴─────┤│
││ Ctrl │ A │ S │ D │ F │ G │ H │ J │ K │ L │: ;│" '│ Enter ││
│├──────┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴─┬─┴────┬───┤│
││ Shift │ Z │ X │ C │ V │ B │ N │ M │< ,│> .│? /│Shift │Fn ││
│└─────┬──┴┬──┴──┬┴───┴───┴───┴───┴───┴──┬┴───┴┬──┴┬─────┴───┘│
│ │Fn │ Alt │ Space │ Alt │Win│ HHKB │
│ └───┴─────┴───────────────────────┴─────┴───┘ │
└─────────────────────────────────────────────────────────────┘
UIE torch版本实现，包含模型预处理/后处理函数。
Author: pankeyu
Date: 2022/10/18
"""
import json
from typing import List
import torch
import torch.nn as nn
import numpy as np
class UIE(nn.Module):
def __init__(self, encoder):
"""
init func.
Args:
encoder (transformers.AutoModel): backbone, 默认使用 ernie 3.0
Reference:
https://github.com/PaddlePaddle/PaddleNLP/blob/a12481fc3039fb45ea2dfac3ea43365a07fc4921/model_zoo/uie/model.py
"""
super().__init__()
self.encoder = encoder
hidden_size = 768
self.linear_start = nn.Linear(hidden_size, 1)
self.linear_end = nn.Linear(hidden_size, 1)
self.sigmoid = nn.Sigmoid()
def forward(
self,
input_ids: torch.tensor,
token_type_ids: torch.tensor,
attention_mask=None,
pos_ids=None,
) -> tuple:
"""
forward 函数，返回开始/结束概率向量。
Args:
input_ids (torch.tensor): (batch, seq_len)
token_type_ids (torch.tensor): (batch, seq_len)
attention_mask (torch.tensor): (batch, seq_len)
pos_ids (torch.tensor): (batch, seq_len)
Returns:
tuple: start_prob -> (batch, seq_len)
end_prob -> (batch, seq_len)
"""
sequence_output = self.encoder(
input_ids=input_ids,
token_type_ids=token_type_ids,
position_ids=pos_ids,
attention_mask=attention_mask,
)["last_hidden_state"]
start_logits = self.linear_start(sequence_output) # (batch, seq_len, 1)
start_logits = torch.squeeze(start_logits, -1) # (batch, seq_len)
start_prob = self.sigmoid(start_logits) # (batch, seq_len)
end_logits = self.linear_end(sequence_output) # (batch, seq_len, 1)
end_logits = torch.squeeze(end_logits, -1) # (batch, seq_len)
end_prob = self.sigmoid(end_logits) # (batch, seq_len)
return start_prob, end_prob
class PositionalEncoding(nn.Module):
def __init__(self, d_model,max_seq_len=5000,dropout=0.1,batch_first=False) -> None:
'''不知道为什么，d_model必须是偶数'''
super().__init__()
self.x_dim=1 if batch_first else 0
self.batch_first=batch_first
pos=torch.arange(0,max_seq_len).unsqueeze(1)
div_term=torch.pow(10000,torch.arange(0,d_model,2)/d_model)
if self.batch_first:
pe=torch.zeros(1,max_seq_len,d_model)
pe[0,:,0::2]=torch.sin(pos/div_term)
pe[0,:,1::2]=torch.cos(pos/div_term)
else:
pe=torch.zeros(max_seq_len,1,d_model)
pe[:,0,0::2]=torch.sin(pos/div_term)
pe[:,0,1::2]=torch.cos(pos/div_term)
self.register_buffer('pe', pe)
self.dropout=nn.Dropout(dropout)
def forward(self,x):
if self.batch_first:
x=x+self.pe[:,:x.shape[self.x_dim]]
else:
x=x+self.pe[:x.shape[self.x_dim]]
return self.dropout(x)
class TimeSeriesTransformer(nn.Module):
def __init__(self,nvars,d_model,d_hid,nheads,nlayers,dropout=0.1) -> None:
'''把d_modal与变量个数做映射
'''
super().__init__()
self.model_name='Transformer Encoder'
# self.max_seq_len=train_win
# encoder
self.encoder_input_mapping=nn.Linear(nvars,d_model)
self.pos_encoding=PositionalEncoding(d_model=d_model,batch_first=True)
encoder_layer=nn.TransformerEncoderLayer(d_model,nheads,d_hid,dropout,batch_first=True)
self.transformer_encoder=nn.TransformerEncoder(encoder_layer,num_layers=nlayers,norm=None)
#norm参数表示每个encoder_layer所需的normalization方法，但是nn.TransformerEncoderLayer已经包含了Layer-normalization，故不需要向norm传递参数，事实上，norm参数是为不具备normalization方法的custorm encoder-layers准备的。
# decoder
self.decoder_input_mapping=nn.Linear(nvars,d_model)
decoder_layer=nn.TransformerDecoderLayer(d_model=d_model,nhead=nheads,dim_feedforward=d_hid,dropout=dropout,batch_first=True)
self.transformer_decoder=nn.TransformerDecoder(decoder_layer,num_layers=nlayers,norm=None)
self.decoder_output_mapping=nn.Linear(d_model,nvars)
# 初始化参数
self.init_weight()
def init_weight(self):
init_range=0.1
self.encoder_input_mapping.weight.data.uniform_(-init_range,init_range)
self.encoder_input_mapping.bias.data.zero_()
self.decoder_input_mapping.weight.data.uniform_(-init_range,init_range)
self.decoder_input_mapping.bias.data.zero_()
self.decoder_output_mapping.weight.data.uniform_(-init_range,init_range)
self.decoder_output_mapping.bias.data.zero_()
def forward(self,src,tgt,memory_mask=None,tgt_mask=None):
'''
Return a tensor of shape:
[batch_size,tgt_win,nvars]
Args:
src:[batch_size,train_win,nvars]
tgt:[batch_size,tgt_win,nvars]
memory_mask:[train_win,train_win]
tgt_mask:[tgt_win,tgt_win]
'''
# encoder
src=self.encoder_input_mapping(src)
src=self.pos_encoding(src)
encoder_output=self.transformer_encoder(src)
# decoder
tgt=self.decoder_input_mapping(tgt)
decoder_output=self.transformer_decoder(tgt=tgt,memory=encoder_output,tgt_mask=tgt_mask,memory_mask=memory_mask)#memory：最后一层encoder layer的输出
decoder_output=self.decoder_output_mapping(decoder_output)
return decoder_output