-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodel.py
298 lines (229 loc) · 8.67 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
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
298
from typing import Callable, Optional, Union
import torch
import math
from torch import Tensor
import torch.nn as nn
from torch.nn import functional as F
from rot_emb import RotaryEmbedding
class PositionalEncoding(nn.Module):
"""
Implement the PE function.
"""
def __init__(self, d_model, max_seq_len=50):
super().__init__()
self.d_model = d_model
# create constant 'pe' matrix with values dependant on
# pos and i
pe = torch.zeros(max_seq_len, d_model)
for pos in range(max_seq_len):
for i in range(0, d_model, 2):
pe[pos, i] = math.sin(pos / (10000**((2 * i) / d_model)))
pe[pos, i + 1] = math.cos(pos / (10000**((2 * (i + 1)) / d_model)))
pe = pe.unsqueeze(0)
self.register_buffer('pe', pe)
def forward(self, x):
# make embeddings relatively larger
x = x * math.sqrt(self.d_model)
#add constant to embedding
seq_len = x.size(1)
x = x + torch.autograd.Variable(self.pe[:, :seq_len], requires_grad=False)
return x
class StrokeNet(nn.Module):
def __init__(self,
user_cnt,
feat_cnt,
key_cnt,
key_emb_size,
dim_ff,
num_heads,
num_layers,
dropout,
causal_att,
use_user_emb) -> None:
super().__init__()
self.positional_encoding = PositionalEncoding(key_emb_size * 3, 51)
self.keycode_embedding = nn.Embedding(key_cnt, key_emb_size)
self.user_embedding = nn.Embedding(user_cnt, key_emb_size * 3)
self.feat_cnt = feat_cnt
self.hidden_dim = key_emb_size
self.dim_ff = dim_ff
self.num_heads = num_heads
self.num_layers = num_layers
self.dropout = dropout
self.causal_att = causal_att
self.use_user_emb = use_user_emb
self.feat_proj = nn.Linear(feat_cnt, key_emb_size, bias=True)
self.feat_bn = nn.BatchNorm1d(key_emb_size)
self.input_mlp = nn.Sequential(
nn.Linear(key_emb_size * 3, key_emb_size * 3),
nn.ReLU(),
nn.Linear(key_emb_size * 3, key_emb_size * 3),
nn.ReLU(),
)
self.trf_cross = nn.TransformerEncoder(nn.TransformerEncoderLayer(
d_model=key_emb_size * 3,
dim_feedforward=dim_ff,
nhead=self.num_heads,
dropout=self.dropout,
batch_first=True,
norm_first=True),
num_layers=self.num_layers)
self.electra_lin = nn.Linear(key_emb_size * 3, 2, bias=False)
if self.use_user_emb:
self.user_lin = nn.Linear(key_emb_size * 3, 2, bias=False)
def forward(self, b0, b1, feat, mask, user, attn_mask=None):
feat = self.feat_proj(feat)
feat = feat.transpose(1, 2)
feat = self.feat_bn(feat)
feat = feat.transpose(1, 2)
b0_emb = self.keycode_embedding(b0)
b1_emb = self.keycode_embedding(b1)
user_emb = self.user_embedding(user)
x = torch.cat([b0_emb, b1_emb, feat], dim=-1)
# append user embedding to the beginning of the sequence
x = torch.cat([user_emb.unsqueeze(1), x], dim=1)
# MLP
x = self.input_mlp(x)
# add positional encoding - including user embedding
x = self.positional_encoding(x)
if self.causal_att:
x = self.trf_cross(src=x, mask=attn_mask, src_key_padding_mask=mask, is_causal=True)
else:
x = self.trf_cross(src=x, src_key_padding_mask=mask, is_causal=False)
# remove user embedding
user_out = x[:, 0]
x = x[:, 1:]
x = self.electra_lin(x)
if self.use_user_emb:
user_out = self.user_lin(user_out)
return x, user_out
else:
return x, None
class MultiHeadRot(nn.Module):
def __init__(self, d_model, nhead, dropout=0.0, bias=False):
super().__init__()
self.rotary_emb = RotaryEmbedding(dim=d_model // nhead)
self.linear_qkv = nn.Linear(d_model, 3 * d_model, bias=bias)
self.linear_out = nn.Linear(d_model, d_model, bias=bias)
self.nhead = nhead
self.d_model = d_model
self.dropout = dropout
self.active_dropout = 0.0
def forward(self, x, src_mask, src_padding, is_causal):
bsz = x.size(0)
q, k, v = torch.split(self.linear_qkv(x), self.d_model, dim=-1)
head_dim = self.d_model // self.nhead
# split heads
q = q.view(bsz, -1, self.nhead, head_dim).transpose(1, 2)
k = k.view(bsz, -1, self.nhead, head_dim).transpose(1, 2)
v = v.view(bsz, -1, self.nhead, head_dim).transpose(1, 2)
# apply rotary embedding
q = self.rotary_emb.rotate_queries_or_keys(q)
k = self.rotary_emb.rotate_queries_or_keys(k)
if is_causal:
dot_prod = F.scaled_dot_product_attention(q, k, v, dropout_p=self.active_dropout, is_causal=is_causal)
else:
dot_prod = F.scaled_dot_product_attention(q, k, v, attn_mask=src_mask, dropout_p=self.active_dropout, is_causal=is_causal)
# concat heads
dot_prod = dot_prod.transpose(1, 2).contiguous().view(bsz, -1, self.d_model)
return self.linear_out(dot_prod)
def train(self, mode=True):
super().train(mode)
if mode:
self.active_dropout = self.dropout
else:
self.active_dropout = 0.0
class TransformerEncoderRotLayer(nn.Module):
def __init__(self,
d_model,
nhead,
dim_feedforward=756,
dropout=0.1):
super().__init__()
self.multihead_rot = MultiHeadRot(d_model=d_model, nhead=nhead, dropout=dropout)
self.linear1 = nn.Linear(d_model, dim_feedforward)
self.dropout = nn.Dropout(dropout)
self.linear2 = nn.Linear(dim_feedforward, d_model)
self.norm1 = nn.LayerNorm(d_model)
self.norm2 = nn.LayerNorm(d_model)
self.dropout1 = nn.Dropout(dropout)
self.dropout2 = nn.Dropout(dropout)
self.activation = nn.SiLU()
def forward(self, x, src_mask, src_key_padding_mask, is_causal):
x = x + self.dropout1(self.multihead_rot(self.norm1(x), src_mask, src_key_padding_mask, is_causal=is_causal))
x = x + self.dropout2(self.linear2(self.dropout(self.activation(self.linear1(self.norm2(x))))))
return x
class TransformerEncoderRot(nn.Module):
def __init__(self,
trf_layer,
num_layers):
super().__init__()
self.layers = nn.ModuleList([trf_layer for _ in range(num_layers)])
def forward(self, x, src_mask, src_key_padding_mask, is_causal):
for layer in self.layers:
x = layer(x, src_mask, src_key_padding_mask, is_causal)
return x
class StrokeNetRot(nn.Module):
def __init__(self,
user_cnt,
feat_cnt,
key_cnt,
key_emb_size,
dim_ff,
num_heads,
num_layers,
dropout,
causal_att,
use_user_emb) -> None:
super().__init__()
self.keycode_embedding = nn.Embedding(key_cnt, key_emb_size)
self.user_embedding = nn.Embedding(user_cnt, key_emb_size * 3)
self.feat_cnt = feat_cnt
self.hidden_dim = key_emb_size
self.dim_ff = dim_ff
self.num_heads = num_heads
self.num_layers = num_layers
self.dropout = dropout
self.causal_att = causal_att
self.use_user_emb = use_user_emb
self.feat_proj = nn.Linear(feat_cnt, key_emb_size, bias=True)
self.feat_bn = nn.BatchNorm1d(key_emb_size)
# self.input_rff = GaussianFourierFeatureTransform(num_input_feats=4, mapping_size=key_emb_size // 2, scale=2)
self.input_mlp = nn.Sequential(
nn.Linear(key_emb_size * 3, key_emb_size * 3),
nn.SiLU(),
nn.Linear(key_emb_size * 3, key_emb_size * 3),
nn.SiLU(),
)
self.trf_cross = TransformerEncoderRot(
TransformerEncoderRotLayer(d_model=key_emb_size * 3, nhead=self.num_heads,
dim_feedforward=dim_ff, dropout=self.dropout),
num_layers=self.num_layers
)
self.electra_lin = nn.Linear(key_emb_size * 3, 2, bias=False)
if self.use_user_emb:
self.user_lin = nn.Linear(key_emb_size * 3, 2, bias=False)
def forward(self, b0, b1, feat, mask, user, attn_mask=None):
feat = self.feat_proj(feat)
# feat = self.input_rff(feat)
feat = feat.transpose(1, 2)
feat = self.feat_bn(feat)
feat = feat.transpose(1, 2)
b0_emb = self.keycode_embedding(b0)
b1_emb = self.keycode_embedding(b1)
user_emb = self.user_embedding(user)
x = torch.cat([b0_emb, b1_emb, feat], dim=-1)
# append user embedding to the beginning of the sequence
x = torch.cat([user_emb.unsqueeze(1), x], dim=1)
# MLP
x = self.input_mlp(x)
x = self.trf_cross(x, src_mask=attn_mask, src_key_padding_mask=mask, is_causal=self.causal_att)
# remove user embedding
user_out = x[:, 0]
x = x[:, 1:]
x = self.electra_lin(x)
if self.use_user_emb:
user_out = self.user_lin(user_out)
return x, user_out
else:
return x, None