AmazonML/model.py at main · greenfish8090/AmazonML · GitHub

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
from torch import nn
import torch
from transformers import BertModel, BertTokenizer, RobertaModel, RobertaTokenizer


class Regressor(nn.Module):
    def __init__(self, num_feature):
        super(Regressor, self).__init__()

        self.layer_1 = nn.Sequential(nn.Linear(num_feature, 256))
        self.reg = nn.Linear(256, 1)

        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(p=0.2)

        self.batchnorm = nn.BatchNorm1d(256)

    def forward(self, x):
        x = self.layer_1(x)
        x = self.batchnorm(x)
        x = self.relu(x)
        x = self.dropout(x)

        return self.reg(x)


class Regressor2(nn.Module):
    def __init__(self, num_feature):
        super(Regressor2, self).__init__()

        self.layer_1 = nn.Sequential(nn.Linear(num_feature, 512))
        self.layer_2 = nn.Sequential(nn.Linear(512, 256))
        self.reg = nn.Linear(256, 1)

        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(p=0.2)

        self.batchnorm1 = nn.BatchNorm1d(512)
        self.batchnorm2 = nn.BatchNorm1d(256)

    def forward(self, x):
        x = self.layer_1(x)
        x = self.batchnorm1(x)
        x = self.relu(x)
        x = self.dropout(x)

        x = self.layer_2(x)
        x = self.batchnorm2(x)
        x = self.relu(x)
        x = self.dropout(x)
        return self.reg(x)


class EntityEmbedding(nn.Module):
    def __init__(self, embedding_dim, num_embeddings):
        super(EntityEmbedding, self).__init__()
        self.embedding = nn.Embedding(num_embeddings, embedding_dim)
        self.embedding_dim = embedding_dim
        self.num_embeddings = num_embeddings
        self.regressor = Regressor(self.embedding_dim)

    def forward(self, x):
        e = self.embedding(x)
        return self.regressor(e)


class TransformerRegressor(nn.Module):
    def __init__(self, transformer):
        super().__init__()
        if transformer == "bert-base-uncased":
            self.transformer = BertModel.from_pretrained(transformer)
            self.tokenizer = BertTokenizer.from_pretrained(transformer)
            self.num_feature = 768
        elif transformer == "roberta-base":
            self.transformer = RobertaModel.from_pretrained(transformer)
            self.tokenizer = RobertaTokenizer.from_pretrained(transformer)
            self.num_feature = 768

        self.regressor = Regressor(self.num_feature)

    def forward(self, x, device):
        inp = self.tokenizer(
            x["string"],
            return_tensors="pt",
            padding="max_length",
            truncation=True,
        )
        inp = {k: v.to(device) for k, v in inp.items()}
        output = self.transformer(**inp)
        cls = output[0][:, 0, :]
        x = self.regressor(cls)
        return x


class TransformerEntityRegressorOld(nn.Module):
    def __init__(self, transformer, embedding_dim, num_embeddings):
        super().__init__()
        if transformer == "bert-base-uncased":
            self.transformer = BertModel.from_pretrained(transformer)
            self.tokenizer = BertTokenizer.from_pretrained(transformer)
            self.num_feature = 768
        elif transformer == "roberta-base":
            self.transformer = RobertaModel.from_pretrained(transformer)
            self.tokenizer = RobertaTokenizer.from_pretrained(transformer)
            self.num_feature = 768

        self.embedding = nn.Embedding(num_embeddings, embedding_dim)
        self.embedding_dim = embedding_dim
        self.num_embeddings = num_embeddings

        self.regressor = Regressor(self.num_feature + self.embedding_dim)

    def forward(self, string, type_id, device):
        inp = self.tokenizer(
            string,
            return_tensors="pt",
            padding="max_length",
            truncation=True,
        )
        inp = {k: v.to(device) for k, v in inp.items()}
        output = self.transformer(**inp)
        cls = output[0][:, 0, :]
        x = self.embedding(type_id.to(device))
        x = self.regressor(torch.cat([cls, x], dim=1))
        return x


class TransformerEntityRegressor(nn.Module):
    def __init__(self, transformer, embedding_dim, num_embeddings):
        super().__init__()
        if transformer == "bert-base-uncased":
            self.transformer = BertModel.from_pretrained(transformer)
            self.tokenizer = BertTokenizer.from_pretrained(transformer)
            self.num_feature = 768
        elif transformer == "roberta-base":
            self.transformer = RobertaModel.from_pretrained(transformer)
            self.tokenizer = RobertaTokenizer.from_pretrained(transformer)
            self.num_feature = 768

        self.embedding = nn.Embedding(num_embeddings, embedding_dim)
        self.embedding_dim = embedding_dim
        self.num_embeddings = num_embeddings

        self.regressor = Regressor2(self.num_feature + self.embedding_dim)

    def forward(self, inp, type_id):
        output = self.transformer(**inp)
        cls = output[0][:, 0, :]
        x = self.embedding(type_id)
        x = self.regressor(torch.cat([cls, x], dim=1))
        return x