File size: 8,465 Bytes
bdce880
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import numpy as np
import torch.nn as nn
from timm.models.layers import trunc_normal_
from einops import rearrange, repeat

ACTIVATION = {'gelu': nn.GELU, 'tanh': nn.Tanh, 'sigmoid': nn.Sigmoid, 'relu': nn.ReLU, 'leaky_relu': nn.LeakyReLU(0.1),
              'softplus': nn.Softplus, 'ELU': nn.ELU, 'silu': nn.SiLU}


class Physics_Attention_Irregular_Mesh(nn.Module):
    def __init__(self, dim, heads=8, dim_head=64, dropout=0., slice_num=64):
        super().__init__()
        inner_dim = dim_head * heads
        self.dim_head = dim_head
        self.heads = heads
        self.scale = dim_head ** -0.5
        self.softmax = nn.Softmax(dim=-1)
        self.dropout = nn.Dropout(dropout)
        self.temperature = nn.Parameter(torch.ones([1, heads, 1, 1]) * 0.5)

        self.in_project_x = nn.Linear(dim, inner_dim)
        self.in_project_fx = nn.Linear(dim, inner_dim)
        self.in_project_slice = nn.Linear(dim_head, slice_num)
        for l in [self.in_project_slice]:
            torch.nn.init.orthogonal_(l.weight)  # use a principled initialization
        self.to_q = nn.Linear(dim_head, dim_head, bias=False)
        self.to_k = nn.Linear(dim_head, dim_head, bias=False)
        self.to_v = nn.Linear(dim_head, dim_head, bias=False)
        self.to_out = nn.Sequential(
            nn.Linear(inner_dim, dim),
            nn.Dropout(dropout)
        )

    def forward(self, x):
        # B N C
        B, N, C = x.shape

        ### (1) Slice
        fx_mid = self.in_project_fx(x).reshape(B, N, self.heads, self.dim_head) \
            .permute(0, 2, 1, 3).contiguous()  # B H N C
        x_mid = self.in_project_x(x).reshape(B, N, self.heads, self.dim_head) \
            .permute(0, 2, 1, 3).contiguous()  # B H N C
        slice_weights = self.softmax(self.in_project_slice(x_mid) / self.temperature)  # B H N G
        slice_norm = slice_weights.sum(2)  # B H G
        slice_token = torch.einsum("bhnc,bhng->bhgc", fx_mid, slice_weights)
        slice_token = slice_token / ((slice_norm + 1e-5)[:, :, :, None].repeat(1, 1, 1, self.dim_head))

        ### (2) Attention among slice tokens
        q_slice_token = self.to_q(slice_token)
        k_slice_token = self.to_k(slice_token)
        v_slice_token = self.to_v(slice_token)
        dots = torch.matmul(q_slice_token, k_slice_token.transpose(-1, -2)) * self.scale
        attn = self.softmax(dots)
        attn = self.dropout(attn)
        out_slice_token = torch.matmul(attn, v_slice_token)  # B H G D

        ### (3) Deslice
        out_x = torch.einsum("bhgc,bhng->bhnc", out_slice_token, slice_weights)
        out_x = rearrange(out_x, 'b h n d -> b n (h d)')
        return self.to_out(out_x)


class MLP(nn.Module):
    def __init__(self, n_input, n_hidden, n_output, n_layers=1, act='gelu', res=True):
        super(MLP, self).__init__()

        if act in ACTIVATION.keys():
            act = ACTIVATION[act]
        else:
            raise NotImplementedError
        self.n_input = n_input
        self.n_hidden = n_hidden
        self.n_output = n_output
        self.n_layers = n_layers
        self.res = res
        self.linear_pre = nn.Sequential(nn.Linear(n_input, n_hidden), act())
        self.linear_post = nn.Linear(n_hidden, n_output)
        self.linears = nn.ModuleList([nn.Sequential(nn.Linear(n_hidden, n_hidden), act()) for _ in range(n_layers)])

    def forward(self, x):
        x = self.linear_pre(x)
        for i in range(self.n_layers):
            if self.res:
                x = self.linears[i](x) + x
            else:
                x = self.linears[i](x)
        x = self.linear_post(x)
        return x


class Transolver_block(nn.Module):
    """Transformer encoder block."""

    def __init__(
            self,
            num_heads: int,
            hidden_dim: int,
            dropout: float,
            act='gelu',
            mlp_ratio=4,
            last_layer=False,
            out_dim=1,
            slice_num=32,
    ):
        super().__init__()
        self.last_layer = last_layer
        self.ln_1 = nn.LayerNorm(hidden_dim)
        self.Attn = Physics_Attention_Irregular_Mesh(hidden_dim, heads=num_heads, dim_head=hidden_dim // num_heads,
                                                     dropout=dropout, slice_num=slice_num)
        self.ln_2 = nn.LayerNorm(hidden_dim)
        self.mlp = MLP(hidden_dim, hidden_dim * mlp_ratio, hidden_dim, n_layers=0, res=False, act=act)
        if self.last_layer:
            self.ln_3 = nn.LayerNorm(hidden_dim)
            self.mlp2 = nn.Linear(hidden_dim, out_dim)

    def forward(self, fx):
        fx = self.Attn(self.ln_1(fx)) + fx
        fx = self.mlp(self.ln_2(fx)) + fx
        if self.last_layer:
            return self.mlp2(self.ln_3(fx))
        else:
            return fx


class Model(nn.Module):
    def __init__(self,
                 space_dim=1,
                 n_layers=5,
                 n_hidden=256,
                 dropout=0,
                 n_head=8,
                 act='gelu',
                 mlp_ratio=1,
                 fun_dim=1,
                 out_dim=1,
                 slice_num=32,
                 ref=8,
                 unified_pos=False
                 ):
        super(Model, self).__init__()
        self.__name__ = 'UniPDE_3D'
        self.ref = ref
        self.unified_pos = unified_pos
        if self.unified_pos:
            self.preprocess = MLP(fun_dim + self.ref * self.ref * self.ref, n_hidden * 2, n_hidden, n_layers=0,
                                  res=False, act=act)
        else:
            self.preprocess = MLP(fun_dim + space_dim, n_hidden * 2, n_hidden, n_layers=0, res=False, act=act)

        self.n_hidden = n_hidden
        self.space_dim = space_dim

        self.blocks = nn.ModuleList([Transolver_block(num_heads=n_head, hidden_dim=n_hidden,
                                                      dropout=dropout,
                                                      act=act,
                                                      mlp_ratio=mlp_ratio,
                                                      out_dim=out_dim,
                                                      slice_num=slice_num,
                                                      last_layer=(_ == n_layers - 1))
                                     for _ in range(n_layers)])
        self.initialize_weights()
        self.placeholder = nn.Parameter((1 / (n_hidden)) * torch.rand(n_hidden, dtype=torch.float))

    def initialize_weights(self):
        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=0.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, (nn.LayerNorm, nn.BatchNorm1d)):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)

    def get_grid(self, my_pos):
        # my_pos 1 N 3
        batchsize = my_pos.shape[0]

        gridx = torch.tensor(np.linspace(-1.5, 1.5, self.ref), dtype=torch.float)
        gridx = gridx.reshape(1, self.ref, 1, 1, 1).repeat([batchsize, 1, self.ref, self.ref, 1])
        gridy = torch.tensor(np.linspace(0, 2, self.ref), dtype=torch.float)
        gridy = gridy.reshape(1, 1, self.ref, 1, 1).repeat([batchsize, self.ref, 1, self.ref, 1])
        gridz = torch.tensor(np.linspace(-4, 4, self.ref), dtype=torch.float)
        gridz = gridz.reshape(1, 1, 1, self.ref, 1).repeat([batchsize, self.ref, self.ref, 1, 1])
        grid_ref = torch.cat((gridx, gridy, gridz), dim=-1).cuda().reshape(batchsize, self.ref ** 3, 3)  # B 4 4 4 3

        pos = torch.sqrt(
            torch.sum((my_pos[:, :, None, :] - grid_ref[:, None, :, :]) ** 2,
                      dim=-1)). \
            reshape(batchsize, my_pos.shape[1], self.ref * self.ref * self.ref).contiguous()
        return pos

    def forward(self, data):
        cfd_data, geom_data = data
        x, fx, T = cfd_data.x, None, None
        x = x[None, :, :]
        if self.unified_pos:
            new_pos = self.get_grid(cfd_data.pos[None, :, :])
            x = torch.cat((x, new_pos), dim=-1)

        if fx is not None:
            fx = torch.cat((x, fx), -1)
            fx = self.preprocess(fx)
        else:
            fx = self.preprocess(x)
            fx = fx + self.placeholder[None, None, :]

        for block in self.blocks:
            fx = block(fx)

        return fx[0]