PDE-Net/pdelearner.py at master · hsalehipour/PDE-Net · 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
#%%
import numpy as np
from numpy import *
import torch
from torch.autograd import Variable
import aTEAM
import FD

__all__ = ['VariantCoeLinear2d', 'SingleNonLinear2d']

#%%
class VariantCoeLinear2d(torch.nn.Module):
    def __init__(self, kernel_size, max_order, dx, constraint, xy, interp_degree, interp_mesh_size, dt=1e-2):
        super(VariantCoeLinear2d, self).__init__()
        self.id = FD.FD2d(kernel_size, 0, dx=dx, constraint=constraint, boundary='Periodic')
        self.fd2d = FD.FD2d(kernel_size, max_order, dx=dx, constraint=constraint, boundary='Periodic')
        N = self.fd2d.MomentBank.moment.size()[0]
        self._N = N
        for i in range(N):
            fitter = aTEAM.nn.modules.LagrangeInterpFixInputs(xy, interp_dim=2, interp_degree=interp_degree, mesh_bound=[[0,0],[2*pi,2*pi]], mesh_size=interp_mesh_size)
            fitter.double()
            self.add_module('coe'+str(i), fitter)
        self.register_buffer('dt', torch.DoubleTensor(1).fill_(dt))

    @property
    def coes(self):
        for i in range(self._N):
            yield self.__getattr__('coe'+str(i))
    @property
    def xy(self):
        return Variable(next(self.coes).inputs)
    @xy.setter
    def xy(self, v):
        for fitter in self.coes:
            fitter.inputs = v
    def diff_params(self):
        return list(self.id.parameters())+list(self.fd2d.parameters())
    def coe_params(self):
        params = []
        for coe in self.coes:
            params = params+list(coe.parameters())
        return params
    def forward(self, init, stepnum):
        idkernel = self.id.MomentBank.kernel()
        fdkernel = self.fd2d.MomentBank.kernel()
        coe = []
        for fitter in self.coes:
            coe.append(fitter())
        if init.dim() == 2:
            coe = torch.stack(coe,dim=0)[newaxis,...]
            u = init[newaxis,...]
        else:
            assert init.dim() == 3
            coe = torch.stack(coe, dim=1)
            u = init
        dt = Variable(self.dt)
        for i in range(stepnum):
            uid = self.id(u, idkernel).squeeze(1)
            ufd = self.fd2d(u, fdkernel)
            u = uid+dt*(coe*ufd).sum(dim=1)
        return u.view(init.size())
    def step(self, u):
        return self.forward(u, stepnum=1)

class SingleNonLinear2d(torch.nn.Module):
    def __init__(self, kernel_size, max_order, dx, constraint, xy, interp_degree, interp_mesh_size, nonlinear_interp_degree, nonlinear_interp_mesh_bound, nonlinear_interp_mesh_size, dt=1e-2):
        super(SingleNonLinear2d, self).__init__()
        self.id = FD.FD2d(kernel_size, 0, dx=dx, constraint=constraint, boundary='Dirichlet')
        self.fd2d = FD.FD2d(kernel_size, max_order, dx=dx, constraint=constraint, boundary='Dirichlet')
        N = self.fd2d.MomentBank.moment.size()[0]
        self._N = N
        for i in range(1,N):
            fitter = aTEAM.nn.modules.LagrangeInterpFixInputs(xy, interp_dim=2, interp_degree=interp_degree, mesh_bound=[[0,0],[2*pi,2*pi]], mesh_size=interp_mesh_size)
            fitter.double()
            self.add_module('coe'+str(i), fitter)
        self.nonlinearfitter = aTEAM.nn.modules.LagrangeInterp(interp_dim=1, interp_degree=nonlinear_interp_degree, mesh_bound=nonlinear_interp_mesh_bound, mesh_size=nonlinear_interp_mesh_size)
        self.nonlinearfitter.double()
        self.register_buffer('dt', torch.DoubleTensor(1).fill_(dt))

    @property
    def coes(self):
        for i in range(1, self._N):
            yield self.__getattr__('coe'+str(i))
    @property
    def xy(self):
        return Variable(next(self.coes).inputs)
    @xy.setter
    def xy(self, v):
        for fitter in self.coes:
            fitter.inputs = v
    def diff_params(self):
        return list(self.id.parameters())+list(self.fd2d.parameters())
    def coe_params(self):
        params = []
        for coe in self.coes:
            params = params+list(coe.parameters())
        params = params+list(self.nonlinearfitter.parameters())
        return params
    def forward(self, init, stepnum):
        idkernel = self.id.MomentBank.kernel()
        fdkernel = self.fd2d.MomentBank.kernel()
        coe = []
        for fitter in self.coes:
            coe.append(fitter())
        if init.dim() == 2:
            coe = torch.stack(coe,dim=0)[newaxis,...]
            u = init[newaxis,...]
        else:
            assert init.dim() == 3
            coe = torch.stack(coe,dim=1)
            u = init
        dt = Variable(self.dt)
        for i in range(stepnum):
            uid = self.id(u, idkernel).squeeze(1)
            ufd = self.fd2d(u, fdkernel)
            u = uid+dt*((coe*ufd[:,1:]).sum(dim=1)+self.nonlinearfitter(ufd[:,0]))
        return u.view(init.size())
    def step(self, u):
        return self.forward(u, stepnum=1)
#%%
def test():
    import pdedata
    import torchvision
    trans = torchvision.transforms.Compose([pdedata.DownSample(4), pdedata.ToTensor()])
    d = pdedata.variantcoelinear2d(0.6, mesh_size=[200,200], initfreq=4, transform=trans)
    dataloader = torch.utils.data.DataLoader(d, batch_size=2, num_workers=2)
    dataloader = iter(dataloader)
    sample = next(dataloader)
    sample = pdedata.ToVariable()(sample)
    xy = torch.stack([sample['x'],sample['y']],dim=3)
    u0 = sample['u0']
    linpdelearner = VariantCoeLinear2d(kernel_size=[7,7],max_order=4,dx=2*pi/50,constraint='moment',xy=xy,interp_degree=2,interp_mesh_size=[20,20],dt=1e-2)
    linpdelearner.xy = xy.clone()
    ut = linpdelearner(u0,20)

    trans = torchvision.transforms.Compose([pdedata.DownSample(4, boundary='Dirichlet'), pdedata.ToTensor()])
    d = pdedata.singlenonlinear2d(0.6, mesh_size=[200,200], transform=trans)
    dataloader = torch.utils.data.DataLoader(d, batch_size=2, num_workers=2)
    dataloader = iter(dataloader)
    sample = next(dataloader)
    sample = pdedata.ToVariable()(sample)
    xy = torch.stack([sample['x'],sample['y']],dim=3)
    u0 = sample['u0']
    nonlinearlearner = SingleNonLinear2d(kernel_size=[7,7],max_order=2,dx=2*pi/50,constraint='moment',xy=xy,interp_degree=2,interp_mesh_size=[5,5],nonlinear_interp_degree=3,nonlinear_interp_mesh_bound=[-30,30],nonlinear_interp_mesh_size=40,dt=1e-2)
    nonlinearlearner.xy = xy.clone()
    ut = nonlinearlearner(u0,20)


#%%