-
Notifications
You must be signed in to change notification settings - Fork 26
/
data_utils.py
126 lines (114 loc) · 4.83 KB
/
data_utils.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
import torch
from torch.utils.data.dataset import Dataset
import numpy as np
import sys
sys.path.insert(0,'/mnt/home/siyuh/Project/Recon/analysis')
#from genwaves import genwaves
class SimuData(Dataset):
def __init__(self,base_path,lIndex,hIndex,aug):
self.datafiles = []
for x in np.arange(lIndex,hIndex,1000):
y = [base_path+str(x)+'_'+str(i)+'.npy' for i in range(1000)]
self.datafiles+=y
self.aug=aug
def __getitem__(self, index):
return get_mini_batch(self.datafiles[index],self.aug)
def __len__(self):
return len(self.datafiles)
def swap(x,index1,index2):
temp = x[index1].copy()
x[index1] = x[index2]
x[index2] = temp
return x
def get_mini_batch(fname,aug):
x = np.load(fname)
x = np.einsum('ijkl->lijk', x)
LPT = x[1:4]
Nbody = x[7::]
if(aug==1):
if np.random.rand() < .5:
LPT = LPT[:,::-1,...]
LPT[0] = -LPT[0]
Nbody = Nbody[:,::-1,...]
Nbody[0] = -Nbody[0]
if np.random.rand() < .5:
LPT = LPT[:,:,::-1,...]
LPT[1] = -LPT[1]
Nbody = Nbody[:,:,::-1,...]
Nbody[1] = -Nbody[1]
if np.random.rand() < .5:
LPT = LPT[:,:,:,::-1]
LPT[2] = -LPT[2]
Nbody = Nbody[:,:,:,::-1]
Nbody[2] = -Nbody[2]
prand = np.random.rand()
if prand < 1./6:
LPT = np.transpose(LPT, axes = (0,2,3,1))
LPT = swap(LPT,0,2)
LPT = swap(LPT,0,1)
Nbody = np.transpose(Nbody, axes = (0,2,3,1))
Nbody = swap(Nbody,0,2)
Nbody = swap(Nbody,0,1)
elif prand < 2./6:
LPT = np.transpose(LPT, axes = (0,2,1,3))
LPT = swap(LPT,0,1)
Nbody = np.transpose(Nbody, axes = (0,2,1,3))
Nbody = swap(Nbody,0,1)
elif prand < 3./6:
LPT = np.transpose(LPT, axes = (0,1,3,2))
LPT = swap(LPT,1,2)
Nbody = np.transpose(Nbody, axes = (0,1,3,2))
Nbody = swap(Nbody,1,2)
elif prand < 4./6:
LPT = np.transpose(LPT, axes = (0,3,1,2))
LPT = swap(LPT,1,2)
LPT = swap(LPT,0,1)
Nbody = np.transpose(Nbody, axes = (0,3,1,2))
Nbody = swap(Nbody,1,2)
Nbody = swap(Nbody,0,1)
elif prand < 5./6:
LPT = np.transpose(LPT, axes = (0,3,2,1))
LPT = swap(LPT,0,2)
Nbody = np.transpose(Nbody, axes = (0,3,2,1))
Nbody = swap(Nbody,0,2)
return torch.from_numpy(LPT.copy()).float(),torch.from_numpy(Nbody.copy()).float()
#def crop_tensor(x):
# return x.narrow(2,1,x.shape[2]-1).narrow(3,1,x.shape[3]-1).narrow(4,1,x.shape[4]-1).contiguous()
def crop_tensor(x):
x = x.narrow(2,1,x.shape[2]-3).narrow(3,1,x.shape[3]-3).narrow(4,1,x.shape[4]-3).contiguous()
return x
def test_prediction(path,model,TestLoader):
net = torch.load(model)
net.cuda()
net.eval()
for t, data in enumerate(TestLoader, 0):
print (t)
NetInput = torch.autograd.Variable(data[0],requires_grad=False).cuda()
Y_pred = net(NetInput)
np.save(path+'test_'+str(t)+'.npy',np.concatenate((np.squeeze(Y_pred.data.cpu().numpy()),np.squeeze(data[1].numpy())),axis=0))
#data = np.fromfile('/mnt/home/siyuh/Project/Recon/data/data_version3/32-pancake/00-00-03-phi090/00000000-00001000.32.10.f4',dtype='f4').reshape([-1,32,32,32,10])
#data = np.fromfile('/mnt/home/siyuh/Project/Recon/data/data_version3/32-pancake/01.80/00000000-00001000.32.10.f4',dtype='f4').reshape([-1,32,32,32,10])
#data = np.fromfile('/mnt/home/siyuh/Project/Recon/data/data_version3/32-pancake/Om/om-0.32-00000000-00001000.32.10.f4',dtype='f4').reshape([-1,32,32,32,10])
#data = np.fromfile('/mnt/home/siyuh/Project/Recon/data/data_version3/32-pancake/32-sm/sm-24.00/00000000-00001000.32.10.f4',dtype='f4').reshape([-1,32,32,32,10])
#data = np.fromfile('/mnt/home/siyuh/Project/Recon/data/data_version3/32-pancake/dual_pancake/00-00-03-phi090-00-00-10-phi090/00000000-00001000.32.10.f4',dtype='f4').reshape([-1,32,32,32,10])
#for t in range(0,1):
# print (t)
# data_temp = torch.autograd.Variable(torch.from_numpy(np.expand_dims(np.einsum('ijkl->lijk', data[t][:,:,:,1:4]),axis=0)).float(),requires_grad=False).cuda()
# NetInput = torch.autograd.Variable(data_temp,requires_grad=False).cuda()
# Y_pred = net(NetInput)
# #np.save(path+'pancake_00-00-03-test_'+str(t)+'.npy',np.squeeze(Y_pred.data.cpu().numpy()))
# #np.save(path+'01.80_test_'+str(t)+'.npy',np.squeeze(Y_pred.data.cpu().numpy()))
# #np.save(path+'sm/sm_24_test_'+str(t)+'.npy',np.squeeze(Y_pred.data.cpu().numpy()))
# np.save(path+'dual_pancake/k-00-00-03_k-00-00-10_'+str(t)+'.npy',np.squeeze(Y_pred.data.cpu().numpy()))
def analysis(path,model,size, A, phi, k):
#data = genwaves(size, A, phi, k)
#data = np.einsum('ijkl->lijk',data)
data = np.zeros([3,32,32,32])
data = np.expand_dims(data,axis=0)
NetInput = torch.autograd.Variable(torch.from_numpy(data).float(),requires_grad=False).cuda()
net = torch.load(path+model)
net.cuda()
net.eval()
Y_pred = net(NetInput)
print (size,A,phi,k)
np.save(path+'A_'+str(A)+'_k_'+str(k).replace(" ","")+'_phi_'+str(phi)+'.npy',np.concatenate((np.squeeze(Y_pred.data.cpu().numpy()),np.squeeze(data)),axis=0))