-
Notifications
You must be signed in to change notification settings - Fork 0
/
FK_Data_QPT.py
155 lines (140 loc) · 6.77 KB
/
FK_Data_QPT.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
from FK_DMFT import DMFT, op_loc
from FK_Data import Ham
import time, os
import matplotlib.pyplot as plt
import numpy as np
import torch
import torch.multiprocessing as mp
from functools import partial
from utils import mymkdir, myceil
import warnings
warnings.filterwarnings('ignore')
os.environ['OMP_NUM_THREADS'] = '1'
os.environ['OPENBLAS_NUM_THREADS'] = '1'
os.environ['MKL_NUM_THREADS'] = '1'
os.environ['VECLIB_MAXIMUM_THREADS'] = '1'
os.environ['NUMEXPR_NUM_THREADS'] = '1'
torch.set_num_threads(1)
def genData(L, t2, mu_devi, DMFT_data=False, phase_set=1):
torch.manual_seed(0)
'''construct DMFT'''
count = 20
iota = 0.
momentum = 0.5
momDisor = 0.
maxEpoch = 10
milestone = 50
f_filling = 0.5
d_filling = 0.5
tol_sc = 1e-6
tol_bi = 1e-7
gap = 5.
T_float = 0.005
scf = DMFT(count, iota, momentum, momDisor, maxEpoch, milestone, f_filling, d_filling, tol_sc, tol_bi, gap, device)
dtype = torch.double
'''gen Ham data'''
# t2 = torch.arange(0, 0.1, 0.05, device=device)
t2_amount = t2.shape[0]
T = T_float * torch.ones(t2_amount, device=device, dtype=dtype).unsqueeze(-1) # (bz, 1)
mu0 = torch.zeros((t2_amount, 1, 1), device=device, dtype=dtype) # (t2_amount, 1, 1)
H0 = torch.stack([Ham(L, mu=0., tp=i.item()) for i in t2], dim=0).unsqueeze(1).to(device) # (t2_amount, 1, size, size)
mu = scf.bisearch_dec(partial(scf.fix_filling, f_ele=False), mu0, H0, T) # (t2_amount, 1, 1)
# mu_devi = torch.arange(-0.1, 0.5, 0.05, device=device)
H0_samples = []
mu_samples = []
for i in range(t2_amount):
mu_sample = mu[i] + mu_devi[:, None, None].to(device=device, dtype=dtype) # (mu_devi_amount, 1, 1)
H0_sample = H0[i] - torch.diag_embed(mu_sample.tile(1, 1, L ** 2)) # (mu_devi_amount, 1, size, size)
mu_samples.append(mu_sample)
H0_samples.append(H0_sample)
bz = mu_devi.shape[0] * t2_amount # bz = mu_devi_amount * t2_amount
H0_samples = torch.cat(H0_samples, dim=0) # (bz, 1, size, size)
T_samples = T_float * torch.ones(bz, device=device, dtype=dtype) # (bz,)
U_samples = torch.ones(bz, device=device, dtype=dtype) # (bz,)
# mu_samples = torch.cat(mu_samples, dim=0).squeeze() # (bz,)
# t2_samples = torch.repeat_interleave(t2, mu_devi.shape[0]) # (bz,)
if not DMFT_data:
phase_labels = phase_set * torch.ones_like(T_samples)
labels = torch.stack((U_samples, T_samples, phase_labels), dim=-1) # (bz, 3)
return H0_samples.cpu(), labels.cpu()
else:
'''gen DMFT data'''
train_batchsize = 100
OPs = []
nfs = []
bad_idxs = []
errors = []
for i in range(myceil(bz / train_batchsize)):
H0_batch = H0_samples[i * train_batchsize:(i + 1) * train_batchsize].to(device)
T_batch = T_samples[i * train_batchsize:(i + 1) * train_batchsize].to(device)
U_batch = U_samples[i * train_batchsize:(i + 1) * train_batchsize].to(device)
SE_batch, OP_batch, nf_batch, bad_batch = scf(T_batch, H0_batch, U_batch, reOP=True, reNf=True, reBad=True,
OPfuns=(op_loc,), prinfo=True) # (bz, 1, size)
bad_idx_batch, bad_error_batch = bad_batch
OPs.append(OP_batch)
nfs.append(nf_batch)
bad_idxs.append(bad_idx_batch + i * train_batchsize)
errors.append(bad_error_batch)
OP = torch.cat(OPs, dim=0)
nf = torch.cat(nfs, dim=0)
idx = torch.cat(bad_idxs, dim=0)
errors = torch.cat(errors, dim=0)
labels = torch.stack((U_samples, T_samples, OP), dim=-1) # (amount,3)
return H0_samples, labels
# OP_np = torch.cat(OPs, dim=0).cpu().numpy()
# nf_np = torch.cat(nfs, dim=0).cpu().numpy()
# idx_np = torch.cat(bad_idxs, dim=0).cpu().numpy()
# errors_np = torch.cat(errors, dim=0).cpu().numpy()
# mu_np = mu_samples.cpu().numpy()
# t2_np = t2_samples.cpu().numpy()
# '去除bad data'
# mu_np = np.delete(mu_np, idx_np)
# t2_np = np.delete(t2_np, idx_np)
# OP_np = np.delete(OP_np, idx_np, axis=1)
# nf_np = np.delete(nf_np, idx_np, axis=0)
# np.savez("data_result", mu=mu_np, t2=t2_np, OP=OP_np, nf=nf_np, idx_bad=idx_np, error=errors_np)
# for i in np.arange(mu_np.shape[0]):
# print(f"mu={mu_np[i]},t2={t2_np[i]},OP={OP_np[:,i]},best_op={np.argmax(OP_np[:,i])}")
# print(np.round(nf_np[i,:].reshape(12,12),2))
# print(time.time() - t)
def main(L, t2_amount, mu_amount):
# generate data without DMFT
t2 = torch.linspace(0, 0.35, t2_amount)
mu_devi = torch.linspace(-0.1, 0.5, mu_amount)
Hs, labels = genData(L, t2, mu_devi, DMFT_data=False, phase_set=0) # label=0, checkboard
indices = torch.randperm(len(labels))
H0_test, l0_test = Hs[indices[:int(len(labels) / 11)]], labels[indices[:int(len(labels) / 11)]]
H0_train, l0_train = Hs[indices[int(len(labels) / 11):]], labels[indices[int(len(labels) / 11):]]
t2 = torch.linspace(0.8, 1.2, t2_amount)
Hs, labels = genData(L, t2, mu_devi, DMFT_data=False, phase_set=1) # label=1, stripe
indices = torch.randperm(len(labels))
H1_test, l1_test = Hs[indices[:int(len(labels) / 11)]], labels[indices[:int(len(labels) / 11)]]
H1_train, l1_train = Hs[indices[int(len(labels) / 11):]], labels[indices[int(len(labels) / 11):]]
H_train = torch.cat((H0_train, H1_train), dim=0)
H_test = torch.cat((H0_test, H1_test), dim=0)
l_train = torch.cat((l0_train, l1_train), dim=0)
l_test = torch.cat((l0_test, l1_test), dim=0)
return H_train, l_train, H_test, l_test
if __name__ == '__main__':
device = torch.device("cpu")
L = 12
t2_amount, mu_amount = 110, 50 # total_amount = t2_amount * mu_amount * 2
path = 'datasets/FK_{}_QPT'.format(L)
mymkdir(path)
mymkdir('{}/train'.format(path))
mymkdir('{}/test'.format(path))
t = time.time()
H_train, l_train, H_test, l_test = main(L, t2_amount, mu_amount)
delta_t = time.time() - t
print(delta_t, '\n', L, H_train.shape, l_train.shape, H_test.shape, l_test.shape)
f = open('{}/train/info.txt'.format(path), 'w')
f.write('time={}\nL={}\ndataset.shape={}\nlabels.shape={}'.format(delta_t, L, H_train.shape, l_train.shape))
f.close()
f = open('{}/test/info.txt'.format(path), 'w')
f.write('time={}\nL={}\ndataset.shape={}\nlabels.shape={}'.format(delta_t, L, H_test.shape, l_test.shape))
f.close()
# save
torch.save(H_train, '{}/train/dataset.pt'.format(path)) # (bz, 1, L ** 2, L ** 2)
torch.save(l_train, '{}/train/labels.pt'.format(path)) # (bz, 3)
torch.save(H_test, '{}/test/dataset.pt'.format(path)) # (bz, 1, L ** 2, L ** 2)
torch.save(l_test, '{}/test/labels.pt'.format(path)) # (bz, 3)