-
Notifications
You must be signed in to change notification settings - Fork 0
/
QLTTrain.py
531 lines (482 loc) · 24.7 KB
/
QLTTrain.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
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
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import Dataset, DataLoader
from torch import optim
from torch.optim import lr_scheduler
import os
import numpy as np
import matplotlib.pyplot as plt
import math
from utils import compute_grads, AverageMeter
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)
torch.manual_seed(0)
Net = 'Sig_'
train = False
batchsz = 256
order = 2
class Model_relu(nn.Module):
def __init__(self, input_size, embedding_size=100, hidden_size=64, output_size=2):
super(Model_relu, self).__init__()
self.pre = nn.Linear(input_size, embedding_size)
self.emb = nn.Linear(embedding_size, hidden_size)
self.fc1 = nn.Linear(hidden_size, hidden_size)
self.fc2 = nn.Linear(hidden_size, hidden_size)
self.fc3 = nn.Linear(hidden_size, hidden_size)
self.fc4 = nn.Linear(hidden_size, hidden_size)
self.out = nn.Linear(hidden_size, output_size)
self.pre.apply(self.weights_init)
self.emb.apply(self.weights_init)
self.fc1.apply(self.weights_init)
self.fc2.apply(self.weights_init)
self.fc3.apply(self.weights_init)
self.fc4.apply(self.weights_init)
self.out.apply(self.weights_init)
def weights_init(self, m):
if isinstance(m, nn.Linear):
# nn.init.xavier_uniform_(m.weight)
# nn.init.xavier_normal_(m.weight)
nn.init.kaiming_uniform_(m.weight)
# nn.init.kaiming_normal_(m.weight)
if m.bias is not None:
torch.nn.init.zeros_(m.bias)
def forward(self, x):
x = F.relu(self.pre(x))
x = F.relu(self.emb(x))
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = F.relu(self.fc3(x))
x = F.relu(self.fc4(x))
return self.out(x)
class Model_sig(nn.Module):
def __init__(self, input_size, embedding_size=100, hidden_size=64, output_size=2):
super(Model_sig, self).__init__()
self.pre = nn.Linear(input_size, embedding_size)
self.emb = nn.Linear(embedding_size, hidden_size)
self.fc1 = nn.Linear(hidden_size, hidden_size)
self.fc2 = nn.Linear(hidden_size, hidden_size)
self.fc3 = nn.Linear(hidden_size, hidden_size)
self.fc4 = nn.Linear(hidden_size, hidden_size)
self.out = nn.Linear(hidden_size, output_size)
self.pre.apply(self.weights_init)
self.emb.apply(self.weights_init)
self.fc1.apply(self.weights_init)
self.fc2.apply(self.weights_init)
self.fc3.apply(self.weights_init)
self.fc4.apply(self.weights_init)
self.out.apply(self.weights_init)
def weights_init(self, m):
if isinstance(m, nn.Linear):
# nn.init.xavier_uniform_(m.weight)
# nn.init.xavier_normal_(m.weight)
nn.init.kaiming_uniform_(m.weight)
# nn.init.kaiming_normal_(m.weight)
if m.bias is not None:
torch.nn.init.zeros_(m.bias)
def forward(self, x):
x = F.sigmoid(self.pre(x))
x = F.sigmoid(self.emb(x))
x = F.sigmoid(self.fc1(x))
x = F.sigmoid(self.fc2(x))
x = F.sigmoid(self.fc3(x))
x = F.sigmoid(self.fc4(x))
return self.out(x)
def vectorized_result(j):
"""Return a 2-dimensional unit vector with a 1.0 in the j'th position
and zeroes elsewhere.
"""
e = np.zeros((2, 1))
e[j] = 1.0
return e
def load_data_wrapper(tr_d, tr_r, te_d, te_r):
training_data = list(zip(tr_d, [vectorized_result(y) for y in tr_r]))
test_data = list(zip(te_d, te_r))
return (training_data, test_data)
class Loadtrain(Dataset):
def __init__(self, dataload):
self.dataload = dataload
def __getitem__(self, index):
g = self.dataload[index]
label = torch.tensor(np.argmax(g[-1]), dtype=torch.long)
g = torch.from_numpy(np.squeeze(g[0])).float()
return g, label
def __len__(self):
return len(self.dataload)
class Loadtest(Dataset):
def __init__(self, dataload):
self.dataload = dataload
def __getitem__(self, index):
g = self.dataload[index]
label = torch.tensor(g[-1], dtype=torch.long)
g = torch.from_numpy(np.squeeze(g[0])).float()
return g, label
def __len__(self):
return len(self.dataload)
def generate_insulator_data(count, qlt, kappa=1.0, delta=0.25, dmax=1, L=12, autocorr=4*12*12):
"""Generate a specific number of samples of the Chern/normal model with model parameter delta = 0.5 and kappa.
The default kappa is at the critical kappa_c. dmax controls the cut-off length scale of the operators. qlt is a list of local triangles."""
N = int(L*L/2)
training_data = []
conf = -np.ones((L, L), dtype=int)
index = np.zeros((L, L), dtype=int)
corr=np.zeros((L,L,2*dmax+1,2*dmax+1,3),dtype=complex)
sample = np.zeros((L,L,len(qlt)),dtype=float)
for x in range(L):
for y in range(L):
if (x-y)%2 == 0:
conf[x,y]*=-1
index[x,y] = int((y*L+x)/2)
""" Initial configuration. Checkerboard occupation of electrons, 1 for occupied and -1 for unoccupied.
Index keep track the row number of the (x, y) fermion in the Slater determinant (and its supplement). """
SD = np.zeros((N, N), dtype=complex)
SD2= np.zeros((N, N), dtype=complex)
""" Initialize the Slater determinant (for the occupied sites) and its supplement (for the unoccupied sites).
Changing the configuration will switch the corresponding rows of the Slater determinant and its supplement."""
for kx in range(L):
for ky in range(int(L/2)):
energy = math.sqrt(math.cos(kx*2*np.pi/L)**2 + math.cos(ky*2*np.pi/L)**2 + (2*delta*math.sin(ky*2*np.pi/L)*(1-kappa+kappa*math.sin(kx*2*np.pi/L)))**2)
for x in range(L):
for y in range(L):
if conf[x,y] == 1:
"""Occupied site, go to Slater determinant."""
if (y%2 == 0):
"""First sublattice."""
SD[index[x,y], ky*L+kx] = complex(math.cos((kx*x+ky*y)*2*np.pi/L), math.sin((kx*x+ky*y)*2*np.pi/L)) * \
complex(math.cos(ky*2*np.pi/L), -2*delta*math.sin(ky*2*np.pi/L) * (1-kappa+kappa*math.sin(kx*2*np.pi/L))) / \
math.sqrt((math.cos(kx*2*np.pi/L)+energy)**2 + math.cos(ky*2*np.pi/L) ** 2 + (2*delta*math.sin(ky*2*np.pi/L) * (1-kappa+kappa*math.sin(kx*2*np.pi/L)))**2)
else:
"""Second sublattice."""
SD[index[x,y], ky*L+kx] = complex(math.cos((kx*x+ky*y)*2*np.pi/L), math.sin((kx*x+ky*y)*2*np.pi/L)) * \
(math.cos(kx*2*np.pi/L)+energy) / \
math.sqrt((math.cos(kx*2*np.pi/L)+energy)**2 + math.cos(ky*2*np.pi/L) ** 2 + (2*delta*math.sin(ky*2*np.pi/L) * (1-kappa+kappa*math.sin(kx*2*np.pi/L)))**2)
else:
"""Unoccupied site, go to supplement."""
if (y%2 == 0):
"""First sublattice."""
SD2[index[x,y], ky*L+kx] = complex(math.cos((kx*x+ky*y)*2*np.pi/L), math.sin((kx*x+ky*y)*2*np.pi/L)) * \
complex(math.cos(ky*2*np.pi/L), -2*delta*math.sin(ky*2*np.pi/L) * (1-kappa+kappa*math.sin(kx*2*np.pi/L))) / \
math.sqrt((math.cos(kx*2*np.pi/L)+energy)**2 + math.cos(ky*2*np.pi/L) ** 2 + (2*delta*math.sin(ky*2*np.pi/L) * (1-kappa+kappa*math.sin(kx*2*np.pi/L)))**2)
else:
"""Second sublattice."""
SD2[index[x,y], ky*L+kx] = complex(math.cos((kx*x+ky*y)*2*np.pi/L), math.sin((kx*x+ky*y)*2*np.pi/L)) * \
(math.cos(kx*2*np.pi/L)+energy) / \
math.sqrt((math.cos(kx*2*np.pi/L)+energy)**2 + math.cos(ky*2*np.pi/L) ** 2 + (2*delta*math.sin(ky*2*np.pi/L) * (1-kappa+kappa*math.sin(kx*2*np.pi/L)))**2)
SDinv = np.linalg.inv(SD)
#print(np.amax(abs(np.dot(SD, SDinv)-np.diag(np.ones(N)))))
"""Also keep the inverse of Slater determinant. Initialization complete."""
for icount in range(count*30+10):
for iautocorr in range(autocorr):
""" A number of steps between measurements to ensure (approximately) independent samples."""
while True:
x, y = np.random.randint(L, size=2)
if conf[x,y] == 1:
break
while True:
xp, yp = np.random.randint(L, size=2)
if conf[xp,yp] == -1:
break
if (np.random.rand()<abs(1+np.dot(SD2[index[xp,yp],:]-SD[index[x,y],:],SDinv[:, index[x,y]]))**2):
"""Calculate the update probability. Accept and update configuration."""
conf[x,y] *=-1
conf[xp,yp] *=-1
"""Update matrix inverse and matrices."""
SDinv[:,index[x,y]] /= np.dot(SD2[index[xp,yp],:],SDinv[:,index[x,y]])
tempnp = np.reshape(np.dot(SD2[index[xp,yp],:], SDinv),N)
for i in range(N):
if i != index[x,y]:
SDinv[:,i]-=SDinv[:,index[x,y]]*tempnp[i]
tempnp = SD[index[x,y],:]
SD[index[x,y],:] = SD2[index[xp,yp],:]
SD2[index[xp,yp],:] = tempnp
tempi = index[x,y]
index[x,y] = index[xp,yp]
index[xp,yp] = tempi
if(icount > 9):
for x in range(L):
for y in range(L):
for dx in range(2*dmax+1):
for dy in range(2*dmax+1):
xp = modpbc(x+dx-dmax,L)
yp = modpbc(y+dy-dmax,L)
if(conf[x,y]==1 and conf[xp,yp]==-1):
corr[x,y,dx,dy,(icount-10)%3] = 1+np.dot(SD2[index[xp,yp],:]-SD[index[x,y],:],SDinv[:,index[x,y]])
else:
corr[x,y,dx,dy,(icount-10)%3] = 0
if ((icount-10)%3 == 2):
for x in range(L):
for y in range(L):
for iqlt in range(len(qlt)):
xp = modpbc(x+qlt[iqlt][0],L)
yp = modpbc(y+qlt[iqlt][1],L)
xpp = modpbc(xp+qlt[iqlt][2],L)
ypp = modpbc(yp+qlt[iqlt][3],L)
sample[x,y,iqlt]+=(corr[x, y, qlt[iqlt][0]+dmax, qlt[iqlt][1]+dmax,0] * corr[xp, yp, qlt[iqlt][2]+dmax, qlt[iqlt][3]+dmax,1]+corr[xpp, ypp, dmax-qlt[iqlt][0]-qlt[iqlt][2], dmax-qlt[iqlt][1]-qlt[iqlt][3], 2]).imag \
+ (corr[x, y, qlt[iqlt][0]+dmax, qlt[iqlt][1]+dmax,1] * corr[xp, yp, qlt[iqlt][2]+dmax, qlt[iqlt][3]+dmax,2]+corr[xpp, ypp, dmax-qlt[iqlt][0]-qlt[iqlt][2], dmax-qlt[iqlt][1]-qlt[iqlt][3], 0]).imag \
+ (corr[x, y, qlt[iqlt][0]+dmax, qlt[iqlt][1]+dmax,2] * corr[xp, yp, qlt[iqlt][2]+dmax, qlt[iqlt][3]+dmax,0]+corr[xpp, ypp, dmax-qlt[iqlt][0]-qlt[iqlt][2], dmax-qlt[iqlt][1]-qlt[iqlt][3], 1]).imag \
+ (corr[x, y, qlt[iqlt][0]+dmax, qlt[iqlt][1]+dmax,0] * corr[xp, yp, qlt[iqlt][2]+dmax, qlt[iqlt][3]+dmax,2]+corr[xpp, ypp, dmax-qlt[iqlt][0]-qlt[iqlt][2], dmax-qlt[iqlt][1]-qlt[iqlt][3], 1]).imag \
+ (corr[x, y, qlt[iqlt][0]+dmax, qlt[iqlt][1]+dmax,2] * corr[xp, yp, qlt[iqlt][2]+dmax, qlt[iqlt][3]+dmax,1]+corr[xpp, ypp, dmax-qlt[iqlt][0]-qlt[iqlt][2], dmax-qlt[iqlt][1]-qlt[iqlt][3], 0]).imag \
+ (corr[x, y, qlt[iqlt][0]+dmax, qlt[iqlt][1]+dmax,1] * corr[xp, yp, qlt[iqlt][2]+dmax, qlt[iqlt][3]+dmax,0]+corr[xpp, ypp, dmax-qlt[iqlt][0]-qlt[iqlt][2], dmax-qlt[iqlt][1]-qlt[iqlt][3], 2]).imag
if ((icount-10)%30 == 29):
training_data.append(np.reshape(sample, (L**2*len(qlt), 1)))
sample = np.zeros((L,L,len(qlt)),dtype=float)
#print(np.amax(abs(np.dot(SD, SDinv)-np.diag(np.ones(N)))))
return training_data
def modpbc(x, L=28):
"""Ensure the coordinate is between [0, L-1] by applying periodic boundary condition. """
if (x<0):
return modpbc(x+L, L)
elif (x>L-1):
return modpbc(x-L, L)
else:
return x
def countriangle(dmax):
"""The list of triangles within a cut-off scale dmax."""
qlt = []
for dx in range(-dmax, dmax+1):
for dy in range(dmax+1):
for dx2 in range(-dmax, dmax+1):
for dy2 in range(-dy, dmax+1):
if(abs(dx+dx2) <= dmax and abs(dy+dy2) <= dmax and (dy2*dx-dy*dx2)>0 and (dy>0 or dx>0) and (dx+dx2>0 or dy+dy2>0)):
qlt.append([dx, dy, dx2, dy2])
return qlt
def plot_fig(x, p0, p1, xlabel, title):
fig = plt.figure()
ax = fig.add_subplot(111)
ax.plot(x, p0, '-b', label='p0')
ax2 = ax.twinx()
ax2.plot(x, p1, '-r', label='p1')
ax.legend(loc=6)
ax.grid()
ax.set_xlabel(xlabel)
ax.set_ylabel("p0")
ax.set_ylim(-0.05, 1.05)
ax2.set_ylabel("p1")
ax2.set_ylim(-0.05, 1.05)
ax2.legend(loc=7)
plt.title(title)
plt.savefig("{}.png".format(title))
def plotgrads(X, title, g=None):
plt.figure()
if g is not None: plt.plot(g, c='k', label='grads')
for i, x in enumerate(X, 1):
plt.plot(x, label='grads({})'.format(i))
plt.xlim([-0.5, X.shape[1] - 0.5])
plt.legend(loc='upper right')
plt.ylabel('Gradient')
plt.xlabel('Epoch')
plt.title(title)
plt.savefig("models/QLT/{}/{}.jpg".format(Net, title))
plt.show()
plt.close()
def plotbargrads(x, title):
plt.figure()
plt.xlim(0, len(x) + 1)
plt.bar(np.arange(1, len(x) + 1), x, width=0.8)
plt.ylabel('Gradient')
plt.xlabel('Layer')
plt.title(title)
plt.savefig("models/QLT/{}/{}.jpg".format(Net, title))
plt.show()
plt.close()
def accuracy(dataloader, model):
model.eval()
class_correct = list(0. for i in range(2))
class_total = list(0. for i in range(2))
with torch.no_grad():
for data in dataloader:
x, labels = data
x, labels = x.to(device), labels.to(device)
outputs = model(x)
predicted = torch.max(outputs, 1)[1]
c = (predicted == labels).squeeze()
for i in range(len(c)):
label = labels[i]
class_correct[label] += c[i].item()
class_total[label] += 1
print('Accuracy: %.5f %%' % (100.0 * sum(class_correct) / sum(class_total)))
for i in range(2):
print('Accuracy of %5s : %.5f %%' % (classes[i], 100.0 * class_correct[i] / class_total[i]))
if __name__ == "__main__":
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
classes = ('Normal', 'Chern')
if not os.path.exists("models/QLT/{}".format(Net)):
os.mkdir("models/QLT/{}".format(Net))
if train:
if os.path.exists("datasets/Insulator_dataset.npy"):
training_data, val_data = np.load("datasets/Insulator_dataset.npy", allow_pickle=True)
print("Datasets loading finished!")
else:
train_count, test_count = 10000, 1000
tr_d, tr_r, te_d, te_r = [], [], [], []
tr_d += generate_insulator_data(int(train_count / 2), countriangle(1), kappa=0.1)
tr_r += int(train_count / 2) * [0]
tr_d += generate_insulator_data(int(train_count / 2), countriangle(1), kappa=1.0)
tr_r += int(train_count / 2) * [1]
te_d += generate_insulator_data(int(test_count / 2), countriangle(1), kappa=0.1)
te_r += int(test_count / 2) * [0]
te_d += generate_insulator_data(int(test_count / 2), countriangle(1), kappa=1.0)
te_r += int(test_count / 2) * [1]
training_data, val_data = load_data_wrapper(tr_d, tr_r, te_d, te_r)
np.save("datasets/Insulator_dataset.npy", (training_data, val_data))
print("Datasets generation finished!")
trainset, valset = Loadtrain(training_data), Loadtest(val_data)
trainloader = DataLoader(trainset, batch_size=batchsz, shuffle=True, num_workers=0)
valloader = DataLoader(valset, batch_size=batchsz, shuffle=False, num_workers=0)
criterion = nn.CrossEntropyLoss().to(device)
if Net.startswith('Relu'):
model = Model_relu(training_data[0][0].shape[0], 100, 64, 2).to(device)
else:
model = Model_sig(training_data[0][0].shape[0], 100, 64, 2).to(device)
print(len(trainset), len(trainloader), len(valset), len(valloader), '\n', model)
optimizer = optim.Adam(model.parameters(), lr=1e-3, weight_decay=0)
scheduler = lr_scheduler.StepLR(optimizer, 50, gamma=0.5, last_epoch=-1)
Grads, Agrads, Grads1, Agrads1, Grads2, Agrads2, Grads_1, Agrads_1 = [], [],[], [], [], [], [], []
Grads3, Agrads3, Grads4, Agrads4, Grads5, Agrads5, Grads6, Agrads6 = [], [], [], [], [], [], [], []
best = 1e5
for epoch in range(21):
# train
grads = AverageMeter('Grad', ':.4e')
agrads = AverageMeter('AGrad', ':.4e')
grads1 = AverageMeter('Grad1', ':.4e')
agrads1 = AverageMeter('AGrad1', ':.4e')
grads2 = AverageMeter('Grad2', ':.4e')
agrads2 = AverageMeter('AGrad2', ':.4e')
grads3 = AverageMeter('Grad3', ':.4e')
agrads3 = AverageMeter('AGrad3', ':.4e')
grads4 = AverageMeter('Grad4', ':.4e')
agrads4 = AverageMeter('AGrad4', ':.4e')
grads5 = AverageMeter('Grad5', ':.4e')
agrads5 = AverageMeter('AGrad5', ':.4e')
grads6 = AverageMeter('Grad6', ':.4e')
agrads6 = AverageMeter('AGrad6', ':.4e')
grads_1 = AverageMeter('Grad-1', ':.4e')
agrads_1 = AverageMeter('AGrad-1', ':.4e')
model.train()
running_loss = 0.
for batchidx, (x, label) in enumerate(trainloader):
x, label = x.to(device), label.to(device)
logits = model(x)
loss = criterion(logits, label)
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_norm, avg_norm = compute_grads(model, order)
grads.update(total_norm, x.size(0))
agrads.update(avg_norm, x.size(0))
total_norm, avg_norm = compute_grads(model.pre, order)
grads1.update(total_norm, x.size(0))
agrads1.update(avg_norm, x.size(0))
total_norm, avg_norm = compute_grads(model.emb, order)
grads2.update(total_norm, x.size(0))
agrads2.update(avg_norm, x.size(0))
total_norm, avg_norm = compute_grads(model.fc1, order)
grads3.update(total_norm, x.size(0))
agrads3.update(avg_norm, x.size(0))
total_norm, avg_norm = compute_grads(model.fc2, order)
grads4.update(total_norm, x.size(0))
agrads4.update(avg_norm, x.size(0))
total_norm, avg_norm = compute_grads(model.fc3, order)
grads5.update(total_norm, x.size(0))
agrads5.update(avg_norm, x.size(0))
total_norm, avg_norm = compute_grads(model.fc4, order)
grads6.update(total_norm, x.size(0))
agrads6.update(avg_norm, x.size(0))
total_norm, avg_norm = compute_grads(model.out, order)
grads_1.update(total_norm, x.size(0))
agrads_1.update(avg_norm, x.size(0))
running_loss += loss.item() * len(label)
scheduler.step()
epoch_loss = running_loss / len(trainset)
Grads.append(float('{:.4e}'.format(grads.avg)))
Agrads.append(float('{:.4e}'.format(agrads.avg)))
Grads1.append(float('{:.4e}'.format(grads1.avg)))
Agrads1.append(float('{:.4e}'.format(agrads1.avg)))
Grads2.append(float('{:.4e}'.format(grads2.avg)))
Agrads2.append(float('{:.4e}'.format(agrads2.avg)))
Grads3.append(float('{:.4e}'.format(grads3.avg)))
Agrads3.append(float('{:.4e}'.format(agrads3.avg)))
Grads4.append(float('{:.4e}'.format(grads4.avg)))
Agrads4.append(float('{:.4e}'.format(agrads4.avg)))
Grads5.append(float('{:.4e}'.format(grads5.avg)))
Agrads5.append(float('{:.4e}'.format(agrads5.avg)))
Grads6.append(float('{:.4e}'.format(grads6.avg)))
Agrads6.append(float('{:.4e}'.format(agrads6.avg)))
Grads_1.append(float('{:.4e}'.format(grads_1.avg)))
Agrads_1.append(float('{:.4e}'.format(agrads_1.avg)))
# validation
model.eval()
with torch.no_grad():
running_loss = 0.
for batchidx, (x, label) in enumerate(valloader):
x, label = x.to(device), label.to(device)
logits = model(x)
loss = criterion(logits, label)
running_loss += loss.item() * len(label)
val_loss = running_loss / len(valset)
print(epoch, 'epoch loss:', epoch_loss, 'val loss:', val_loss)
# accuarcy
print("Train accuarcy:")
accuracy(trainloader, model)
print("Validation accuarcy:")
accuracy(valloader, model)
# save model
if best > val_loss:
best = val_loss
torch.save(model, "models/QLT/{}/model.pkl".format(Net))
print('Finished Training')
grads, agrads = np.array(Grads), np.array(Agrads)
grads1, agrads1 = np.array(Grads1), np.array(Agrads1)
grads2, agrads2 = np.array(Grads2), np.array(Agrads2)
grads3, agrads3 = np.array(Grads3), np.array(Agrads3)
grads4, agrads4 = np.array(Grads4), np.array(Agrads4)
grads5, agrads5 = np.array(Grads5), np.array(Agrads5)
grads6, agrads6 = np.array(Grads6), np.array(Agrads6)
grads_1, agrads_1 = np.array(Grads_1), np.array(Agrads_1)
Grads = np.stack((grads1, grads2, grads3, grads4, grads5, grads6, grads_1), axis=0)
Agrads = np.stack((agrads1, agrads2, agrads3, agrads4, agrads5, agrads6, agrads_1), axis=0)
np.save("models/QLT/{}/grads.npy".format(Net), grads)
np.save("models/QLT/{}/agrads.npy".format(Net), agrads)
np.save("models/QLT/{}/grads_layers.npy".format(Net), Grads)
np.save("models/QLT/{}/agrads_layers.npy".format(Net), Agrads)
plotgrads(Grads, 'QLT_Gradients', grads)
plotgrads(Agrads, 'QLT_Average Gradients', agrads)
Grads = np.mean(Grads[:, 1:2], axis=1)
NGrads = Grads / np.max(Grads)
Agrads = np.mean(Agrads[:, 1:2], axis=1)
NAgrads = Agrads / np.max(Agrads)
plotbargrads(NGrads, 'QLT_Normalized Gradients ({})'.format(Net[:Net.index('_')]))
plotbargrads(NAgrads, 'QLT_Normalized Average Gradients ({})'.format(Net[:Net.index('_')]))
else:
# model = torch.load("models/QLT/{}/model.pkl".format(Net), map_location=device)
if os.path.exists("models/QLT/{}/grads.npy".format(Net)):
grads = np.load("models/QLT/{}/grads.npy".format(Net))
agrads = np.load("models/QLT/{}/agrads.npy".format(Net))
Grads = np.load("models/QLT/{}/grads_layers.npy".format(Net))
Agrads = np.load("models/QLT/{}/agrads_layers.npy".format(Net))
plotgrads(Grads, 'QLT_Gradients', grads)
plotgrads(Agrads, 'QLT_Average Gradients', agrads)
Grads = np.mean(Grads[:, :3], axis=1)
NGrads = Grads / np.max(Grads)
Agrads = np.mean(Agrads[:, :3], axis=1)
NAgrads = Agrads / np.max(Agrads)
NGrads, NAgrads = NGrads[:-1], NAgrads[:-1]
plotbargrads(NGrads, 'QLT_Normalized Gradients ({})'.format(Net[:Net.index('_')]))
plotbargrads(NAgrads, 'QLT_Normalized Average Gradients ({})'.format(Net[:Net.index('_')]))
# # plot phase diagram
# model.eval()
# Kap, p0, p1, group = np.arange(0.1, 1.01, 0.05), [], [], 10
# for k in Kap:
# samples = torch.from_numpy(
# np.concatenate(generate_insulator_data(group, countriangle(1), kappa=k, dmax=1), axis=1).T).float().to(
# device)
# output = F.softmax(model(samples), dim=1)
# p0.append(torch.mean(output[:, 0]).item())
# p1.append(torch.mean(output[:, 1]).item())
# plot_fig(Kap, p0, p1, "Kappa", "Normal VS Topological")