-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrial_matrix_E.py
151 lines (140 loc) · 6.43 KB
/
trial_matrix_E.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
import argparse
from collections import defaultdict
import os
import pickle
import matplotlib.pyplot as plt
import numpy as np
from scipy import spatial
import argparse
import os
from pathlib import Path
import pickle
import timeit
import platform
import numpy as np
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.utils.data
from scipy import spatial
from model_E import GOODLE_E
def cal_dist(positions, accessible_R, ks, Rx, Ry, Rz):
num = len(positions)
tmp = np.repeat(np.array(positions), 4, axis=0)
tmp_x, tmp_y, tmp_z = np.zeros(tmp.shape), np.zeros(tmp.shape), np.zeros(tmp.shape)
tmp_x[:, 0] = tmp[:, 0]
tmp_y[:, 1] = tmp[:, 1]
tmp_z[:, 2] = tmp[:, 2]
tmp_kx = np.repeat(np.expand_dims(np.repeat(ks[:, 0].unsqueeze(1).cpu().numpy(), tmp.shape[0], axis=1), 1),
tmp.shape[0], axis=1)
tmp_ky = np.repeat(np.expand_dims(np.repeat(ks[:, 1].unsqueeze(1).cpu().numpy(), tmp.shape[0], axis=1), 1),
tmp.shape[0], axis=1)
tmp_kz = np.repeat(np.expand_dims(np.repeat(ks[:, 2].unsqueeze(1).cpu().numpy(), tmp.shape[0], axis=1), 1),
tmp.shape[0], axis=1)
distance_matrices, dist_xs, dist_ys, dist_zs, kRs = [], [], [], [], []
ori_dist = spatial.distance_matrix(tmp, tmp)
for R in accessible_R:
dist_R = spatial.distance_matrix(tmp, tmp - R)
distance_matrices.append(dist_R)
dist_x = (np.repeat(np.expand_dims(tmp_x[:, 0], axis=1), 4 * num, axis=1) -
np.repeat(np.expand_dims(tmp_x[:, 0], axis=0), 4 * num, axis=0))
dist_xs.append(dist_x)
dist_y = (np.repeat(np.expand_dims(tmp_y[:, 1], axis=1), 4 * num, axis=1) -
np.repeat(np.expand_dims(tmp_y[:, 1], axis=0), 4 * num, axis=0))
dist_ys.append(dist_y)
dist_z = (np.repeat(np.expand_dims(tmp_z[:, 2], axis=1), 4 * num, axis=1) -
np.repeat(np.expand_dims(tmp_z[:, 2], axis=0), 4 * num, axis=0))
dist_zs.append(dist_z)
kR = tmp_kx * dist_x / Rx + tmp_ky * dist_y / Ry + tmp_kz * dist_z / Rz
kRs.append(kR * (-np.pi))
return distance_matrices, dist_xs, dist_ys, dist_zs, kRs, ori_dist
def read_file(filename):
with open(filename) as file:
data = file.readlines()
# print(filename)
flag = True
# read position from file
Rx, Ry, Rz, alpha, beta, gamma = data[0].split()[:]
Rx, Ry, Rz, alpha, beta, gamma = float(Rx), float(Ry), float(Rz), np.deg2rad(float(alpha)), np.deg2rad(float(beta)), np.deg2rad(float(gamma))
R2 = np.array([1, 0, 0])
R3 = np.array([np.cos(gamma), np.sin(gamma), 0])
tmp_x = np.cos(beta)
tmp_y = (np.cos(alpha) - tmp_x * np.cos(gamma)) / np.sin(gamma)
tmp_z = np.sqrt(1 - tmp_x**2 - tmp_y ** 2)
R4 = np.array([tmp_x, tmp_y, tmp_z])
accessible_R = []
# motice that the original lattice vector has been normalized.
for i in [-1, 0, 1]:
for j in [-1, 0, 1]:
for k in [-1, 0, 1]:
accessible_R.append(R2 * i + R3 * j + R4 * k)
positions = []
for line in range(1, 30):
info = data[line].split()
if info[0] != 'C':
break
atom, x, y, z = data[line].split()
x, y, z = float(x), float(y), float(z)
positions.append(np.array([x, y, z]))
ks = [[0,0,0]]
true_bands = [[float(data[line])]]
return ks, true_bands, positions, accessible_R, Rx, Ry, Rz
#training process
torch.set_default_tensor_type(torch.DoubleTensor)
device = torch.device('cuda:2') if torch.cuda.is_available() else torch.device('cpu')
# model = GOODLE_E(device=device)
model = torch.load("./GOODLE_model.pt")
model.to(device)
# model.eval()
print("## GOODLE model loaded.")
lr = 1e-4
optimizer = optim.Adam(model.parameters(), lr)
epoch_num = 100
train_sets = [10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25]
# train_sets = [10, 11, 12, 13]
test_sets = [20, 21, 22]
# test_sets = [11,]
optimizer = optim.Adam(model.parameters(), lr)
# square case
def preprocess():
for num in train_sets:
all_names = os.listdir("./all_data/" + str(num))
for name in all_names:
ks, true_bands, positions, accessible_R, Rx, Ry, Rz = read_file("all_data/" + str(num) + "/" + name)
ks = torch.tensor(ks)
distance_matrices, dist_xs, dist_ys, dist_zs, kRs, ori_dist = cal_dist(positions, accessible_R, ks, Rx, Ry, Rz)
np.save("./pre_processed/"+name, np.array([positions, distance_matrices, ori_dist, dist_xs, dist_ys, dist_zs, kRs, accessible_R, ks.numpy(), true_bands]))
# # preprocess()
out_file = open("record.txt", "w")
for i in range(epoch_num + 1):
# change to batch ?
optimizer.zero_grad()
time0 = time.time()
print("epoch {} starts".format(i))
training_loss = []
for num in train_sets:
all_names = os.listdir("./all_data/"+str(num))
for name in all_names:
positions, distance_matrices, ori_dist, dist_xs, dist_ys, dist_zs, kRs, accessible_R, ks, true_bands = np.load("pre_processed/" + name + ".npy", allow_pickle=True)
predict_bands = model.forward(torch.tensor(positions, device=device),
distance_matrices, ori_dist, dist_xs, dist_ys, dist_zs, kRs,
torch.tensor(accessible_R,device=device),
torch.tensor(ks, device=device))
true_bands = torch.tensor(true_bands, device=device)
loss = F.smooth_l1_loss(predict_bands, true_bands)
loss.backward()
optimizer.step()
optimizer.zero_grad()
training_loss.append(loss.cpu().detach().numpy())
print(f"{num} completed, time: {time.time()- time0}, loss:{np.average(training_loss)}")
with open("record.txt", "a") as out_file:
out_file.write(f"{num} completed, time: {time.time()- time0}, loss until now:{np.average(training_loss)/num}\n")
print(i, time.time() - time0,
np.average(training_loss))
with open("record.txt", "a") as out_file:
out_file.write(f"epoch {i} completed, time: {time.time() - time0}, loss:{np.average(training_loss) / num}\n")
if i % 5 == 0:
torch.save(model, f"20220303_original_save_E_epoch_{i}.pt")
torch.save(model, '20220303_original_save_E.pt')