-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathget_performance.py
182 lines (163 loc) · 10.2 KB
/
get_performance.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
import numpy as np
import torch
import torch.nn as nn
# import matlab.engine
import matplotlib
import matplotlib.pyplot as plt
import scipy.io
import seaborn as sns
import os
from tqdm import tqdm
from solver import test
# matlabeng = matlab.engine.start_matlab() #启动matlab
sns.set_style('ticks')
sns.set_context("poster")
plt.rcParams['font.sans-serif'] = 'Arial'
matplotlib.use('AGG')
class PostProcess():
def __init__(self, model, dataset, x_inp=None, x_out=None) -> None:
self.model = model
self.dataset = dataset
self.x_inp = x_inp
self.x_out = x_out
def __performance(self, label, pred, type='mean'):
if type == 'mean':
if len(label.shape) == 3:
MSE = np.sqrt(np.mean((pred - label) ** 2))
label_std = np.sqrt(np.mean((label - np.mean(label, axis=1)[:, None, :]) ** 2, axis=1))
RMSE = np.mean(np.sqrt(np.mean((pred - label) ** 2, axis=1)) / label_std)
MAE = np.mean(np.abs(pred - label))
RMAE = np.mean(np.max(np.abs(pred - label), axis=1) / label_std)
r_all = np.zeros(label.shape[0])
for i in range(label.shape[0]):
r_all[i] = np.corrcoef(label[i, :, :].ravel(), pred[i, :, :].ravel())[0, 1]
r = np.mean(r_all)
elif len(label.shape) == 2:
MSE = np.sqrt(np.mean(pred - label) ** 2)
label_std = np.sqrt(np.mean((label - np.mean(label, axis=0)[None, :]) ** 2), axis=0)
RMSE = np.mean(np.sqrt(np.mean((pred - label) ** 2, axis=0)) / label_std)
MAE = np.mean(np.abs(pred - label))
RMAE = np.mean(np.max(np.abs(pred - label), axis=0) / label_std)
r = np.corrcoef(label.ravel(), pred.ravel())[0, 1]
performance = {'MSE': MSE, 'RMSE': RMSE, 'MAE': MAE, 'RMAE': RMAE, 'r': r}
else:
MSE = np.sqrt(np.mean((pred - label) ** 2, axis=1))
label_std = np.sqrt(np.mean((label - np.mean(label, axis=1)[:, None, :]) ** 2, axis=1))
RMSE = MSE / label_std
MAE = np.mean(np.abs(pred - label), axis=1)
RMAE = np.max(np.abs(pred - label), axis=1) / label_std
r = np.zeros(label.shape[0])
for i in range(label.shape[0]):
r[i] = np.corrcoef(label[i, :, :].ravel(), pred[i, :, :].ravel())[0, 1]
performance = {'MSE': MSE, 'RMSE': RMSE, 'MAE': MAE, 'RMAE': RMAE, 'r': r}
return performance
def __pred(self, args, data_type):
if data_type == 'train':
data, label = self.dataset.train_data, self.dataset.train_label
pred, _ = test(self.model, nn.MSELoss(), torch.utils.data.DataLoader(self.dataset.train_dataset), self.dataset.batch)
elif data_type == 'valid':
data, label = self.dataset.valid_data, self.dataset.valid_label
pred, _ = test(self.model, nn.MSELoss(), torch.utils.data.DataLoader(self.dataset.valid_dataset))
else:
data, label = self.dataset.test_data, self.dataset.test_label
pred, _ = test(self.model, nn.MSELoss(), torch.utils.data.DataLoader(self.dataset.test_dataset))
pred = np.array(pred)
if args.normalize == 'minmax':
for i in range(len(data)):
data[i] = data[i] * (self.dataset.data_max[i] - self.dataset.data_min[i]) + self.dataset.data_min[i]
label = label * (self.dataset.label_max - self.dataset.label_min) + self.dataset.label_min
pred = pred * (self.dataset.label_max - self.dataset.label_min) + self.dataset.label_min
elif args.normalize == 'standard':
for i in range(len(data)):
data[i] = data[i] * self.dataset.data_std[i] + self.dataset.data_mean[i]
label = label * self.dataset.label_std + self.dataset.label_mean
pred = pred * self.dataset.label_std + self.dataset.label_mean
if args.noisy > 0:
label = self.dataset.out_data[self.dataset.train_idx, :, :]
return data, label, pred
def getResults(self, args, results_path, train_time, train_loss, valid_loss, recover=None):
self.dataset.train_data, self.dataset.train_label, self.dataset.train_pred = self.__pred(args, 'train')
if recover is not None:
self.dataset.train_data, self.dataset.train_label, self.dataset.train_pred = self.dataset.recoverData(self.dataset.train_data, self.dataset.train_label, self.dataset.train_pred, recover)
self.train_performance = self.__performance(self.dataset.train_label, self.dataset.train_pred)
print('Train set | ', end='')
for key, value in self.train_performance.items():
print('%s: %.3E, ' % (key, value), end='')
print('\n')
self.dataset.valid_data, self.dataset.valid_label, self.dataset.valid_pred = self.__pred(args, 'valid')
if recover is not None:
self.dataset.valid_data, self.dataset.valid_label, self.dataset.valid_pred = self.dataset.recoverData(self.dataset.valid_data, self.dataset.valid_label, self.dataset.valid_pred, recover)
self.valid_performance = self.__performance(self.dataset.valid_label, self.dataset.valid_pred)
print('Valid set | ', end='')
for key, value in self.valid_performance.items():
print('%s: %.3E, ' % (key, value), end='')
print('\n')
self.dataset.test_data, self.dataset.test_label, self.dataset.test_pred = self.__pred(args, 'test')
if recover is not None:
self.dataset.test_data, self.dataset.test_label, self.dataset.test_pred = self.dataset.recoverData(self.dataset.test_data, self.dataset.test_label, self.dataset.test_pred, recover)
self.test_performance = self.__performance(self.dataset.test_label, self.dataset.test_pred)
print('Test set | ', end='')
for key, value in self.test_performance.items():
print('%s: %.3E, ' % (key, value), end='')
print('\n')
## 输出performance数据
perfile = open(os.path.join(results_path, 'performance.out'), 'w')
datatype = ['Train', 'Valid', 'Test']
allperformance = [self.train_performance, self.valid_performance, self.test_performance]
perfile.write('训练总次数:\t\t%d\n' % args.epochs)
perfile.write('训练总时间:\t\t%.2f\n' % train_time)
perfile.write('训练最好次数:\t\t%d\n' % np.argmin(train_loss))
perfile.write('验证最好次数:\t\t%d\n' % np.argmin(valid_loss))
for i, per in enumerate(allperformance):
for key, value in per.items():
if key == 'r':
perfile.write(datatype[i] + '-' + key + ':\t\t%.1f\n' % (100 * value))
else:
perfile.write(datatype[i] + '-' + key + ':\t\t%.3E\n' % value)
perfile.close()
# 保存结果数据
scipy.io.savemat(os.path.join(results_path, 'result.mat'),
{'train_pred': self.dataset.train_pred, 'test_pred': self.dataset.test_pred, 'valid_pred': self.dataset.valid_pred, 'train_loss': train_loss, 'valid_loss': valid_loss, 'train_performance': self.train_performance, 'valid_performance': self.valid_performance, 'test_performance': self.test_performance})
def __plotOneSet(self, results_path, label, pred, legend, xlabel, ylabel, dim=0, lgd_each=None, scale='linear'):
performance = self.__performance(label, pred, type='all')
if label.shape[0] > 200:
pbar = tqdm(range(0, label.shape[0], int(label.shape[0] / 200)), desc=legend + ' Plotting', ncols=100)
else:
pbar = tqdm(range(0, label.shape[0]), desc=legend + ' Plotting', ncols=100)
for i in pbar:
plt.figure(figsize=(8, 6))
if self.x_out is not None:
if scale == 'linear':
plt.plot(self.x_out, label[i, :, dim], 'k', label='Label')
plt.plot(self.x_out, pred[i, :, dim], 'r--', label='Prediction')
elif scale == 'logx':
plt.semilogx(self.x_out, label[i, :, dim], 'k', label='Label')
plt.semilogx(self.x_out, pred[i, :, dim], 'r--', label='Prediction')
elif scale == 'logy':
plt.semilogy(self.x_out, label[i, :, dim], 'k', label='Label')
plt.semilogy(self.x_out, pred[i, :, dim], 'r--', label='Prediction')
elif scale == 'loglog':
plt.loglog(self.x_out, label[i, :, dim], 'k', label='Label')
plt.loglog(self.x_out, pred[i, :, dim], 'r--', label='Prediction')
else:
print('Plot scale error!')
else:
plt.plot(label[i, :, dim], 'k', label='Label')
plt.plot(pred[i, :, dim], 'r--', label='Prediction')
plt.legend()
# plt.xlim([0.3, 20])
# plt.ylim([0.5, 30])
plt.xlabel(xlabel)
plt.ylabel(ylabel)
if lgd_each is None:
plt.savefig(os.path.join(results_path, 'figures', '%s%d_e%.3f_r%.1f.svg' % (legend, i, performance['MSE'][i], 100 * performance['r'][i])), bbox_inches='tight')
else:
plt.savefig(os.path.join(results_path, 'figures', '%s%d_%s_e%.3f_r%.1f.svg' % (legend, i, lgd_each[i], performance['MSE'][i], 100 * performance['r'][i])), bbox_inches='tight')
plt.close('all')
return performance
def plotResults(self, results_path, xlabel, ylabel, dim=0, lgd_each=None, scale='linear'):
train_per = self.__plotOneSet(results_path, self.dataset.train_label, self.dataset.train_pred, 'Train', xlabel, ylabel, dim, lgd_each, scale)
valid_per = self.__plotOneSet(results_path, self.dataset.valid_label, self.dataset.valid_pred, 'Valid', xlabel, ylabel, dim, lgd_each, scale)
test_per = self.__plotOneSet(results_path, self.dataset.test_label, self.dataset.test_pred, 'Test', xlabel, ylabel, dim, lgd_each, scale)
scipy.io.savemat(os.path.join(results_path, 'performance.mat'),
{'train-performance': train_per, 'valid-performance': valid_per, 'test-performance': test_per})