forked from tomgoldstein/loss-landscape
-
Notifications
You must be signed in to change notification settings - Fork 0
/
plot_1D.py
170 lines (137 loc) · 6.22 KB
/
plot_1D.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
"""
1D plotting routines
"""
from matplotlib import pyplot as pp
import h5py
import argparse
import numpy as np
def plot_1d_loss_err(surf_file, xmin=-1.0, xmax=1.0, loss_max=5, log=False, show=False):
print('------------------------------------------------------------------')
print('plot_1d_loss_err')
print('------------------------------------------------------------------')
f = h5py.File(surf_file,'r')
print(f.keys())
x = f['xcoordinates'][:]
assert 'train_loss' in f.keys(), "'train_loss' does not exist"
train_loss = f['train_loss'][:]
train_acc = f['train_acc'][:]
print("train_loss")
print(train_loss)
print("train_acc")
print(train_acc)
xmin = xmin if xmin != -1.0 else min(x)
xmax = xmax if xmax != 1.0 else max(x)
# loss and accuracy map
fig, ax1 = pp.subplots()
ax2 = ax1.twinx()
if log:
tr_loss, = ax1.semilogy(x, train_loss, 'b-', label='Training loss', linewidth=1)
else:
tr_loss, = ax1.plot(x, train_loss, 'b-', label='Training loss', linewidth=1)
tr_acc, = ax2.plot(x, train_acc, 'r-', label='Training accuracy', linewidth=1)
if 'test_loss' in f.keys():
test_loss = f['test_loss'][:]
test_acc = f['test_acc'][:]
if log:
te_loss, = ax1.semilogy(x, test_loss, 'b--', label='Test loss', linewidth=1)
else:
te_loss, = ax1.plot(x, test_loss, 'b--', label='Test loss', linewidth=1)
te_acc, = ax2.plot(x, test_acc, 'r--', label='Test accuracy', linewidth=1)
pp.xlim(xmin, xmax)
ax1.set_ylabel('Loss', color='b', fontsize='xx-large')
ax1.tick_params('y', colors='b', labelsize='x-large')
ax1.tick_params('x', labelsize='x-large')
ax1.set_ylim(0, loss_max)
ax2.set_ylabel('Accuracy', color='r', fontsize='xx-large')
ax2.tick_params('y', colors='r', labelsize='x-large')
ax2.set_ylim(0, 100)
pp.savefig(surf_file + '_1d_loss_acc' + ('_log' if log else '') + '.pdf',
dpi=300, bbox_inches='tight', format='pdf')
# train_loss curve
pp.figure()
if log:
pp.semilogy(x, train_loss)
else:
pp.plot(x, train_loss)
pp.ylabel('Training Loss', fontsize='xx-large')
pp.xlim(xmin, xmax)
pp.ylim(0, loss_max)
pp.savefig(surf_file + '_1d_train_loss' + ('_log' if log else '') + '.pdf',
dpi=300, bbox_inches='tight', format='pdf')
# train_err curve
pp.figure()
pp.plot(x, 100 - train_acc)
pp.xlim(xmin, xmax)
pp.ylim(0, 100)
pp.ylabel('Training Error', fontsize='xx-large')
pp.savefig(surf_file + '_1d_train_err.pdf', dpi=300, bbox_inches='tight', format='pdf')
if show: pp.show()
f.close()
def plot_1d_loss_err_repeat(prefix, idx_min=1, idx_max=10, xmin=-1.0, xmax=1.0,
loss_max=5, show=False):
"""
Plotting multiple 1D loss surface with different directions in one figure.
"""
fig, ax1 = pp.subplots()
ax2 = ax1.twinx()
for idx in range(idx_min, idx_max + 1):
# The file format should be prefix_{idx}.h5
f = h5py.File(prefix + '_' + str(idx) + '.h5','r')
x = f['xcoordinates'][:]
train_loss = f['train_loss'][:]
train_acc = f['train_acc'][:]
test_loss = f['test_loss'][:]
test_acc = f['test_acc'][:]
xmin = xmin if xmin != -1.0 else min(x)
xmax = xmax if xmax != 1.0 else max(x)
tr_loss, = ax1.plot(x, train_loss, 'b-', label='Training loss', linewidth=1)
te_loss, = ax1.plot(x, test_loss, 'b--', label='Testing loss', linewidth=1)
tr_acc, = ax2.plot(x, train_acc, 'r-', label='Training accuracy', linewidth=1)
te_acc, = ax2.plot(x, test_acc, 'r--', label='Testing accuracy', linewidth=1)
pp.xlim(xmin, xmax)
ax1.set_ylabel('Loss', color='b', fontsize='xx-large')
ax1.tick_params('y', colors='b', labelsize='x-large')
ax1.tick_params('x', labelsize='x-large')
ax1.set_ylim(0, loss_max)
ax2.set_ylabel('Accuracy', color='r', fontsize='xx-large')
ax2.tick_params('y', colors='r', labelsize='x-large')
ax2.set_ylim(0, 100)
pp.savefig(prefix + '_1d_loss_err_repeat.pdf', dpi=300, bbox_inches='tight', format='pdf')
if show: pp.show()
def plot_1d_eig_ratio(surf_file, xmin=-1.0, xmax=1.0, val_1='min_eig', val_2='max_eig', ymax=1, show=False):
print('------------------------------------------------------------------')
print('plot_1d_eig_ratio')
print('------------------------------------------------------------------')
f = h5py.File(surf_file,'r')
x = f['xcoordinates'][:]
Z1 = np.array(f[val_1][:])
Z2 = np.array(f[val_2][:])
abs_ratio = np.absolute(np.divide(Z1, Z2))
pp.plot(x, abs_ratio)
pp.xlim(xmin, xmax)
pp.ylim(0, ymax)
pp.savefig(surf_file + '_1d_eig_abs_ratio.pdf', dpi=300, bbox_inches='tight', format='pdf')
ratio = np.divide(Z1, Z2)
pp.plot(x, ratio)
pp.xlim(xmin, xmax)
pp.ylim(0, ymax)
pp.savefig(surf_file + '_1d_eig_ratio.pdf', dpi=300, bbox_inches='tight', format='pdf')
f.close()
if show: pp.show()
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Plott 1D loss and error curves')
parser.add_argument('--surf_file', '-f', default='', help='The h5 file contains loss values')
parser.add_argument('--log', action='store_true', default=False, help='logarithm plot')
parser.add_argument('--xmin', default=-1, type=float, help='xmin value')
parser.add_argument('--xmax', default=1, type=float, help='xmax value')
parser.add_argument('--loss_max', default=5, type=float, help='ymax value')
parser.add_argument('--show', action='store_true', default=False, help='show plots')
parser.add_argument('--prefix', default='', help='The common prefix for surface files')
parser.add_argument('--idx_min', default=1, type=int, help='min index for the surface file')
parser.add_argument('--idx_max', default=10, type=int, help='max index for the surface file')
args = parser.parse_args()
if args.prefix:
plot_1d_loss_err_repeat(args.prefix, args.idx_min, args.idx_max,
args.xmin, args.xmax, args.loss_max, args.show)
else:
plot_1d_loss_err(args.surf_file, args.xmin, args.xmax, args.loss_max, args.log, args.show)