-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathdebug.py
164 lines (131 loc) · 5.48 KB
/
debug.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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
Useful debugging functions.
"""
import numpy as np
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
from layers import SpectralConv
from paths import DEBUG_DIR_MNIST_012, DEBUG_DIR_MNIST_rot, DEBUG_DIR_ETH80, FIGURES_DIR
from plot import show_filters_single_window, colorbar
titles = ['PyTorch', 'Theano']
size_dict = {DEBUG_DIR_MNIST_012 + 'constant_weights/':20, DEBUG_DIR_MNIST_012 + 'pretrained_weights/':20, DEBUG_DIR_MNIST_rot:26, DEBUG_DIR_ETH80:50}
def get_min_max(images):
"""Return the min and max pixel values for a list of images."""
vmin = []
vmax = []
for i in images:
vmin.append(np.min(i))
vmax.append(np.max(i))
return np.min(vmin), np.max(vmax)
def plot_pytorch_theano_image(images, dir, name='temp'):
"""Plot and save image comparison between PyTorch and Theano framework."""
size = size_dict[dir]
fig = plt.figure(figsize=(12,5))
vmin, vmax = get_min_max(images)
delta = 1e-0
if vmin==vmax:
vmin -= delta
for i in range(2):
ax = fig.add_subplot(1, 2, i+1)
im = ax.imshow(images[i].astype(int).reshape(size,size), cmap='jet', vmin=vmin, vmax=vmax)
ax.set_axis_off()
colorbar(im)
plt.tight_layout(h_pad=1)
plt.savefig(dir + 'figures/' + name + '.pdf')
#plt.show()
def plot_pytorch_theano_image_diff(images, dir, name='temp'):
"""Plot and save image comparison between PyTorch and Theano framework."""
size = size_dict[dir]
fig = plt.figure(figsize=(6,5))
diff = np.abs(images[0]-images[1])
ax = fig.add_subplot(1, 1, 1)
im = ax.imshow(diff.reshape(size,size), cmap='jet')
ax.set_axis_off()
colorbar(im)
plt.tight_layout(h_pad=1)
plt.savefig(dir + 'figures/' + name + '.pdf')
def plot_pytorch_theano_statistic(images, shape, dir, name='temp'):
"""Plot and save feature vector (from statistical layer) comparison between PyTorch and Theano framework."""
h, w = shape
if w==28:
fig = plt.figure(figsize=(10,3))
elif w==24:
fig = plt.figure(figsize=(8,3))
else:
fig = plt.figure(figsize=(6,5))
vmin, vmax = get_min_max(images)
for i in range(2):
ax = fig.add_subplot(1, 2, i+1)
im = ax.imshow(images[i].reshape(h,w), cmap='jet', vmin=vmin, vmax=vmax)
ax.set_axis_off()
colorbar(im)
plt.tight_layout(h_pad=1)
plt.savefig(dir + 'figures/' + name + '.pdf')
#plt.show()
def plot_pytorch_theano_statistic_diff(images, shape, dir, name='temp'):
"""Plot and save feature vector (from statistical layer) comparison between PyTorch and Theano framework."""
h, w = shape
if w==28:
#MNIST_ROT
fig = plt.figure(figsize=(5,4))
elif w==24:
#ETH80
fig = plt.figure(figsize=(5,4))
else:
fig = plt.figure(figsize=(3,5))
diff = np.abs(images[0]-images[1])
ax = fig.add_subplot(1, 1, 1)
im = ax.imshow(diff.reshape(h,w), cmap='jet')
ax.set_axis_off()
colorbar(im)
plt.tight_layout(h_pad=1)
plt.savefig(dir + 'figures/' + name + '.pdf')
#plt.show()
def plot_pytorch_theano_filter_operator(images, dir, name='temp'):
"""Plot and save filter operator (from spectral conv layer) comparison between PyTorch and Theano framework."""
size = size_dict[dir]
fig = plt.figure(figsize=(15,10))
for i in range(2):
ax = fig.add_subplot(1, 2, i+1)
im = ax.imshow(images[i].reshape(size**2,size**2), cmap='jet')
ax.set_title(titles[i])
ax.set_axis_off()
colorbar(im)
plt.suptitle('filter operator')
plt.tight_layout(h_pad=1)
plt.savefig(dir + 'figures/' + name + '.pdf')
plt.show()
def init_weights_constant(model, constant=1):
"""Initialize weights of the model with the constant passed in argument."""
for m in model.modules():
if isinstance(m, SpectralConv):
nn.init.constant_(m.alpha.weight, constant)
nn.init.constant_(m.beta.weight, constant)
elif isinstance(m, nn.Sequential):
for m in m.modules():
if isinstance(m, nn.Linear):
nn.init.constant_(m.weight, constant)
nn.init.constant_(m.bias, constant)
def display_weights(model):
"""Display the weights of the model."""
for name, param in model.named_parameters():
print('- - - - - - - - - - - - - - - - - - - \n{} \n\n{} \n\n{} \n- - - - - - - - - - - - - - - - - - - '.format(name, param, param.size()))
def plot_dataset():
"""Plot the training, validation and testing datasets for the mnist_rot, mnist_trans and eth80 datasets."""
num_images = 6
dataset_shapes = {'mnist_rot':26, 'mnist_trans':30, 'eth80':50}
for dataset_name in ['mnist_rot', 'mnist_trans', 'eth80']:
dataset_shape = dataset_shapes[dataset_name]
fig = plt.figure(figsize=(10,7))
fig.suptitle(t='train, valid, test images for ' + dataset_name)
for i, d in enumerate(['train', 'val', 'test']):
data = np.load('{}{}/dataset/{}_{}_signals.npy'.format(SAVED_DATA, dataset_name, dataset_name, d))
labels = np.load('{}{}/dataset/{}_{}_labels.npy'.format(SAVED_DATA, dataset_name, dataset_name, d))
for j in range(num_images):
ax = fig.add_subplot(3, num_images, i*6 + j+1) # this line adds sub-axes
im = ax.imshow(data[j].reshape(dataset_shape,dataset_shape), cmap='jet')
ax.set_axis_off()
plt.show()