-
Notifications
You must be signed in to change notification settings - Fork 3
/
utils.py
212 lines (170 loc) · 5.86 KB
/
utils.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
import os
import matplotlib.pyplot as plt
import numpy as np
import torch
from torch.nn import init
from types import SimpleNamespace
import json
from tqdm import tqdm
import PIL.Image
from copy import deepcopy
def save_image_grid(img, fname, drange, grid_size, ids_to_prepend=None):
lo, hi = drange
img = np.asarray(img, dtype=np.float32)
img = (img - lo) * (255 / (hi - lo))
img = np.rint(img).clip(0, 255).astype(np.uint8)
gw, gh = grid_size
_N, C, H, W = img.shape
img = img.reshape(gh, gw, C, H, W)
img = img.transpose(0, 3, 1, 4, 2)
img = img.reshape(gh * H, gw * W, C)
if ids_to_prepend is not None:
# Load number images
nrow = gh
assert len(ids_to_prepend) == nrow
num_ims = np.concatenate([plt.imread(f'results/number_imgs/{num}.png') for num in ids_to_prepend], 0)
num_ims = np.rint(num_ims * 255).clip(0, 255).astype(np.uint8)
img = np.concatenate([num_ims, img], 1)
assert C in [1, 3]
if C == 1:
PIL.Image.fromarray(img[:, :, 0], 'L').save(fname)
if C == 3:
PIL.Image.fromarray(img, 'RGB').save(fname)
def torch_mvn_logp(x, m, C):
"""
input
x - (N, k) data matrix torch.Tensor
m - (1, k) mean torch.Tensor
C - (k, k) covariance torch.Tensor
output
(N, ) logp = N(x; m, C), torch.Tensor
"""
assert len(x.shape) == 2
assert x.shape[1] == m.shape[-1]
assert m.shape[0] == 1
assert m.shape[1] == C.shape[0] == C.shape[1]
k = x.shape[1]
Z = -(k/2) * np.log(2*np.pi) - (1/2) * torch.logdet(C)
Cinv = torch.inverse(C)
s = -(1/2) * (((x - m) @ Cinv) * (x - m)).sum(-1)
return Z + s
def euclidean_dist(x, y):
# x: N x D
# y: M x D
n = x.size(0)
m = y.size(0)
d = x.size(1)
assert d == y.size(1)
return torch.pow(x.unsqueeze(1).expand(n, m, d) - y.unsqueeze(0).expand(n, m, d), 2).sum(2)
def save_checkpoint(args, state):
torch.save(state, args.ckpt)
def maybe_load_checkpoint(args):
if args.resume and os.path.exists(args.ckpt):
return torch.load(args.ckpt)
def get_item(v):
if type(v) == int or np.isscalar(v):
return v
elif isinstance(v, torch.Tensor):
return v.item()
else:
raise
def gaussian_logp(mean, logstd, x, detach=False):
"""
lnL = -1/2 * { ln|Var| + ((X - Mu)^T)(Var^-1)(X - Mu) + kln(2*PI) }
k = 1 (Independent)
Var = logstd ** 2
"""
c = np.log(2 * np.pi)
v = -0.5 * (logstd * 2. + ((x - mean) ** 2) / torch.exp(logstd * 2.) + c)
if detach:
v = v.detach()
return v
class Stats:
def __init__(self):
self.items = []
def add(self, x):
self.items.append(x)
def avg(self):
return np.mean(self.items)
def save_args(args, fp):
if type(args) != dict:
args = vars(args)
args = deepcopy(args)
# Remove Known Wrong Types (that don't need to be saved)
if 'device' in args:
args.pop('device')
if 'print' in args:
args.pop('print')
# Check all values are basic types
assert all(map(lambda v: type(v) in [int, float, str, bool], args.values()))
return json.dump(args, open(fp, 'w'),
sort_keys=True, indent=4)
def load_args(fp):
return SimpleNamespace(**json.load(open(fp, 'r')))
def generate_n_samples(generator, generator_args, device, N):
all_samples = []
for start in tqdm(range(0, N, generator_args.num_gen_images), desc='generate_n_samples'):
with torch.no_grad():
fake = generator(torch.randn(
generator_args.num_gen_images, generator_args.nz, 1, 1, device=device))
all_samples.append(fake)
return torch.cat(all_samples, 0)
def plot_kde(samples, epoch, name, title='', cmap='Blues', save_path=None):
samples = samples.cpu().numpy()
sns.set(font_scale=2)
f, ax = plt.subplots(figsize=(4, 4))
sns.kdeplot(samples[:, 0], samples[:, 1], cmap=cmap,
ax=ax, n_levels=20, shade=True)
plt.xlim([-5, 5])
plt.ylim([-5, 5])
plt.axis('off')
plt.title(title)
if save_path is not None:
plt.savefig(os.path.join(save_path, '{}_{}.pdf'.format(name, epoch)))
plt.show()
def weights_init(model, method='N02', std=0.02):
for m in model.modules():
classname = m.__class__.__name__
# if (classname.find('Conv') != -1 or
# classname.find('Linear') != -1 or
# classname.find('Embedding') != -1):
if (isinstance(m, torch.nn.Conv2d)
or isinstance(m, torch.nn.Linear)
or isinstance(m, torch.nn.Embedding)):
print(f'Initializing: {classname}')
if method == 'ortho':
init.orthogonal_(m.weight)
elif method == 'N02':
init.normal_(m.weight, 0, 0.02)
elif method == 'N':
init.normal_(m.weight, 0, std)
elif method in ['glorot', 'xavier']:
init.xavier_uniform_(m.weight)
else:
print('Init style not recognized...')
elif classname.find('BatchNorm') != -1:
m.weight.data.normal_(1.0, 0.02)
m.bias.data.fill_(0)
def tonp(x):
if not isinstance(x, np.ndarray):
x = x.detach().cpu().numpy()
return x
def mkdir(p):
if not os.path.exists(p):
os.makedirs(p)
def prepare_random_indices(x, eval_size):
all_indices = np.arange(len(x))
np.random.shuffle(all_indices)
return torch.from_numpy(all_indices[:eval_size]).to(x.device)
def expand_stackedmnist(x):
N, three, W, H = x.shape
y = []
for _x in x:
_y = torch.zeros((3, W*2, H*2))
_y[:, :W, :H] = _x
_y[:, W:, :H] = _x[0].unsqueeze(0).repeat(3, 1, 1)
_y[:, :W, H:] = _x[1].unsqueeze(0).repeat(3, 1, 1)
_y[:, W:, W:] = _x[2].unsqueeze(0).repeat(3, 1, 1)
y.append(_y)
y = torch.stack(y)
return y