-
Notifications
You must be signed in to change notification settings - Fork 0
/
speedtest.py
133 lines (109 loc) · 4.92 KB
/
speedtest.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
import numpy as np
import matplotlib.pyplot as plt
import topnetv2
import imageio
import timeit
import tensorflow as tf
def TopBackpropSpeedtest(f, TopLoss, a, lr, seconds):
# Initialize TensorFlow
tf.reset_default_graph()
# Create input variable initialized with image values
x = tf.get_variable("X", initializer=np.array(f), trainable=True)
# Compute persistence
bad_top = TopLoss(x)
# Compute loss
loss = a * bad_top + (1 - a) * tf.losses.mean_squared_error(f, x)
# Optimization
opt = tf.train.AdamOptimizer(learning_rate=lr)
# Train it
train = opt.minimize(loss)
# Training!
init = tf.global_variables_initializer()
start_time = timeit.default_timer()
times, losses = [], []
with tf.Session() as sess:
sess.run(init)
while True:
losses.append(sess.run([train, loss])[1])
times.append(timeit.default_timer() - start_time)
if times[-1] > seconds:
break
pred = sess.run(tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, "X")[0])
return {"pred": pred, "times": times, "losses": losses}
if __name__ == '__main__':
# Grab and normalize blobs image.
f = -1 * np.asarray(imageio.imread('Data/blobs.png')[:256, :256, 0], dtype=np.float32)
f -= f.min()
f /= f.max()
f *= 255
# Create random image
f_rand = np.random.randint(0, 255, (256, 256)).astype('float32')
# Parameters
hom_dim = 0
card = 10000
bad_pers = [float('-inf'), float('inf'), 50, float('inf')]
lr = 5e-2
a = (1 - 1 / np.prod(f.shape))
seconds = 600
kernel_size = 4
pool_mode = 'simplex'
def update_func(grad_dgm, dgm, cof, x):
bsm, dsm = topnetv2.compute_dgm_grad(grad_dgm, cof, x)
return dsm
def VanillaTopLoss(x):
dgm = topnetv2.Cubical(x, card, hom_dim, update_func)[0]
bad_top = topnetv2.SqPersInRegion(dgm, bad_pers)
return bad_top
vanilla_L2 = TopBackpropSpeedtest(f, VanillaTopLoss, a, lr, seconds)
rand_vanilla_L2 = TopBackpropSpeedtest(f_rand, VanillaTopLoss, a, lr, seconds)
def VanillaTopLoss(x):
dgm = topnetv2.Cubical(x, card, hom_dim, update_func)[0]
bad_top = topnetv2.AbsPersInRegion(dgm, bad_pers)
return bad_top
vanilla_L1 = TopBackpropSpeedtest(f, VanillaTopLoss, a, lr, seconds)
rand_vanilla_L1 = TopBackpropSpeedtest(f_rand, VanillaTopLoss, a, lr, seconds)
def SpawnTopLoss(x):
x_noisy = topnetv2.UniformNoise(x, 50)
x_down = topnetv2.Spool(x_noisy, kernel_size, pool_mode)[0]
dgm = topnetv2.Cubical(x_down, card, hom_dim, update_func)[0]
bad_top = topnetv2.AbsPersInRegion(dgm, bad_pers)
return bad_top
spawn_noise = TopBackpropSpeedtest(f, SpawnTopLoss, a, lr, seconds)
rand_spawn_noise = TopBackpropSpeedtest(f_rand, SpawnTopLoss, a, lr, seconds)
def SpawnTopLoss(x):
x_down = topnetv2.Spool(x, kernel_size, pool_mode)[0]
dgm = topnetv2.Cubical(x_down, card, hom_dim, update_func)[0]
bad_top = topnetv2.AbsPersInRegion(dgm, bad_pers)
return bad_top
spawn_no_noise = TopBackpropSpeedtest(f, SpawnTopLoss, a, lr, seconds)
rand_spawn_no_noise = TopBackpropSpeedtest(f_rand, SpawnTopLoss, a, lr, seconds)
plt.figure(figsize=(10, 10))
plt.plot(spawn_noise["times"],
(spawn_noise["losses"][0] - spawn_noise["losses"]) / spawn_noise["losses"][0])
plt.plot(spawn_no_noise["times"],
(spawn_no_noise["losses"][0] - spawn_no_noise["losses"]) / spawn_no_noise["losses"][0])
plt.plot(vanilla_L2["times"],
(vanilla_L2["losses"][0] - vanilla_L2["losses"]) / vanilla_L2["losses"][0])
plt.plot(vanilla_L1["times"],
(vanilla_L1["losses"][0] - vanilla_L1["losses"]) / vanilla_L1["losses"][0])
plt.legend(["STUMP w/ noise", "STUMP w/o noise", "Vanilla W2", "Vanilla W1"], fontsize=18)
plt.xlabel('Seconds', fontsize=20)
plt.ylabel('Percentage Reduction of Starting Loss', fontsize=20)
plt.xticks(fontsize=18)
plt.yticks(fontsize=18)
plt.show()
plt.figure(figsize=(10, 10))
plt.plot(rand_spawn_noise["times"],
(rand_spawn_noise["losses"][0] - rand_spawn_noise["losses"]) / rand_spawn_noise["losses"][0])
plt.plot(rand_spawn_no_noise["times"],
(rand_spawn_no_noise["losses"][0] - rand_spawn_no_noise["losses"]) / rand_spawn_no_noise["losses"][0])
plt.plot(rand_vanilla_L2["times"],
(rand_vanilla_L2["losses"][0] - rand_vanilla_L2["losses"]) / rand_vanilla_L2["losses"][0])
plt.plot(rand_vanilla_L1["times"],
(rand_vanilla_L1["losses"][0] - rand_vanilla_L1["losses"]) / rand_vanilla_L1["losses"][0])
plt.legend(["STUMP w/ noise", "STUMP w/o noise", "Vanilla W2", "Vanilla W1"], fontsize=18)
plt.xlabel('Seconds', fontsize=20)
plt.ylabel('Percentage Reduction of Starting Loss', fontsize=20)
plt.xticks(fontsize=18)
plt.yticks(fontsize=18)
plt.show()