/
exp4_robust_lr_time_2.py
141 lines (108 loc) · 4.58 KB
/
exp4_robust_lr_time_2.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
# -*- coding: utf-8 -*-
#############################################################
# Experiment for testing algorithm speed of four different lr
#############################################################
import numpy as np
import matplotlib.pyplot as plt
import time
import matplotlib.ticker as ticker
from PRW import ProjectionRobustWasserstein
from Optimization.riemann_adap import RiemmanAdaptive
import pickle
def T(x, d, dim=2):
assert dim <= d
assert dim >= 1
assert dim == int(dim)
return x + 2 * np.sign(x) * np.array(dim * [1] + (d - dim) * [0])
def fragmented_hypercube(n, d, dim):
assert dim <= d
assert dim >= 1
assert dim == int(dim)
a = (1. / n) * np.ones(n)
b = (1. / n) * np.ones(n)
# First measure : uniform on the hypercube
X = np.random.uniform(-1, 1, size=(n, d))
# Second measure : fragmentation
Y = T(np.random.uniform(-1, 1, size=(n, d)), d, dim)
return a, b, X, Y
ds = [25, 50, 100, 250, 500] # , 1000] # Dimensions for which to compute the SRW computation time
nb_ds = len(ds)
n = 100 # Number of points in the measures
k = 2 # Dimension parameter
reg = 0.2 # Entropic regularization strength
max_iter = 500 # Maximum number of iterations
max_iter_sinkhorn = 30 # Maximum number of iterations in Sinkhorn
threshold = 1e-3 # Stopping threshold
threshold_sinkhorn = 1e-3 # Stopping threshold in Sinkhorn
nb_exp = 50 # Number of experiments
lrs = [0.005, 0.01, 0.05, 0.1]
times_PRW = np.zeros((2, len(lrs), nb_exp, nb_ds))
np.random.seed(357)
tic = time.time()
tac = time.time()
if 1 == 1:
for t in range(nb_exp):
print('iter', t)
for ind_lr in range(len(lrs)):
for ind_d in range(nb_ds):
d = ds[ind_d]
lr = lrs[ind_lr]
# print(d)
a, b, X, Y = fragmented_hypercube(n, d, dim=2)
reg = 0.2
if d >= 250:
reg = 0.5
if lr > 0.01:
reg *= 10
#print('PRW(1)', lr)
algo = RiemmanAdaptive(reg=reg, step_size_0=None, max_iter=max_iter,
max_iter_sinkhorn=max_iter_sinkhorn,
threshold=threshold, threshold_sinkhorn=threshold_sinkhorn, use_gpu=False)
PRW = ProjectionRobustWasserstein(X, Y, a, b, algo, k)
tic = time.time()
PRW.run(0, lr=lr, beta=None)
tac = time.time()
times_PRW[0, ind_lr, t, ind_d] = tac - tic
#print('PRW(2)',lr)
algo = RiemmanAdaptive(reg=reg, step_size_0=None, max_iter=max_iter,
max_iter_sinkhorn=max_iter_sinkhorn,
threshold=threshold, threshold_sinkhorn=threshold_sinkhorn, use_gpu=False)
PRW = ProjectionRobustWasserstein(X, Y, a, b, algo, k)
tic = time.time()
PRW.run(1, lr=lr, beta=0.8)
tac = time.time()
times_PRW[1, ind_lr, t, ind_d] = tac - tic
with open('./results/exp4_robust_lr_time_2.pkl', 'wb') as f:
pickle.dump(times_PRW, f)
else:
with open('./results/exp4_robust_lr_time_2.pkl', 'rb') as f:
times_PRW = pickle.load(f)
line_styles = ['-', '-',':', ':']
captions = ['PRW (RGAS) ', 'PRW (RAGAS)']
for t in range(2):
plt.figure(figsize=(17, 9))
for ind_lr in range(len(lrs)):
cap = '%s lr=%.3f' % (captions[t], lrs[ind_lr])
time_t_lr = times_PRW[t, ind_lr, :, :]
times_mean = np.mean(time_t_lr, axis=0)
times_min = np.min(time_t_lr, axis=0)
times_max = np.max(time_t_lr, axis=0)
if ind_lr in [1, 3]:
mean, = plt.loglog(ds, times_mean, 'C%d' % (t + 1,), ls=line_styles[ind_lr], lw=6, ms=8, label=cap)
elif t == 0:
mean, = plt.loglog(ds, times_mean, ls=line_styles[ind_lr], c='m', lw=6, ms=8, label=cap)
elif t == 1:
mean, = plt.loglog(ds, times_mean, ls=line_styles[ind_lr], c='b', lw=6, ms=8, label=cap)
col = mean.get_color()
plt.fill_between(ds, times_min, times_max, facecolor=col, alpha=0.15)
plt.xlabel('Dimension', fontsize=25)
plt.ylabel('Execution time in seconds', fontsize=25)
plt.legend(loc='best', fontsize=25, handlelength=3)
plt.xticks(ds, fontsize=20)
plt.ylim((1e-3, 10))
# plt.yticks(fontsize=20)
plt.gca().xaxis.set_major_formatter(ticker.FormatStrFormatter('%0.0f'))
plt.grid(ls=':')
plt.savefig('figs/exp4_computation_time_lr_%d.png' % (t,))
plt.show()
plt.close()