|
1
|
|
|
def warn(*args, **kwargs): |
|
2
|
|
|
pass |
|
3
|
|
|
|
|
4
|
|
|
|
|
5
|
|
|
import warnings |
|
6
|
|
|
|
|
7
|
|
|
warnings.warn = warn |
|
8
|
|
|
|
|
9
|
|
|
import os |
|
10
|
|
|
import gc |
|
11
|
|
|
import glob |
|
12
|
|
|
|
|
13
|
|
|
import numpy as np |
|
14
|
|
|
import pandas as pd |
|
15
|
|
|
from tqdm import tqdm |
|
16
|
|
|
|
|
17
|
|
|
import matplotlib as mpl |
|
18
|
|
|
import matplotlib.pyplot as plt |
|
19
|
|
|
|
|
20
|
|
|
plt.rcParams["figure.facecolor"] = "w" |
|
21
|
|
|
|
|
22
|
|
|
mpl.use("agg") |
|
23
|
|
|
|
|
24
|
|
|
from gradient_free_optimizers.optimizers.core_optimizer.converter import Converter |
|
25
|
|
|
|
|
26
|
|
|
|
|
27
|
|
|
def plot_search_paths( |
|
28
|
|
|
path, |
|
29
|
|
|
optimizer, |
|
30
|
|
|
opt_para, |
|
31
|
|
|
n_iter_max, |
|
32
|
|
|
objective_function, |
|
33
|
|
|
search_space, |
|
34
|
|
|
initialize, |
|
35
|
|
|
random_state, |
|
36
|
|
|
title, |
|
37
|
|
|
): |
|
38
|
|
|
if opt_para == {}: |
|
39
|
|
|
show_opt_para = False |
|
40
|
|
|
else: |
|
41
|
|
|
show_opt_para = True |
|
42
|
|
|
|
|
43
|
|
|
opt = optimizer( |
|
44
|
|
|
search_space, initialize=initialize, random_state=random_state, **opt_para |
|
45
|
|
|
) |
|
46
|
|
|
opt.search( |
|
47
|
|
|
objective_function, |
|
48
|
|
|
n_iter=n_iter_max, |
|
49
|
|
|
# memory=False, |
|
50
|
|
|
verbosity=False, |
|
51
|
|
|
) |
|
52
|
|
|
|
|
53
|
|
|
conv = Converter(search_space) |
|
54
|
|
|
for n_iter in tqdm(range(1, n_iter_max + 1)): |
|
55
|
|
|
|
|
56
|
|
|
def objective_function_np(args): |
|
57
|
|
|
params = {} |
|
58
|
|
|
for i, para_name in enumerate(search_space): |
|
59
|
|
|
params[para_name] = args[i] |
|
60
|
|
|
|
|
61
|
|
|
return objective_function(params) |
|
62
|
|
|
|
|
63
|
|
|
plt.figure(figsize=(7, 7)) |
|
64
|
|
|
plt.set_cmap("jet_r") |
|
65
|
|
|
# jet_r |
|
66
|
|
|
|
|
67
|
|
|
x_all, y_all = search_space["x0"], search_space["x1"] |
|
68
|
|
|
xi, yi = np.meshgrid(x_all, y_all) |
|
69
|
|
|
zi = objective_function_np((xi, yi)) |
|
70
|
|
|
|
|
71
|
|
|
zi = np.rot90(zi, k=1) |
|
72
|
|
|
|
|
73
|
|
|
plt.imshow( |
|
74
|
|
|
zi, |
|
75
|
|
|
alpha=0.15, |
|
76
|
|
|
# interpolation="antialiased", |
|
77
|
|
|
# vmin=z.min(), |
|
78
|
|
|
# vmax=z.max(), |
|
79
|
|
|
# origin="lower", |
|
80
|
|
|
extent=[x_all.min(), x_all.max(), y_all.min(), y_all.max()], |
|
81
|
|
|
) |
|
82
|
|
|
|
|
83
|
|
|
for n, opt_ in enumerate(opt.optimizers): |
|
84
|
|
|
n_optimizers = len(opt.optimizers) |
|
85
|
|
|
n_iter_tmp = int(n_iter / n_optimizers) |
|
86
|
|
|
n_iter_mod = n_iter % n_optimizers |
|
87
|
|
|
|
|
88
|
|
|
if n_iter_mod > n: |
|
89
|
|
|
n_iter_tmp += 1 |
|
90
|
|
|
if n_iter_tmp == 0: |
|
91
|
|
|
continue |
|
92
|
|
|
|
|
93
|
|
|
pos_list = np.array(opt_.pos_new_list) |
|
94
|
|
|
score_list = np.array(opt_.score_new_list) |
|
95
|
|
|
|
|
96
|
|
|
# print("\n pos_list \n", pos_list, "\n") |
|
97
|
|
|
# print("\n score_list \n", score_list, "\n") |
|
98
|
|
|
|
|
99
|
|
|
if len(pos_list) == 0: |
|
100
|
|
|
continue |
|
101
|
|
|
|
|
102
|
|
|
values_list = conv.positions2values(pos_list) |
|
103
|
|
|
values_list = np.array(values_list) |
|
104
|
|
|
|
|
105
|
|
|
plt.plot( |
|
106
|
|
|
values_list[:n_iter_tmp, 0], |
|
107
|
|
|
values_list[:n_iter_tmp, 1], |
|
108
|
|
|
linestyle="--", |
|
109
|
|
|
marker=",", |
|
110
|
|
|
color="black", |
|
111
|
|
|
alpha=0.33, |
|
112
|
|
|
label=n, |
|
113
|
|
|
linewidth=0.5, |
|
114
|
|
|
) |
|
115
|
|
|
plt.scatter( |
|
116
|
|
|
values_list[:n_iter_tmp, 0], |
|
117
|
|
|
values_list[:n_iter_tmp, 1], |
|
118
|
|
|
c=score_list[:n_iter_tmp], |
|
119
|
|
|
marker="H", |
|
120
|
|
|
s=15, |
|
121
|
|
|
vmin=np.amin(score_list[:n_iter_tmp]), |
|
122
|
|
|
vmax=np.amax(score_list[:n_iter_tmp]), |
|
123
|
|
|
label=n, |
|
124
|
|
|
edgecolors="black", |
|
125
|
|
|
linewidth=0.3, |
|
126
|
|
|
) |
|
127
|
|
|
|
|
128
|
|
|
plt.xlabel("x") |
|
129
|
|
|
plt.ylabel("y") |
|
130
|
|
|
|
|
131
|
|
|
nth_iteration = "\n\nnth Iteration: " + str(n_iter) |
|
132
|
|
|
opt_para_name = "" |
|
133
|
|
|
opt_para_value = "\n\n" |
|
134
|
|
|
|
|
135
|
|
|
if show_opt_para: |
|
136
|
|
|
opt_para_name += "\n Parameter:" |
|
137
|
|
|
for para_name, para_value in opt_para.items(): |
|
138
|
|
|
opt_para_name += "\n " + " " + para_name + ": " |
|
139
|
|
|
opt_para_value += "\n " + str(para_value) + " " |
|
140
|
|
|
|
|
141
|
|
|
if title == True: |
|
142
|
|
|
title_name = opt.name + "\n" + opt_para_name |
|
143
|
|
|
plt.title(title_name, loc="left") |
|
144
|
|
|
plt.title(opt_para_value, loc="center") |
|
145
|
|
|
elif isinstance(title, str): |
|
146
|
|
|
plt.title(title, loc="left") |
|
147
|
|
|
|
|
148
|
|
|
plt.title(nth_iteration, loc="right", fontsize=8) |
|
149
|
|
|
|
|
150
|
|
|
# plt.xlim((-101, 201)) |
|
151
|
|
|
# plt.ylim((-101, 201)) |
|
152
|
|
|
clb = plt.colorbar() |
|
153
|
|
|
clb.set_label("score", labelpad=-50, y=1.03, rotation=0) |
|
154
|
|
|
|
|
155
|
|
|
# plt.legend(loc="upper left", bbox_to_anchor=(-0.10, 1.2)) |
|
156
|
|
|
|
|
157
|
|
|
# plt.axis("off") |
|
158
|
|
|
|
|
159
|
|
|
if show_opt_para: |
|
160
|
|
|
plt.subplots_adjust(top=0.75) |
|
161
|
|
|
plt.tight_layout() |
|
162
|
|
|
|
|
163
|
|
|
# plt.margins(0, 0) |
|
164
|
|
|
plt.savefig( |
|
165
|
|
|
path + "/_plots/" + opt._name_ + "_" + "{0:0=3d}".format(n_iter) + ".jpg", |
|
166
|
|
|
dpi=150, |
|
167
|
|
|
pad_inches=0, |
|
168
|
|
|
# bbox_inches="tight", |
|
169
|
|
|
) |
|
170
|
|
|
|
|
171
|
|
|
plt.ioff() |
|
172
|
|
|
# Clear the current axes. |
|
173
|
|
|
plt.cla() |
|
174
|
|
|
# Clear the current figure. |
|
175
|
|
|
plt.clf() |
|
176
|
|
|
# Closes all the figure windows. |
|
177
|
|
|
plt.close("all") |
|
178
|
|
|
|
|
179
|
|
|
gc.collect() |
|
180
|
|
|
|
|
181
|
|
|
|
|
182
|
|
|
def search_path_gif( |
|
183
|
|
|
path, |
|
184
|
|
|
optimizer, |
|
185
|
|
|
opt_para, |
|
186
|
|
|
name, |
|
187
|
|
|
n_iter, |
|
188
|
|
|
objective_function, |
|
189
|
|
|
search_space, |
|
190
|
|
|
initialize, |
|
191
|
|
|
random_state=0, |
|
192
|
|
|
title=True, |
|
193
|
|
|
): |
|
194
|
|
|
path = os.path.join(os.getcwd(), path) |
|
195
|
|
|
|
|
196
|
|
|
print("\n\nname", name) |
|
197
|
|
|
plots_dir = path + "/_plots/" |
|
198
|
|
|
print("plots_dir", plots_dir) |
|
199
|
|
|
os.makedirs(plots_dir, exist_ok=True) |
|
200
|
|
|
|
|
201
|
|
|
plot_search_paths( |
|
202
|
|
|
path=path, |
|
203
|
|
|
optimizer=optimizer, |
|
204
|
|
|
opt_para=opt_para, |
|
205
|
|
|
n_iter_max=n_iter, |
|
206
|
|
|
objective_function=objective_function, |
|
207
|
|
|
search_space=search_space, |
|
208
|
|
|
initialize=initialize, |
|
209
|
|
|
random_state=random_state, |
|
210
|
|
|
title=title, |
|
211
|
|
|
) |
|
212
|
|
|
|
|
213
|
|
|
### ffmpeg |
|
214
|
|
|
framerate = str(n_iter / 10) |
|
215
|
|
|
# framerate = str(10) |
|
216
|
|
|
_framerate = " -framerate " + framerate + " " |
|
217
|
|
|
|
|
218
|
|
|
_opt_ = optimizer(search_space) |
|
219
|
|
|
_input = " -i " + path + "/_plots/" + str(_opt_._name_) + "_" + "%03d.jpg " |
|
220
|
|
|
_scale = " -vf scale=1200:-1:flags=lanczos " |
|
221
|
|
|
_output = os.path.join(path, name) |
|
222
|
|
|
|
|
223
|
|
|
ffmpeg_command = ( |
|
224
|
|
|
"ffmpeg -hide_banner -loglevel error -y" |
|
225
|
|
|
+ _framerate |
|
226
|
|
|
+ _input |
|
227
|
|
|
+ _scale |
|
228
|
|
|
+ _output |
|
229
|
|
|
) |
|
230
|
|
|
print("\n -----> ffmpeg_command \n", ffmpeg_command, "\n") |
|
231
|
|
|
print("create " + name) |
|
232
|
|
|
|
|
233
|
|
|
os.system(ffmpeg_command) |
|
234
|
|
|
|
|
235
|
|
|
### remove _plots |
|
236
|
|
|
rm_files = glob.glob(path + "/_plots/*.jpg") |
|
237
|
|
|
for f in rm_files: |
|
238
|
|
|
os.remove(f) |
|
239
|
|
|
os.rmdir(plots_dir) |
|
240
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
