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
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
|
import argparse
import copy
import csv
import os.path as osp
import random
import numpy as np
import yaml
from torch_geometric.graphgym.utils.comp_budget import match_baseline_cfg
from torch_geometric.graphgym.utils.io import (
makedirs_rm_exist,
string_to_python,
)
random.seed(123)
def parse_args():
"""Parses the arguments."""
parser = argparse.ArgumentParser()
parser.add_argument('--config', dest='config',
help='the base configuration file used for edit',
default=None, type=str)
parser.add_argument('--grid', dest='grid',
help='configuration file for grid search',
required=True, type=str)
parser.add_argument('--sample_alias', dest='sample_alias',
help='configuration file for sample alias',
default=None, required=False, type=str)
parser.add_argument('--sample_num', dest='sample_num',
help='Number of random samples in the space',
default=10, type=int)
parser.add_argument('--out_dir', dest='out_dir',
help='output directory for generated config files',
default='configs', type=str)
parser.add_argument(
'--config_budget', dest='config_budget',
help='the base configuration file used for matching computation',
default=None, type=str)
return parser.parse_args()
def get_fname(string):
if string is not None:
return string.split('/')[-1].split('.')[0]
else:
return 'default'
def grid2list(grid):
list_in = [[]]
for grid_temp in grid:
list_out = []
for val in grid_temp:
for list_temp in list_in:
list_out.append(list_temp + [val])
list_in = list_out
return list_in
def lists_distance(l1, l2):
assert len(l1) == len(l2)
dist = 0
for i in range(len(l1)):
if l1[i] != l2[i]:
dist += 1
return dist
def grid2list_sample(grid, sample=10):
configs = []
while len(configs) < sample:
config = []
for grid_temp in grid:
config.append(random.choice(grid_temp))
if config not in configs:
configs.append(config)
return configs
def load_config(fname):
if fname is not None:
with open(fname) as f:
return yaml.load(f, Loader=yaml.FullLoader)
else:
return {}
def load_search_file(fname):
with open(fname) as f:
out_raw = csv.reader(f, delimiter=' ')
outs = []
out = []
for row in out_raw:
if '#' in row:
continue
elif len(row) > 0:
assert len(row) == 3, \
'Exact 1 space between each grid argument file' \
'And no spaces within each argument is allowed'
out.append(row)
else:
if len(out) > 0:
outs.append(out)
out = []
if len(out) > 0:
outs.append(out)
return outs
def load_alias_file(fname):
with open(fname) as f:
file = csv.reader(f, delimiter=' ')
for line in file:
break
return line
def exclude_list_id(list, id):
return [list[i] for i in range(len(list)) if i != id]
def gen_grid(args, config, config_budget={}):
task_name = f'{get_fname(args.config)}_grid_{get_fname(args.grid)}'
fname_start = get_fname(args.config)
out_dir = f'{args.out_dir}/{task_name}'
makedirs_rm_exist(out_dir)
config['out_dir'] = osp.join(config['out_dir'], task_name)
outs = load_search_file(args.grid)
for i, out in enumerate(outs):
vars_label = [row[0].split('.') for row in out]
vars_alias = [row[1] for row in out]
vars_value = grid2list([string_to_python(row[2]) for row in out])
if i == 0:
print(f'Variable label: {vars_label}')
print(f'Variable alias: {vars_alias}')
for vars in vars_value:
config_out = config.copy()
fname_out = fname_start
for id, var in enumerate(vars):
if len(vars_label[id]) == 1:
config_out[vars_label[id][0]] = var
elif len(vars_label[id]) == 2:
if vars_label[id][0] in config_out: # if key1 exist
config_out[vars_label[id][0]][vars_label[id][1]] = var
else:
config_out[vars_label[id][0]] = {
vars_label[id][1]: var
}
else:
raise ValueError('Only 2-level config files are supported')
var_repr = str(var).strip("[]").strip("''")
fname_out += f'-{vars_alias[id]}={var_repr}'
if len(config_budget) > 0:
config_out = match_baseline_cfg(config_out, config_budget)
with open(f'{out_dir}/{fname_out}.yaml', 'w') as f:
yaml.dump(config_out, f, default_flow_style=False)
print(f'{len(vars_value)} configurations saved to: {out_dir}')
def gen_grid_sample(args, config, config_budget={}, compare_alias_list=[]):
task_name = f'{get_fname(args.config)}_grid_{get_fname(args.grid)}'
fname_start = get_fname(args.config)
out_dir = f'{args.out_dir}/{task_name}'
makedirs_rm_exist(out_dir)
config['out_dir'] = osp.join(config['out_dir'], task_name)
outs = load_search_file(args.grid)
counts = []
for out in outs:
vars_grid = [string_to_python(row[2]) for row in out]
count = 1
for var in vars_grid:
count *= len(var)
counts.append(count)
counts = np.array(counts)
print('Total size of each chunk of experiment space:', counts)
counts = counts / np.sum(counts)
counts = np.round(counts * args.sample_num)
counts[0] += args.sample_num - np.sum(counts)
print('Total sample size of each chunk of experiment space:', counts)
for i, out in enumerate(outs):
vars_label = [row[0].split('.') for row in out]
vars_alias = [row[1] for row in out]
if i == 0:
print(f'Variable label: {vars_label}')
print(f'Variable alias: {vars_alias}')
vars_grid = [string_to_python(row[2]) for row in out]
for alias in compare_alias_list:
alias_id = vars_alias.index(alias)
vars_grid_select = copy.deepcopy(vars_grid[alias_id])
vars_grid[alias_id] = [vars_grid[alias_id][0]]
vars_value = grid2list_sample(vars_grid, counts[i])
vars_value_new = []
for vars in vars_value:
for grid in vars_grid_select:
vars[alias_id] = grid
vars_value_new.append(copy.deepcopy(vars))
vars_value = vars_value_new
vars_grid[alias_id] = vars_grid_select
for vars in vars_value:
config_out = config.copy()
fname_out = fname_start + f'-sample={vars_alias[alias_id]}'
for id, var in enumerate(vars):
if len(vars_label[id]) == 1:
config_out[vars_label[id][0]] = var
elif len(vars_label[id]) == 2:
if vars_label[id][0] in config_out: # if key1 exist
config_out[vars_label[id][0]][vars_label[id]
[1]] = var
else:
config_out[vars_label[id][0]] = {
vars_label[id][1]: var
}
else:
raise ValueError(
'Only 2-level config files are supported')
var_repr = str(var).strip("[]").strip("''")
fname_out += f'-{vars_alias[id]}={var_repr}'
if len(config_budget) > 0:
config_out = match_baseline_cfg(config_out, config_budget,
verbose=False)
with open(f'{out_dir}/{fname_out}.yaml', "w") as f:
yaml.dump(config_out, f, default_flow_style=False)
print(f'Chunk {i + 1}/{len(outs)}: '
f'Perturbing design dimension {alias}, '
f'{len(vars_value)} configurations saved to: {out_dir}')
args = parse_args()
config = load_config(args.config)
config_budget = load_config(args.config_budget)
if args.sample_alias is None:
gen_grid(args, config, config_budget)
else:
alias_list = load_alias_file(args.sample_alias)
gen_grid_sample(args, config, config_budget, alias_list)
|
