-
Notifications
You must be signed in to change notification settings - Fork 0
/
runner.py
267 lines (228 loc) · 11.8 KB
/
runner.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
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
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
from cmaes import CMAESAlgorithm
from multiprocessing import Pool
from experimentdatabase import Database
from logger import Logger
import constraints_generator as cg
import os
import subprocess
from functools import reduce
from frozendict import frozendict
import psutil
log = Logger(name='runner')
def flat(l):
return [item for sublist in l for item in sublist]
class SlurmPool:
def __init__(self):
try:
os.makedirs("./scripts")
except:
None
self.run = open("./run.sh", "w")
self.run.write("#!/bin/bash\n")
self.run.write("export PATH=\"/home/inf116360/anaconda3/bin:$PATH\"\n")
def execute(self, cmd, arguments, script_filename):
sbatch = open("./scripts/%s.sh" % script_filename, "w")
sbatch.write("#!/bin/bash\n")
sbatch.write("#SBATCH -p idss-student")
sbatch.write("#SBATCH -c 1 --mem=1475\n")
sbatch.write("#SBATCH -t 24:00:00\n")
sbatch.write("#SBATCH -Q\n")
sbatch.write("date\n")
sbatch.write("hostname\n")
sbatch.write("echo %s %s\n" % (cmd, arguments))
sbatch.write("srun %s %s && srun rm \"./scripts/%s.sh\"\n" % (cmd, arguments, script_filename))
sbatch.close()
self.run.write("sbatch \"./scripts/%s.sh\"\n" % script_filename)
def close(self):
self.run.close()
ps = subprocess.Popen(["/bin/sh", "./run.sh"])
ps.wait()
class AlgorithmRunner:
@property
def sql(self):
return "SELECT count(*) FROM experiments WHERE constraints_generator=? AND margin=? AND sigma=? AND k=? AND n=? AND seed=? AND name=? AND clustering=? AND standardized=? AND (train_tp + train_fn)=?"
def check_if_table_is_empty(self, database: Database):
check_if_table_exists = database.engine.execute("SELECT count(*) FROM main.experiments").fetchone()[0]
return check_if_table_exists > 0
def filter_algorithms(self, experiments: iter, database: Database) -> list:
if not self.check_if_table_is_empty(database=database):
return experiments
seq_of_params = [self.convert_to_sql_params(experiment) for experiment in experiments]
db_experiments = [database.engine.execute(self.sql, params).fetchone()[0] for params in seq_of_params]
filtered = list()
existing = 0
for exists, algorithm in zip(db_experiments, experiments):
if exists == 0:
filtered.append(algorithm)
else:
existing = existing + 1
log.info("Number of existing experiments: {}/{}".format(existing, len(experiments)))
return filtered
def convert_to_sql_params(self, algorithm_params: dict):
return (algorithm_params['constraints_generator'],
algorithm_params['margin'],
algorithm_params['sigma0'],
algorithm_params['k'],
algorithm_params['n'],
algorithm_params['seed'],
algorithm_params['model_name'],
algorithm_params['clustering_k_min'],
algorithm_params['scaler'],
algorithm_params['train_sample'])
def data_source(self, scaler: bool = True,
constraints_generator: callable = cg.f_n3,
clustering_k_min: int = 2,
sigma0: float = 0.125,
margin: float = 1.0,
benchmark_mode: bool = False,
seeds: range = range(0, 30), K: range = range(1, 3), N: range = range(2, 8), train_sample: int = 500, models=['ball', 'simplex', 'cube']):
experiments = []
for seed in seeds:
for k in K:
for n in N:
for model in models:
inopts = frozendict({
'constraints_generator': constraints_generator.__name__,
'sigma0': sigma0,
'k': k,
'n': n,
'scaler': scaler,
'margin': margin,
'clustering_k_min': clustering_k_min,
'seed': seed,
'model_name': model,
'benchmark_mode': benchmark_mode,
'train_sample': train_sample
})
experiments.append(inopts)
return experiments
def experiments_1(self, seeds: range = range(0, 30), K: range = range(1, 3), N: range = range(2, 8), models=['ball', 'simplex', 'cube']) -> list:
return [self.data_source(seeds=seeds, K=K, N=N,
scaler=scaler,
constraints_generator=cg.f_2np2,
clustering_k_min=1,
sigma0=0.5,
margin=1.0,
train_sample=300,
models=models) for scaler in [True, False]]
def experiments_2(self, seeds: range = range(0, 30), K: range = range(1, 3), N: range = range(2, 8), models=['ball', 'simplex', 'cube']) -> list:
return [self.data_source(seeds=seeds, K=K, N=N,
scaler=True,
constraints_generator=constraints_generator,
clustering_k_min=1,
sigma0=0.5,
margin=1.0,
train_sample=300,
models=models) for constraints_generator in [cg.f_2n, cg.f_2np2, cg.f_n3, cg.f_2pn]] # cg.f_n1,
def experiments_3(self, seeds: range = range(0, 30), K: range = range(1, 3), N: range = range(2, 8), models=['ball', 'simplex', 'cube']) -> list:
return [self.data_source(seeds=seeds, K=K, N=N,
scaler=True,
constraints_generator=cg.f_2pn,
clustering_k_min=kmin,
sigma0=0.5,
margin=1.0,
train_sample=300,
models=models) for kmin in [0, 1, 2]]
def experiments_4(self, seeds: range = range(0, 30), K: range = range(1, 3), N: range = range(2, 8), models=['ball', 'simplex', 'cube']) -> list:
return [self.data_source(seeds=seeds, K=K, N=N,
scaler=True,
constraints_generator=cg.f_2pn,
clustering_k_min=2,
sigma0=sigma,
margin=1.0,
train_sample=300,
models=models) for sigma in [0.03125, 0.0625, 0.125, 0.25, 0.5, 1.0, 2.0]]
def experiments_5(self, seeds: range = range(0, 30), K: range = range(1, 3), N: range = range(2, 8), models=['ball', 'simplex', 'cube']) -> list:
return [self.data_source(seeds=seeds, K=K, N=N,
scaler=True,
constraints_generator=cg.f_2pn,
clustering_k_min=2,
sigma0=0.03125,
margin=margin,
train_sample=300,
models=models) for margin in [0.9, 1, 1.1]]
def experiments_6(self, seeds: range = range(0, 30), N: range = range(2, 8), K=range(1, 3), models=['ball', 'simplex', 'cube']) -> list:
return [
self.data_source(seeds=seeds, K=K, N=N,
scaler=True,
constraints_generator=cg.f_2pn,
clustering_k_min=2,
sigma0=0.03125,
margin=1.0,
train_sample=ts,
models=models) for ts in [500, 400, 100, 200, 300]]
def experiments_case_study(self, seeds: range = range(0, 30), N=[8], K=[1], models=['case_study']) -> list:
return [
self.data_source(seeds=seeds, K=K, N=N,
scaler=True,
constraints_generator=cg.f_2pn,
clustering_k_min=2,
sigma0=0.03125,
margin=1.0,
train_sample=1022,
models=models)]
def benchmarks(self) -> list:
return [self.data_source(benchmark_mode=True)]
def experiment(self, key, seeds: range = range(0, 30)):
return {
1: self.experiments_1(seeds=seeds),
2: self.experiments_2(seeds=seeds),
3: self.experiments_3(seeds=seeds),
4: self.experiments_4(seeds=seeds),
5: self.experiments_5(seeds=seeds),
6: self.experiments_6(seeds=seeds),
'case_study': self.experiments_case_study(seeds=seeds),
'best': [self.data_source(seeds=seeds)],
'benchmarks': self.benchmarks()
}[key]
def check_existing_experiments(self, seeds: range = range(0, 30)):
db = 'experiments.sqlite'
database = Database(database_filename=db)
for i in range(1, 6):
experiments = set(flat(self.experiment(i, seeds=seeds)))
seq_of_params = [self.convert_to_sql_params(experiment) for experiment in experiments]
db_experiments = [database.engine.execute(self.sql, params).fetchone()[0] for params in seq_of_params]
filtered = list(filter(lambda t: t[0] > 0, zip(db_experiments, experiments)))
log.info("Number of existing experiments in experiment {}: {}/{}".format(i, len(filtered), len(experiments)))
def run_instance(self, inopts: dict):
algorithm = CMAESAlgorithm(**inopts)
algorithm.experiment()
def run(self, experiments: list):
experiments = flat(experiments)
db = 'experiments.sqlite'
database = Database(database_filename=db)
experiments = self.filter_algorithms(experiments, database=database)
cpus = 3 # psutil.cpu_count(logical=False)
pool = Pool(processes=cpus) # start worker processes
pool.map(self.run_instance, experiments, 1)
def run_slurm(self, experiments: list):
experiments = set(flat(experiments))
db = 'experiments.sqlite'
database = Database(database_filename=db)
experiments = self.filter_algorithms(experiments, database=database)
pool = SlurmPool()
for experiment in experiments:
try:
mapped = list(map(lambda item: item[0] + ':' + str(item[1]), experiment.items()))
arguments = "\"" + reduce(lambda key, value: key + ';' + value, mapped) + "\""
pool.execute(cmd='python', arguments='cmaes.py {}'.format(arguments),
script_filename=str(self.convert_to_sql_params(experiment)))
except:
print(experiment)
pool.close()
if __name__ == '__main__':
runner = AlgorithmRunner()
seeds = range(0, 15)
models = ['ball', 'cube', 'simplex']
experiments = flat([
runner.experiments_1(seeds=seeds, N=range(3, 8), K=range(1, 3), models=models),
runner.experiments_2(seeds=seeds, N=range(3, 8), K=range(1, 3), models=models),
runner.experiments_3(seeds=seeds, N=range(3, 8), K=range(1, 3), models=models),
runner.experiments_4(seeds=seeds, N=range(3, 8), K=range(1, 3), models=models),
runner.experiments_5(seeds=seeds, N=range(3, 8), K=range(1, 3), models=models),
runner.experiments_6(seeds=seeds, N=range(3, 8), K=range(1, 3), models=models),
# runner.experiments_case_study(seeds=seeds)
])
# experiments = runner.experiments_1(seeds=seeds)
runner.run(experiments)
# runner.run_slurm(experiments)