-
Notifications
You must be signed in to change notification settings - Fork 1
/
surrogate_ea.py
411 lines (323 loc) · 13.8 KB
/
surrogate_ea.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
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Thu June 21 09:42:07 2018
@author: niko
"""
import numpy as np
import copy as cp
from surrogates import NeuralNet, Kriging, RBFN
from sklearn.svm import NuSVR, SVR
from sklearn.tree import DecisionTreeRegressor
from _utils import calc_cwd_dist, remove_duplication
from multiprocessing.pool import Pool
class SurrogateEAMixin(object):
""" Mixins for Surrogate Assisted EAs
"""
def __init__(self, max_episode=60, embedded_ea=None, verbosity=1,
*args, **kwargs):
self.true_front = []
self.episode = 0
self.max_episode = max_episode
self.embedded_ea = embedded_ea
self.verbosity = verbosity
return None
def _render_pop_by_name(self, name='global'):
""" Render a pop ( or a frontier or a set of solutions) given its name
"""
if not(name.__class__ is str):
raise TypeError('name should be a "str"... ')
# Renders self.global_pop
global_list = ['global', 'population', 'pop']
# Renders self.front
front_list = ['front', 'hat']
# Renders self.true_front
true_front_list = ['pm', 'true']
_s = name.lower()
if any(_pattern in _s for _pattern in global_list):
return self.global_pop
elif any(_pattern in _s for _pattern in front_list):
return self.front
elif any(_pattern in _s for _pattern in true_front_list):
return self.true_front
else:
raise ValueError('Name of the population % not found....' % name )
return None
def generate_init(self, trial_method='random',
trial_criterion='cm', **kwargs):
if self.generation == 0:
self.episode = 0
local = self.trial(method=trial_method, criterion=trial_criterion)
if self.mixinteger:
local = self.trim_mixinteger(local)
for i in local:
if self.cache.find(i, overwrite=True) is None:
i.calc_fitness(fun=self.fitness_fun)
self.global_pop = local
self.elites = []
self.front = []
self.true_front = []
self.bounds = local[0].bounds
self.generation = 1
return None
def recalc_fitness_with(self, fun):
# Re-calculate fitness in population
for i in self.global_pop:
i.calc_fitness(fun)
# Re-calculate fitness in current front
for i in self.front:
i.calc_fitness(fun)
# Re-evaluate the pareto front with PM results
self.front = self.compute_front(on=self.front)
return None
def update_true_front(self, front=None):
"""Use the surrogate front to update the current true Pareto front.
The inaccuracy of the surrogate (overfitting or underfitting) may
result in fake Pareto optimals which override the true ones. In
order to save those true values, we build this so-called true_front
archive to cache the PM's true Pareto optimal.
"""
front = self.front if front is None else front
self.true_front.extend(front)
self.true_front = self.compute_front(pop=self.true_front)
return None
def crossover(self, **kwargs):
""" Perform crossover in self.front (fake front in surrogate EA)
"""
self._crossover(parent_pop=self.front,
offspring_pop=self.global_pop,
mutate=True, calc_fitness=False, **kwargs)
return None
def crossover_in_true_front(self, **kwargs):
self._crossover(parent_pop=self.true_front,
offspring_pop=self.global_pop,
mutate=True, calc_fitness=False, **kwargs)
return None
def render_features(self, pop='global'):
if pop == 'global':
_pop = self.global_pop
elif pop == 'front':
_pop = self.front
elif pop == 'true_front':
_pop = self.true_front
elif pop.__class__.__name__ == 'list':
_pop = pop
else:
raise NotImplementedError('Render features (genes) from sets other'
' than global_pop, front, or true_front '
'not supported...')
return np.array([i.gene for i in _pop])
def render_targets(self, pop='global'):
if pop == 'global':
_pop = self.global_pop
elif pop == 'front':
_pop = self.front
elif pop == 'true_front':
_pop = self.true_front
elif pop.__class__.__name__ == 'list':
_pop = pop
else:
raise NotImplementedError('Render targets (fitnesses) from sets '
'other than global_pop, front, or '
'true_front not supported...')
return np.array([i.fitness for i in _pop])
def expensive_eval(self, candidates, verbose=True):
""" Expensively evaluate the candidates' fitnesses
"""
if self.n_process <= 1:
new_candidates = [i for i in candidates if self._expensive_eval(i)
is not None]
else:
# parallel
no_dup = remove_duplication(candidates)
new_candidates = self._expensive_eval_parallel(no_dup)
if self.verbose and verbose:
print("New expensively evaluations: %s" % len(new_candidates))
return new_candidates
def _expensive_eval(self, i):
if self.cache.find(i, overwrite=True) is None:
i.calc_fitness(fun=self.fitness_fun)
self.cache.save(i)
if hasattr(self, 'sampled_archive'): self.sampled_archive.append(i)
return i
return None
def _expensive_eval_parallel(self, candidates):
pool = Pool(self.n_process)
rs = [pool.apply_async(self._expensive_eval, i) for i in candidates]
candidates = [r.get() for r in rs]
pool.close()
pool.join()
return candidates
def cheap_eval(self, candidates='global'):
""" Evaluate the candidates with the surrogate model
"""
if candidates == 'global':
candidates = self.global_pop
else:
pass
[i.calc_fitness(fun=self.surrogate.render_fitness) for i in candidates]
return None
def train_surrogate(self, samples=None):
""" Train the surrogate(s)
"""
if samples == []: return self.surrogate
# Retraining of the surrogate
if self.surrogate._warm_start and samples is not None:
X = self.render_features(pop=samples)
y = self.render_targets(pop=samples)
else:
X = self.render_features(pop=self.sampled_archive)
y = self.render_targets(pop=self.sampled_archive)
self.surrogate.fit(X, y)
return self.surrogate
def _progressive_revolution(self):
""" Progressive revolution
"""
new_pop = self._crossover(parent_pop=self.true_front,
offspring_pop=cp.deepcopy(self.global_pop),
mutate=True, calc_fitness=False)
new_pop = self.expensive_eval(new_pop)
self.update_true_front(front=new_pop)
if self.surrogate._warm_start:
self.train_surrogate(samples=new_pop)
else:
self.train_surrogate(samples=self.sampled_archive)
return None
def progressive_revolution(self, n_no_improvement, tol):
""" Performance a progressive revolution if no improvement observed
"""
deadlock = False
if len(self.hypervol) < n_no_improvement or n_no_improvement < 2:
return deadlock
diff = np.diff(self.hypervol[-n_no_improvement:])
ratio = np.divide(diff, self.hypervol[-n_no_improvement:-1] + \
np.finfo(float).tiny).__abs__()
if np.less(ratio, tol).all():
deadlock = True
if self.verbose:
print('')
print('No major improvement was observed, performing '
'a progressive revolution...')
self._progressive_revolution()
return deadlock
def find_least_crowded(self, candidates='default', **kwargs):
""" Find the least crowded solution in a set of non-dominated
solutions (PM-evaluated by default)
"""
if candidates in ['default', 'pm', 'PM', 'Pm', 'true_front']:
_candidates = self.true_front
elif candidates in ['local', 'front']:
_candidates = self.front
else:
raise ValueError('Can not find the least crowded solution on "%"'
% candidates)
calc_cwd_dist(pop=_candidates, **kwargs)
_init = True
for i in _candidates:
if _init:
_least_crowded = i
_init = False
# if i._on_edge:
# continue
if _least_crowded._dist < i._dist: _least_crowded = i
self._least_crowded = cp.deepcopy(_least_crowded)
return self._least_crowded
def config_surrogate(self, typ='ANN', params={}, premade=None,
n_models=None, n_process=1, X_scaler=None,
y_scaler=None, warm_start=False, **kwargs):
""" Configurate and initialize a surrogate
"""
if premade is not None:
self.surrogate = premade
return None
_t = typ.lower()
if _t in ['ann', 'mlp', 'neural_network', 'neural-network']:
self.surrogate = NeuralNet(n_objs=self.n_objs, params=params,
n_models=n_models, n_process=n_process,
X_scaler=X_scaler, y_scaler=y_scaler,
warm_start=warm_start)
elif _t in ['svm', 'svr']:
self.surrogate = SVR(**params)
elif _t in ['nusvm', 'nusvr', 'nu-svm', 'nu-svr', 'nu_svm', 'nu_svr']:
self.surrogate = NuSVR(**params)
elif _t[:3] == 'rbf':
self.surrogate = RBFN(n_objs=self.n_objs, params=params,
n_models=n_models, n_process=n_process,
X_scaler=X_scaler, y_scaler=y_scaler,
warm_start=warm_start)
elif 'tree' in _t:
self.surrogate = DecisionTreeRegressor(**params)
elif 'kriging' in _t:
self.surrogate = Kriging(n_objs=self.n_objs, params=params,
n_models=n_models, n_process=n_process,
X_scaler=X_scaler, y_scaler=y_scaler,
warm_start=warm_start)
else:
raise NotImplementedError('Surrogate type % not supported...' %typ)
return None
def max_episode_termination(self):
terminate = False
if self.max_episode <= self.episode: terminate = True
return terminate
def stop(self):
if self.stopping_rule == 'max_generation':
return self.max_generation_termination()
elif self.stopping_rule == 'max_eval':
return self.max_eval_termination()
elif self.stopping_rule == 'max_episode':
return self.max_episode_termination()
else:
raise ValueError("Unknown stopping_rule: %s" % self.stopping_rule)
return True
def config_embedded_ea(self, **kwargs):
if self.embedded_ea is None:
pass
elif hasattr(self.embedded_ea, "_external_moea"):
self.embedded_ea_ = \
self.embedded_ea(problem=self.problem, surrogate=self.surrogate,
size=self.size, generation=self.max_generation,
**kwargs)
else:
raise ValueError("Unknown embedded EA")
return None
def evolve_surrogate(self, **params_ea):
if self.embedded_ea is None:
self._naive_ea(**params_ea)
elif hasattr(self.embedded_ea, "_external_moea"):
self._evolve_embedded_ea(**params_ea)
else:
raise ValueError("Unknown embedded EA")
return None
def _naive_ea(self, **params_ea):
while not self.max_generation_termination():
self.crossover_in_true_front(**params_ea)
self.cheap_eval(candidates='global')
self.select(**params_ea)
self.update_front(**params_ea)
self.generation += 1
return None
def _evolve_embedded_ea(self, **params_ea):
# Apply crossover in the true front for surrogate-assisted optimization
# self.crossover_in_true_front(**params_ea)
# Evolutionary computation over the surrogate
# Prevent neglect of the first population
if self.episode < 1:
self.embedded_ea_.load_external_pop_xf(pop=self.global_pop)
else:
self.embedded_ea_.load_external_pop_x(pop=self.global_pop)
self.embedded_ea_.evolve()
self.embedded_ea_.export_internal_pop(pop=self.global_pop)
# Select and update non-dominated solutions
self.select(**params_ea)
self.update_front(**params_ea)
self.generation = self.max_generation
return None
def report(self, verbosity=None):
verbosity = verbosity or self.verbosity
if self.episode % verbosity == 0:
print("Episode: %s, Total expensive evaluations: %s,\n"
"True front size: %s,\n"
"Surrogate Front size: %s,\n"
"Hypervolume: %.3e \n" %
(self.episode, self.problem.n_evals, len(self.true_front),
self.front.__len__(), self.hypervol[-1]))