-
Notifications
You must be signed in to change notification settings - Fork 2
/
enas_convELM_simp.py
488 lines (371 loc) · 15.2 KB
/
enas_convELM_simp.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
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
'''
Created on April , 2021
@author:
'''
## Import libraries in python
import argparse
import time
import json
import logging
import sys
import os
import math
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import random
import importlib
from scipy.stats import randint, expon, uniform
import glob
import tensorflow as tf
import sklearn as sk
from sklearn import svm
from sklearn.utils import shuffle
from sklearn import metrics
from sklearn import preprocessing
from sklearn import pipeline
from sklearn.metrics import mean_squared_error
from math import sqrt
from utils.elm_network import network_fit
from utils.hpelm import ELM, HPELM
from utils.convELM_task_simp import SimpleNeuroEvolutionTask
from utils.ea_multi import GeneticAlgorithm
import torch
# random seed predictable
jobs = 1
current_dir = os.path.dirname(os.path.abspath(__file__))
data_filedir = os.path.join(current_dir, 'N-CMAPSS')
data_filepath = os.path.join(current_dir, 'N-CMAPSS', 'N-CMAPSS_DS02-006.h5')
sample_dir_path = os.path.join(data_filedir, 'Samples_whole')
model_temp_path = os.path.join(current_dir, 'Models', 'convELM_rep.h5')
torch_temp_path = os.path.join(current_dir, 'torch_model')
pic_dir = os.path.join(current_dir, 'Figures')
# Log file path of EA in csv
# directory_path = current_dir + '/EA_log'
directory_path = os.path.join(current_dir, 'EA_log')
if not os.path.exists(pic_dir):
os.makedirs(pic_dir)
if not os.path.exists(directory_path):
os.makedirs(directory_path)
'''
load array from npz files
'''
def array_tensorlst_data (arry, bs, device):
# arry = arry.reshape(arry.shape[0],arry.shape[2],arry.shape[1])
if bs > arry.shape[0]:
bs = arry.shape[0]
arry = arry.transpose((0,2,1))
print ("arry.shape[0]//bs", arry.shape[0]//bs)
num_train_batch = arry.shape[0]//bs
print (arry.shape)
arry_cut = arry[:num_train_batch*bs]
arrt_rem = arry[num_train_batch*bs:]
print (arry.shape)
arry4d = arry_cut.reshape(int(arry_cut.shape[0]/bs), bs, arry_cut.shape[1], arry_cut.shape[2])
print (arry4d.shape)
arry_lst = list(arry4d)
arry_lst.append(arrt_rem)
print (len(arry_lst))
print (arry_lst[0].shape)
train_batch_lst = []
for batch_sample in arry_lst:
arr_tensor = torch.from_numpy(batch_sample)
if torch.cuda.is_available():
arr_tensor = arr_tensor.to(device)
train_batch_lst.append(arr_tensor)
return train_batch_lst
def array_tensorlst_label (arry, bs, device):
if bs > arry.shape[0]:
bs = arry.shape[0]
print ("arry.shape[0]//bs", arry.shape[0]//bs)
num_train_batch = arry.shape[0]//bs
arry_cut = arry[:num_train_batch*bs]
arrt_rem = arry[num_train_batch*bs:]
arry2d = arry_cut.reshape(int(arry_cut.shape[0]/bs), bs)
arry_lst = list(arry2d)
arry_lst.append(arrt_rem)
print (len(arry_lst))
print (arry_lst[0].shape)
train_batch_lst = []
for batch_sample in arry_lst:
arr_tensor = torch.from_numpy(batch_sample)
if torch.cuda.is_available():
arr_tensor = arr_tensor.to(device)
train_batch_lst.append(arr_tensor)
return train_batch_lst
def load_part_array (sample_dir_path, unit_num, win_len, stride, part_num):
filename = 'Unit%s_win%s_str%s_part%s.npz' %(str(int(unit_num)), win_len, stride, part_num)
filepath = os.path.join(sample_dir_path, filename)
loaded = np.load(filepath)
return loaded['sample'], loaded['label']
def load_part_array_merge (sample_dir_path, unit_num, win_len, win_stride, partition):
sample_array_lst = []
label_array_lst = []
print ("Unit: ", unit_num)
for part in range(partition):
print ("Part.", part+1)
sample_array, label_array = load_part_array (sample_dir_path, unit_num, win_len, win_stride, part+1)
sample_array_lst.append(sample_array)
label_array_lst.append(label_array)
sample_array = np.dstack(sample_array_lst)
label_array = np.concatenate(label_array_lst)
sample_array = sample_array.transpose(2, 0, 1)
print ("sample_array.shape", sample_array.shape)
print ("label_array.shape", label_array.shape)
return sample_array, label_array
def load_array (sample_dir_path, unit_num, win_len, stride):
filename = 'Unit%s_win%s_str%s.npz' %(str(int(unit_num)), win_len, stride)
filepath = os.path.join(sample_dir_path, filename)
loaded = np.load(filepath)
return loaded['sample'].transpose(2, 0, 1), loaded['label']
def shuffle_array(sample_array, label_array):
ind_list = list(range(len(sample_array)))
print("ind_list befor: ", ind_list[:10])
print("ind_list befor: ", ind_list[-10:])
ind_list = shuffle(ind_list)
print("ind_list after: ", ind_list[:10])
print("ind_list after: ", ind_list[-10:])
print("Shuffeling in progress")
shuffle_sample = sample_array[ind_list, :, :]
shuffle_label = label_array[ind_list,]
return shuffle_sample, shuffle_label
def figsave(history, h1,h2,h3,h4, bs, lr, sub):
fig_acc = plt.figure(figsize=(15, 8))
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('Training', fontsize=24)
plt.ylabel('loss', fontdict={'fontsize': 18})
plt.xlabel('epoch', fontdict={'fontsize': 18})
plt.legend(['Training loss', 'Validation loss'], loc='upper left', fontsize=18)
plt.show()
print ("saving file:training loss figure")
fig_acc.savefig(pic_dir + "/elm_enas_training_h1%s_h2%s_h3%s_h4%s_bs%s_sub%s_lr%s.png" %(int(h1), int(h2), int(h3), int(h4), int(bs), int(sub), str(lr)))
return
def score_calculator(y_predicted, y_actual):
# Score metric
h_array = y_predicted - y_actual
s_array = np.zeros(len(h_array))
for j, h_j in enumerate(h_array):
if h_j < 0:
s_array[j] = math.exp(-(h_j / 13)) - 1
else:
s_array[j] = math.exp(h_j / 10) - 1
score = np.sum(s_array)
return score
def release_list(a):
del a[:]
del a
def recursive_clean(directory_path):
"""clean the whole content of :directory_path:"""
if os.path.isdir(directory_path) and os.path.exists(directory_path):
files = glob.glob(directory_path + '*')
for file_ in files:
if os.path.isdir(file_):
recursive_clean(file_ + '/')
else:
os.remove(file_)
units_index_train = [2.0, 5.0, 10.0, 16.0, 18.0, 20.0]
units_index_test = [11.0, 14.0, 15.0]
def tensor_type_checker(tensor, device):
if torch.cuda.is_available():
tensor = tensor.to(device)
print(f"Shape of tensor: {tensor.shape}")
print(f"Datatype of tensor: {tensor.dtype}")
print(f"Device tensor is stored on: {tensor.device}")
return tensor
def main():
# current_dir = os.path.dirname(os.path.abspath(__file__))
parser = argparse.ArgumentParser(description='NAS CNN')
parser.add_argument('-w', type=int, default=50, help='sequence length', required=True)
parser.add_argument('-s', type=int, default=1, help='stride of filter')
parser.add_argument('-bs', type=int, default=512, help='batch size')
parser.add_argument('-pt', type=int, default=30, help='patience')
parser.add_argument('-ep', type=int, default=100, help='epochs')
parser.add_argument('-vs', type=float, default=0.2, help='validation split')
parser.add_argument('-lr', type=float, default=10**(-1*4), help='learning rate')
parser.add_argument('-sub', type=int, default=10, help='subsampling stride')
parser.add_argument('-t', type=int, required=True, help='trial')
parser.add_argument('--pop', type=int, default=20, required=False, help='population size of EA')
parser.add_argument('--gen', type=int, default=20, required=False, help='generations of evolution')
parser.add_argument('--device', type=str, default="cuda", help='Use "basic" if GPU with cuda is not available')
parser.add_argument('--obj', type=str, default="soo", help='Use "soo" for single objective and "moo" for multiobjective')
args = parser.parse_args()
win_len = args.w
win_stride = args.s
lr = args.lr
bs = args.bs
ep = args.ep
pt = args.pt
vs = args.vs
sub = args.sub
device = args.device
print(f"Using {device} device")
obj = args.obj
trial = args.t
# random seed predictable
jobs = 1
seed = trial
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
train_units_samples_lst =[]
train_units_labels_lst = []
for index in units_index_train:
print("Load data index: ", index)
sample_array, label_array = load_array (sample_dir_path, index, win_len, win_stride)
#sample_array, label_array = shuffle_array(sample_array, label_array)
sample_array = sample_array[::sub]
label_array = label_array[::sub]
# sample_array = sample_array.astype(np.float32)
# label_array = label_array.astype(np.float32)
train_units_samples_lst.append(sample_array)
train_units_labels_lst.append(label_array)
sample_array = np.concatenate(train_units_samples_lst)
label_array = np.concatenate(train_units_labels_lst)
print ("samples are aggregated")
release_list(train_units_samples_lst)
release_list(train_units_labels_lst)
train_units_samples_lst =[]
train_units_labels_lst = []
print("Memory released")
sample_array, label_array = shuffle_array(sample_array, label_array)
print("samples are shuffled")
# sample_array = sample_array.reshape(sample_array.shape[0], sample_array.shape[2])
print("sample_array_reshape.shape", sample_array.shape)
print("label_array_reshape.shape", label_array.shape)
window_length = sample_array.shape[1]
feat_len = sample_array.shape[2]
num_samples = sample_array.shape[0]
print ("window_length", window_length)
print("feat_len", feat_len)
train_sample_array = sample_array[:int(num_samples*(1-vs))]
train_label_array = label_array[:int(num_samples*(1-vs))]
val_sample_array = sample_array[int(num_samples*(1-vs))+1:]
val_label_array = label_array[int(num_samples*(1-vs))+1:]
print ("train_sample_array.shape", train_sample_array.shape)
print ("train_label_array.shape", train_label_array.shape)
print ("val_sample_array.shape", val_sample_array.shape)
print ("val_label_array.shape", val_label_array.shape)
sample_array = []
label_array = []
# train_sample_array.shape (84212, 50, 20) = # samples, win_len, feature_len
# train_label_array.shape (84212,)
# val_sample_array.shape (21053, 50, 20)
# val_label_array.shape (21053,)
train_sample_array = array_tensorlst_data(train_sample_array, bs, device)[0]
val_sample_array = array_tensorlst_data(val_sample_array, bs, device)[0]
train_label_array = array_tensorlst_label(train_label_array, bs, device)[0]
val_label_array = array_tensorlst_label(val_label_array, bs, device)[0]
bs = train_sample_array.shape[0]
# tensor_type_checker(train_sample_array[0], device)
# tensor_type_checker(train_label_array[0], device)
# tensor_type_checker(val_sample_array[0], device)
# tensor_type_checker(val_label_array[0], device)
## Parameters for the GA
pop_size = args.pop
n_generations = args.gen
cx_prob = 0.5 # 0.25
mut_prob = 0.5 # 0.7
cx_op = "one_point"
mut_op = "uniform"
if obj == "soo":
sel_op = "best"
other_args = {
'mut_gene_probability': 0.3 # 0.1
}
mutate_log_path = os.path.join(directory_path, 'mute_log_ori_%s_%s_%s_%s.csv' % (pop_size, n_generations, obj, trial))
mutate_log_col = ['idx', 'params_1', 'params_2', 'params_3', 'params_4', 'params_5', 'fitness_1',
'gen']
mutate_log_df = pd.DataFrame(columns=mutate_log_col, index=None)
mutate_log_df.to_csv(mutate_log_path, index=False)
def log_function(population, gen, hv=None, mutate_log_path=mutate_log_path):
for i in range(len(population)):
indiv = population[i]
if indiv == []:
"non_mutated empty"
pass
else:
# print ("i: ", i)
indiv.append(indiv.fitness.values[0])
indiv.append(gen)
temp_df = pd.DataFrame(np.array(population), index=None)
temp_df.to_csv(mutate_log_path, mode='a', header=None)
print("population saved")
return
# elif obj == "moo":
else:
sel_op = "nsga2"
other_args = {
'mut_gene_probability': 0.4 # 0.1
}
mutate_log_path = os.path.join(directory_path, 'mute_log_ori_%s_%s_%s_%s.csv' % (pop_size, n_generations, obj, trial ))
mutate_log_col = ['idx', 'params_1', 'params_2', 'params_3', 'params_4',
'fitness_1', 'fitness_2', 'hypervolume', 'gen']
mutate_log_df = pd.DataFrame(columns=mutate_log_col, index=None)
mutate_log_df.to_csv(mutate_log_path, index=False)
def log_function(population, gen, hv=None, mutate_log_path=mutate_log_path):
for i in range(len(population)):
indiv = population[i]
if indiv == []:
"non_mutated empty"
pass
else:
# print ("i: ", i)
indiv.append(indiv.fitness.values[0])
indiv.append(indiv.fitness.values[1])
# append val_rmse
indiv.append(hv)
indiv.append(gen)
temp_df = pd.DataFrame(np.array(population), index=None)
temp_df.to_csv(mutate_log_path, mode='a', header=None)
print("population saved")
return
prft_path = os.path.join(directory_path, 'prft_out_ori_%s_%s_%s.csv' % (pop_size, n_generations, trial))
start = time.time()
cs = 0.0001
# Assign & run EA
task = SimpleNeuroEvolutionTask(
train_sample_array = train_sample_array,
train_label_array = train_label_array,
val_sample_array = val_sample_array,
val_label_array = val_label_array,
constant = cs,
epochs = ep,
batch=bs,
model_path = model_temp_path,
device = device,
obj = obj
)
# aic = task.evaluate(individual_seed)
ga = GeneticAlgorithm(
task=task,
population_size=pop_size,
n_generations=n_generations,
cx_probability=cx_prob,
mut_probability=mut_prob,
crossover_operator=cx_op,
mutation_operator=mut_op,
selection_operator=sel_op,
jobs=jobs,
log_function=log_function,
cs = cs,
**other_args
)
pop, log, hof, prtf = ga.run()
print("Best individual:")
print(hof[0])
print(prtf)
# Save to the txt file
# hof_filepath = tmp_path + "hof/best_params_fn-%s_ps-%s_ng-%s.txt" % (csv_filename, pop_size, n_generations)
# with open(hof_filepath, 'w') as f:
# f.write(json.dumps(hof[0]))
print("Best individual is saved")
end = time.time()
print("EA time: ", end - start)
print ("#################### EA COMPLETE / HOF TEST ##############################")
if __name__ == '__main__':
main()