-
Notifications
You must be signed in to change notification settings - Fork 2
/
inference_mlp.py
375 lines (301 loc) · 13.5 KB
/
inference_mlp.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
'''
DL models (FNN, 1D CNN and CNN-LSTM) evaluation on N-CMAPSS
12.07.2021
Hyunho Mo
hyunho.mo@unitn.it
'''
## Import libraries in python
import gc
import argparse
import os
import json
import logging
import sys
import h5py
import time
import matplotlib
import numpy as np
import pandas as pd
import seaborn as sns
from pandas import DataFrame
import matplotlib.pyplot as plt
from matplotlib import gridspec
import math
import random
from random import shuffle
from tqdm.keras import TqdmCallback
import importlib
from scipy.stats import randint, expon, uniform
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 tqdm import tqdm
import scipy.stats as stats
# from sklearn.utils.testing import ignore_warnings
# from sklearn.exceptions import ConvergenceWarning
# import keras
import tensorflow as tf
print(tf.__version__)
# import keras.backend as K
import tensorflow.keras.backend as K
from tensorflow.keras import backend
from tensorflow.keras import optimizers
from tensorflow.keras.models import Sequential, load_model, Model
from tensorflow.keras.layers import Input, Dense, Flatten, Dropout, Embedding
from tensorflow.keras.layers import BatchNormalization, Activation, LSTM, TimeDistributed, Bidirectional
from tensorflow.keras.layers import Conv1D
from tensorflow.keras.layers import MaxPooling1D
from tensorflow.keras.layers import concatenate
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint, LearningRateScheduler
from tensorflow.python.framework.convert_to_constants import convert_variables_to_constants_v2_as_graph
from utils.data_preparation_unit import df_all_creator, df_train_creator, df_test_creator, Input_Gen
from utils.dnn import one_dcnn, mlps
# import tensorflow.compat.v1 as tf
# tf.disable_v2_behavior()
seed = 0
random.seed(0)
np.random.seed(seed)
# Ignore tf err log
pd.options.mode.chained_assignment = None # default='warn'
# from tensorflow.compat.v1 import ConfigProto
# from tensorflow.compat.v1 import InteractiveSession
# config = ConfigProto()
# config.gpu_options.allow_growth = True
# session = InteractiveSession(config=config)
#gpus = tf.config.experimental.list_physical_devices('GPU')
#for gpu in gpus:
# tf.config.experimental.set_memory_growth(gpu, True)
# tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
# tf.get_logger().setLevel(logging.ERROR)
# tf.config.set_visible_devices([], 'GPU')
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', 'oned_fnn_rep.h5')
tf_temp_path = os.path.join(current_dir, 'TF_Model_tf')
pic_dir = os.path.join(current_dir, 'Figures')
'''
load array from npz files
'''
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 rmse(y_true, y_pred):
return backend.sqrt(backend.mean(backend.square(y_pred - y_true), axis=-1))
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 + "/mlp_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 get_flops(model):
concrete = tf.function(lambda inputs: model(inputs))
concrete_func = concrete.get_concrete_function(
[tf.TensorSpec([1, *inputs.shape[1:]]) for inputs in model.inputs])
frozen_func, graph_def = convert_variables_to_constants_v2_as_graph(concrete_func)
with tf.Graph().as_default() as graph:
tf.graph_util.import_graph_def(graph_def, name='')
run_meta = tf.compat.v1.RunMetadata()
opts = tf.compat.v1.profiler.ProfileOptionBuilder.float_operation()
flops = tf.compat.v1.profiler.profile(graph=graph, run_meta=run_meta, cmd="op", options=opts)
return flops.total_float_ops
def scheduler(epoch, lr):
if epoch == 30:
print ("lr decay by 10")
return lr * 0.1
elif epoch == 70:
print("lr decay by 10")
return lr * 0.1
else:
return lr
def release_list(a):
del a[:]
del a
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 main():
# current_dir = os.path.dirname(os.path.abspath(__file__))
parser = argparse.ArgumentParser(description='sample creator')
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('-h1', type=int, default=200, help='number of neurons')
parser.add_argument('-h2', type=int, default=200, help='number of neurons')
parser.add_argument('-h3', type=int, default=200, help='number of neurons')
parser.add_argument('-h4', type=int, default=50, help='number of neurons')
parser.add_argument('-bs', type=int, default=256, help='batch size')
parser.add_argument('-ep', type=int, default=30, help='max epoch')
parser.add_argument('-pt', type=int, default=20, help='patience')
parser.add_argument('-vs', type=float, default=0.1, help='validation split')
parser.add_argument('-lr', type=float, default=0.001, help='learning rate')
parser.add_argument('-sub', type=int, default=10, help='subsampling stride')
args = parser.parse_args()
win_len = args.w
win_stride = args.s
partition = 3
h1 = args.h1
h2 = args.h2
h3 = args.h3
h4 = args.h4
lr = args.lr
bs = args.bs
ep = args.ep
pt = args.pt
vs = args.vs
sub = args.sub
amsgrad = optimizers.Adam(learning_rate=lr, beta_1=0.9, beta_2=0.999, epsilon=1e-07, amsgrad=True, name='Adam')
rmsop = optimizers.RMSprop(learning_rate=lr, rho=0.9, momentum=0.0, epsilon=1e-07, centered=False,
name='RMSprop')
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)
print("sample_array.shape", sample_array.shape)
print("label_array.shape", label_array.shape)
sample_array = sample_array[::sub]
label_array = label_array[::sub]
print("sub sample_array.shape", sample_array.shape)
print("sub label_array.shape", label_array.shape)
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")
print("sample_array.shape", sample_array.shape)
print("label_array.shape", label_array.shape)
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)
feat_len = sample_array.shape[1]
print ("feat_len", feat_len)
start = time.time()
fnn_model = mlps(feat_len, h1, h2, h3, h4)
print(fnn_model.summary())
# one_d_cnn_model.compile(loss='mean_squared_error', optimizer=amsgrad, metrics=[rmse, 'mae'])
fnn_model.compile(loss='mean_squared_error', optimizer=amsgrad, metrics='mae')
lr_scheduler = LearningRateScheduler(scheduler)
history = fnn_model.fit(sample_array, label_array, epochs=ep, batch_size=bs, validation_split=vs, verbose=2,
callbacks = [EarlyStopping(monitor='val_loss', min_delta=0, patience=pt, verbose=1, mode='min'),
ModelCheckpoint(model_temp_path, monitor='val_loss', save_best_only=True, mode='min', verbose=1)]
)
# TqdmCallback(verbose=2)
# one_d_cnn_model.save(tf_temp_path,save_format='tf')
figsave(history, h1, h2, h3, h4, bs, lr, sub)
print("The FLOPs is:{}".format(get_flops(fnn_model)), flush=True)
num_train = sample_array.shape[0]
end = time.time()
training_time = end - start
print("Training time: ", training_time)
### Test (inference after training)
start = time.time()
output_lst = []
truth_lst = []
for index in units_index_test:
print ("test idx: ", index)
sample_array, label_array = load_array(sample_dir_path, index, win_len, win_stride)
# estimator = load_model(tf_temp_path, custom_objects={'rmse':rmse})
print("sample_array.shape", sample_array.shape)
print("label_array.shape", label_array.shape)
sample_array = sample_array[::sub]
label_array = label_array[::sub]
print("sub sample_array.shape", sample_array.shape)
print("sub label_array.shape", label_array.shape)
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)
estimator = load_model(model_temp_path)
y_pred_test = estimator.predict(sample_array)
output_lst.append(y_pred_test)
truth_lst.append(label_array)
print(output_lst[0].shape)
print(truth_lst[0].shape)
print(np.concatenate(output_lst).shape)
print(np.concatenate(truth_lst).shape)
output_array = np.concatenate(output_lst)[:, 0]
trytg_array = np.concatenate(truth_lst)
print(output_array.shape)
print(trytg_array.shape)
rms = sqrt(mean_squared_error(output_array, trytg_array))
print(rms)
rms = round(rms, 2)
end = time.time()
inference_time = end - start
num_test = output_array.shape[0]
for idx in range(len(units_index_test)):
print(output_lst[idx])
print(truth_lst[idx])
fig_verify = plt.figure(figsize=(24, 10))
plt.plot(output_lst[idx], color="green")
plt.plot(truth_lst[idx], color="red", linewidth=2.0)
plt.title('Unit%s inference' %str(int(units_index_test[idx])), fontsize=30)
plt.yticks(fontsize=20)
plt.xticks(fontsize=20)
plt.ylabel('RUL', fontdict={'fontsize': 24})
plt.xlabel('Timestamps', fontdict={'fontsize': 24})
plt.legend(['Predicted', 'Truth'], loc='upper right', fontsize=28)
plt.show()
fig_verify.savefig(pic_dir + "/mlps_unit%s_test_h1%s_h2%s_h3%s_h4%s_bs%s_lr%s_sub%s_rmse-%s.png" %(str(int(units_index_test[idx])),
int(h1), int(h2), int(h3), int(h4), int(bs),
str(lr), int(sub), str(rms)))
print("The FLOPs is:{}".format(get_flops(fnn_model)), flush=True)
print("wind length_%s, win stride_%s" %(str(win_len), str(win_stride)))
print("# Training samples: ", num_train)
print("# Inference samples: ", num_test)
print("Training time: ", training_time)
print("Inference time: ", inference_time)
print("Result in RMSE: ", rms)
if __name__ == '__main__':
main()