-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodel_v7.py
251 lines (222 loc) · 10.1 KB
/
model_v7.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
#高精度，以cme起始信号作为触发的触发检测机
#可采用只输入磁场的方式
import kerastuner.tuners
import numpy as np
import tensorflow
import tensorflow as tf
from tensorflow.keras.callbacks import ModelCheckpoint
import h5py
from tensorflow.keras.models import Model, load_model, Sequential
from tensorflow.keras.layers import Dense, Activation, Dropout, Input, Masking, TimeDistributed, LSTM, Conv1D
from tensorflow.keras.layers import GRU, Bidirectional, BatchNormalization, Reshape,Embedding,Masking
from tensorflow.keras.optimizers import Adam
import V1_utils
import getdataset
import matplotlib.pyplot as plt
import tensorflow.keras.regularizers as tfkreg
import aced_utils
from kerastuner import HyperModel
from kerastuner.tuners import RandomSearch
# class MyHyperModel(HyperModel):
#
# def __init__(self,input_shape):
# self.input_shape = input_shape
#
# def build(self,hp):
# lr = hp.Float("lr",min_value=1e-5,max_value=1e-1,sampling="log")
# lambda_l2 = hp.Float("lambda_l2",min_value=1e-5,max_value=1e-1,sampling="log")
# X_input = Input(shape=self.input_shape)
# X = Dense(units=hp.Int('units0',min_value=8,max_value=32,step=4), activation='relu', kernel_regularizer=tfkreg.l2(lambda_l2))(X_input)
# X = BatchNormalization()(X)
# X = Dense(units=hp.Int('units1',min_value=16,max_value=64,step=4), activation='relu', kernel_regularizer=tfkreg.l2(lambda_l2))(X)
# X = BatchNormalization()(X)
# X = Dense(units=hp.Int('units2',min_value=8,max_value=32,step=4), activation='relu', kernel_regularizer=tfkreg.l2(lambda_l2))(X)
# X = BatchNormalization()(X)
# X = Dense(units=hp.Int('units3',min_value=4,max_value=32,step=4), activation='relu', kernel_regularizer=tfkreg.l2(lambda_l2))(X)
# X = BatchNormalization()(X)
# X = Dense(1, activation='sigmoid', kernel_regularizer=tfkreg.l2(lambda_l2))(X)
#
# model = Model(inputs=X_input, outputs=X)
# opt = Adam(learning_rate=lr)
# model.compile(optimizer=opt,loss='binary_crossentropy',metrics=["accuracy"])
# return model
#
#
# fileName = 'data/train_v7_1.mat'
# file = h5py.File(fileName) # "eventSteps","eventTimes","xdata","ydata","means","stds"
# xdata = np.array(file['xdata'])
# means = np.mean(xdata,axis=0)
# maxmins = np.max(xdata,axis=0)-np.min(xdata,axis=0)
# xdata = (xdata-means)/maxmins
# ydata = np.array(file['ydata'])
# eventTimes = file['times']
# eventSteps = np.array(file['eventSteps'])
# devnum = np.sum(eventSteps[0:40])
# testnum = np.sum(eventSteps[40:80])
# xdev = xdata[:devnum]
# ydev = ydata[:devnum]
# xtest = xdata[devnum:(devnum+testnum)]
# ytest = ydata[devnum:(devnum+testnum)]
# xtrain = xdata[(devnum+testnum):]
# ytrain = ydata[(devnum+testnum):]
#
# hypermodel = MyHyperModel(input_shape=xtrain.shape[1:])
#
# class MyTuner(kerastuner.tuners.BayesianOptimization):
# def run_trial(self, trial, *args, **kwargs):
# kwargs['batch_size'] = trial.hyperparameters.Int('batch_size',32,256,step=32)
# super(MyTuner,self).run_trial(trial,*args,**kwargs)
#
# tuner = MyTuner(
# hypermodel,
# objective='val_loss',
# max_trials=10,
# directory='my_dir',
# project_name='model_v7_2',
# )
# tuner.search(xtrain,ytrain,
# epochs=20,
# validation_data=(xdev,ydev),
# )
# tuner.results_summary()
# #tf.compat.v1.disable_v2_behavior() # model trained in tf1
# #model = tf.compat.v1.keras.models.load_model('./model/v1/my_model.h5')
def preprocess(fileName = 'data/train_v7_1.mat'):
file = h5py.File(fileName) # "eventSteps","eventTimes","xdata","ydata","means","stds"
xdata = np.array(file['xdata'])
means = np.mean(xdata,axis=0)
maxmins = np.max(xdata,axis=0)-np.min(xdata,axis=0)
xdata = (xdata-means)/maxmins
ydata = np.array(file['ydata'])
eventTimes = file['times']
eventSteps = np.array(file['eventSteps'])
devnum = np.sum(eventSteps[0:40])
testnum = np.sum(eventSteps[40:80])
xdev = xdata[:devnum]
ydev = ydata[:devnum]
xtest = xdata[devnum:(devnum+testnum)]
ytest = ydata[devnum:(devnum+testnum)]
xtrain = xdata[(devnum+testnum):]
ytrain = ydata[(devnum+testnum):]
return xtrain,ytrain,xdev,ydev,xtest,ytest
def model_v7(input_shape,params): #params: batch_size,lr,lambda_l2
X_input = Input(input_shape)
X = Dense(8, activation='relu',
kernel_regularizer=tfkreg.l2(params['lambda_l2']))(X_input)
X = BatchNormalization()(X)
X = Dense(16, activation='relu',
kernel_regularizer=tfkreg.l2(params['lambda_l2']))(X)
X = BatchNormalization()(X)
X = Dense(8, activation='relu',
kernel_regularizer=tfkreg.l2(params['lambda_l2']))(X)
X = BatchNormalization()(X)
X = Dense(4, activation='relu',
kernel_regularizer=tfkreg.l2(params['lambda_l2']))(X)
X = BatchNormalization()(X)
X = Dense(1, activation='sigmoid', kernel_regularizer=tfkreg.l2(params['lambda_l2']))(X)
model = Model(inputs=X_input, outputs=X)
return model
if __name__=='__main__':
classes = [0, 1]
xtrain,ytrain,xdev,ydev,xtest,ytest = preprocess()
################################# hyperopt model #####################################
from hyperopt import hp, STATUS_OK, Trials, fmin, tpe
from tensorflow.keras.callbacks import EarlyStopping
space = {
'lr': hp.loguniform('lr', -10, -3),
'lambda_l2': hp.loguniform('lambda_l2', -10, -3),
'batch_size': hp.choice('batch_size', [256,])
}
f1 = 0
workidx = 1
print('work {}'.format(workidx))
maxtrailnum = 20
def trainAmodel(params):
global xtrain, xdev, ytrain, ydev
global f1, workidx
print('Params testing: ', params)
aModel = model_v7(xtrain.shape[1:], params)
opt = tensorflow.keras.optimizers.Adam(lr=params['lr'])
aModel.compile(loss="binary_crossentropy", metrics=['accuracy'], optimizer=opt)
steps_per_epoch = (np.shape(xtrain)[0] + params['batch_size'] - 1) // params['batch_size']
# metrics = V1_utils.Metrics(test_data=(X_test[::10], Y_test[::10]), train_data=(X_train[::100], Y_train[::100]))
from sklearn.utils import class_weight
import pandas as pd
class_weight = class_weight.compute_class_weight(class_weight='balanced',
classes=classes,
y=ytrain[:, 0])
cw = dict(enumerate(class_weight))
# early_stopping = EarlyStopping(monitor='val_loss', patience=8, min_delta=0.8, mode='min')
# history = aModel.fit_generator(generator=data_generator(xtrain, ytrain, params['batch_size']),
# steps_per_epoch=steps_per_epoch,
# epochs=30,
# verbose=0,
# validation_data=(xdev[::80], ydev[::80]),
# callbacks=[early_stopping],
# # callbacks=[metrics],
# class_weight=cw,
# )
history = aModel.fit(xtrain, ytrain,
batch_size=params['batch_size'],
epochs=30,
verbose=0,
# validation_data=(xdev[::10000], ydev[::10000]),
class_weight=cw,
)
# 评估模型
batch_size_test = 32
# preds = aModel.evaluate_generator(generator=data_generator(xdev, ydev, batch_size_test, cycle=False),verbose=0)
preds = aModel.evaluate(xdev, ydev, batch_size=batch_size_test, verbose=0)
print("误差值 = " + str(preds[0]))
print("准确度 = " + str(preds[1]))
# cvres = aModel.predict_generator(data_generator(xdev,None,4,cycle=False,givey=False), verbose=0)
# cvf1s, cache = V1_utils.fmeasure(ydev, cvres)
cvres = aModel.predict(xdev, batch_size=batch_size_test, verbose=0)
cvf1s, cache = V1_utils.fmeasure(ydev, cvres)
p, r = cache
print("f1 = {}, precision = {}, recall = {}".format(cvf1s, p, r))
if cvf1s > f1:
f1 = cvf1s
aModel.save('model/v7/model_v7_1_{}.h5'.format(workidx))
print(f1)
plt.figure()
plt.plot(history.history['loss'], 'b', label='Training loss')
# plt.plot(history.history['val_loss'], 'r', label='Validation val_loss')
plt.title('Traing loss')
plt.legend()
plt.xlabel('epoch')
plt.ylabel('loss')
plt.savefig('image/v7/v7_1/log/loss_v2_3_{}.jpg'.format(workidx))
return {
'loss': -cvf1s,
'status': STATUS_OK
}
trials = Trials()
best = fmin(trainAmodel, space, algo=tpe.suggest, max_evals=maxtrailnum, trials=trials)
filename = 'model/v7/log_v7_{}.npz'.format(workidx)
np.savez(filename, trials=trials, best=best)
print('best')
print(best)
trialNum = len(trials.trials)
l2s = np.zeros(trialNum)
lrs = np.zeros(trialNum)
losses = np.zeros(trialNum)
bzs = np.zeros(trialNum)
for trialidx in range(trialNum):
thevals = trials.trials[trialidx]['misc']['vals'] #如果是从文件中读取，这一行改成trials[]，即不需要后面那个.trails,下面losses那一行同理
l2s[trialidx] = thevals['lambda_l2'][0]
lrs[trialidx] = thevals['lr'][0]
bzs[trialidx] = (thevals['batch_size'][0] + 1)
losses[trialidx] = -trials.trials[trialidx]['result']['loss']
plt.figure()
# plt.scatter(np.log(lrs), np.log(l2s), c=bzs, s=losses * 100, cmap=mpl.colors.ListedColormap(
# ["darkorange", "gold", "lawngreen", "lightseagreen"]
# ))
plt.scatter(np.log(lrs), np.log(l2s), c=losses, )
plt.xlabel('ln[lr]')
plt.ylabel('ln[λ]')
plt.title('f1')
cb = plt.colorbar()
# cb.set_label('log2[BatchSize]', labelpad=-1)
plt.savefig('image/v7/v7_1/hyparams_v7_{}.jpg'.format(workidx))
print('done')