-
Notifications
You must be signed in to change notification settings - Fork 2
/
experimentAB.py
346 lines (294 loc) · 15.1 KB
/
experimentAB.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
from keras.backend.tensorflow_backend import set_session
import tensorflow as tf
# set the memory usage
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
set_session(tf.Session(config=tf_config))
import matplotlib.pyplot as plt
from scipy.io import loadmat
import numpy as np
import imp
import pandas as pd
#
import cwru
window_size = 2048
data = cwru.CWRU(['12DriveEndFault'], ['1772', '1750', '1730'], window_size)
print("data.nclasses:",data.nclasses,"data.classes:",data.classes)
print("len(data.X_train):",len(data.X_train),"len(data.X_test):",len(data.X_test))
#
import models
# # imp.reload(models)
# siamese_net = models.load_siamese_net((window_size,2))
# print('\nsiamese_net summary:')
# siamese_net.summary()
# #
# print('\nsequential_3 is WDCNN:')
# siamese_net.layers[2].summary()
# #
# wdcnn_net = models.load_wdcnn_net()
# print('\nwdcnn_net summary:')
# wdcnn_net.summary()
#
import keras
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split
from keras.callbacks import ModelCheckpoint,EarlyStopping
import siamese
imp.reload(siamese)
import utils
imp.reload(utils)
snrs = [-4,-2,0,2,4,6,8,10,None]
settings = {
"N_way": 10, # how many classes for testing one-shot tasks>
"batch_size": 32,
"best": -1,
"evaluate_every": 200, # interval for evaluating on one-shot tasks
"loss_every": 20, # interval for printing loss (iterations)
"n_iter": 15000,
"n_val": 2, #how many one-shot tasks to validate on?
"n": 0,
"save_path":"",
"save_weights_file": "weights-best-10-oneshot-low-data.hdf5"
}
exp_name = "EXP-AB"
exps = [60,90,120,200,300,600,900,1500,6000]
#exps = [1980,6000]
times = 20
#缺少了1500 第9次
is_training = True # enable or disable train models. if enable training, save best models will be update.
def EXPAB_train_and_test(exp_name, exps, is_training):
train_classes = sorted(list(set(data.y_train)))
train_indices = [np.where(data.y_train == i)[0] for i in train_classes]
for exp in exps: #总的训练样本数
scores_1_shot = []
scores_5_shot = []
scores_5_shot_prod = []
scores_wdcnn = []
num = int(exp / len(train_classes))
settings['evaluate_every'] = 300 if exp < 1000 else 600
print("settings['evaluate_every']:",settings['evaluate_every'])
for time_idx in range(10): #重复20次，每次的随机种子不一样
seed = int(time_idx / 4) * 10
np.random.seed(seed)
print('random seed:', seed)
print("\n样本数%s-第%s次训练" % (exp, time_idx) + '*' * 80)
#路径：temp/EXP-AB/size_exp/time_time_idx
settings["save_path"] = "tmp/%s/size_%s/time_%s/" % (exp_name, exp, time_idx)
data._mkdir(settings["save_path"])
train_idxs = []
val_idxs = []
for i, c in enumerate(train_classes):
select_idx = train_indices[i][np.random.choice(len(train_indices[i]), num, replace=False)]
split = int(0.6 * num) #使用60%样本作为训练集
train_idxs.extend(select_idx[:split])
val_idxs.extend(select_idx[split:])
X_train, y_train = data.X_train[train_idxs], data.y_train[train_idxs],
X_val, y_val = data.X_train[val_idxs], data.y_train[val_idxs],
print("训练集前10个元素的下标:",train_idxs[0:10])
print("验证集前10个元素的下标:",val_idxs[0:10])
# load one-shot model and training 修改的
siamese_net = models.load_siamese_net_my_mew()
siamese_loader = siamese.Siamese_Loader(X_train,
y_train,
X_val,
y_val)
if (is_training):
print(siamese.train_and_test_oneshot(settings, siamese_net, siamese_loader))
# 将具体类别数转换为 向量化
y_train = keras.utils.to_categorical(y_train, data.nclasses)
y_val = keras.utils.to_categorical(y_val, data.nclasses)
y_test = keras.utils.to_categorical(data.y_test, data.nclasses)
earlyStopping = EarlyStopping(monitor='val_loss', patience=20, verbose=0, mode='min')
# checkpoint
# filepath="tmp/weights-best-cnn-{epoch:02d}-{val_acc:.2f}.hdf5"
filepath = "%sweights-best-10-cnn-low-data.hdf5" % (settings["save_path"])
checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=0, save_best_only=True, mode='max')
callbacks_list = [earlyStopping, checkpoint]
wdcnn_net = models.load_wdcnn_net()
if (is_training):
wdcnn_net.fit(X_train, y_train,
batch_size=32,
epochs=300,
verbose=0,
callbacks=callbacks_list,
validation_data=(X_val, y_val))
# loading best weights and testing
print("load best weights", settings["save_path"] + settings['save_weights_file'])
siamese_net.load_weights(settings["save_path"] + settings['save_weights_file'])
print("load best weights", filepath)
wdcnn_net.load_weights(filepath)
for snr in snrs:
print("\n样本数%s_第%s次训练_噪音为%s" % (exp, time_idx, snr) + '*' * 80)
X_test_noise = []
if snr != None:
for x in data.X_test:
X_test_noise.append(utils.noise_rw(x, snr))
X_test_noise = np.array(X_test_noise)
else:
X_test_noise = data.X_test
# test 1_shot and 5_shot
siamese_loader.set_val(X_test_noise, data.y_test)
s = 'val'
preds_5_shot = []
prods_5_shot = []
scores = []
for k in range(5):
#val_acc 是一个ontshot的正确率，preds是预测类别和真实类别的组合
#prods是 预测某一个样本在 10个类别上的概率
val_acc, preds, prods = siamese_loader.test_oneshot2(siamese_net, N=len(siamese_loader.classes[s]),
k=len(siamese_loader.data[s]), verbose=False)
# utils.confusion_plot(preds[:,1],preds[:,0])
print("测试集的正确率:",val_acc, preds.shape, prods.shape)
scores.append(val_acc)
preds_5_shot.append(preds[:, 1]) #list 每个是ndarray shape=750
prods_5_shot.append(prods) #list 每个是ndarray shape=750 10 1
preds = []
for line in np.array(preds_5_shot).T:
pass
#np.argmax(np.bincount(line)) 就是 一个样本 5次ont-shot 中选择出现次数最多的
#预测类别比如54555 那么就选5 ，代表是第5类
preds.append(np.argmax(np.bincount(line)))
# utils.confusion_plot(np.array(preds),data.y_test)
#这里就是 某个样本 进行5次预测，每次预测产生 10个类别的概率，5次 10个类别概率相加找最大的
prod_preds = np.argmax(np.sum(prods_5_shot, axis=0), axis=1).reshape(-1)
score_5_shot = accuracy_score(data.y_test, np.array(preds)) * 100
print('5_shot:', score_5_shot)
score_5_shot_prod = accuracy_score(data.y_test, prod_preds) * 100
print('5_shot_prod:', score_5_shot_prod)
scores_1_shot.append(scores[0])
scores_5_shot.append(score_5_shot)
scores_5_shot_prod.append(score_5_shot_prod)
# test wdcnn
score = wdcnn_net.evaluate(X_test_noise, y_test, verbose=0)[1] * 100
print('wdcnn:', score)
scores_wdcnn.append(score)
a = pd.DataFrame(np.array(scores_1_shot).reshape(-1, len(snrs)))
a.columns = snrs
a.to_csv("tmp/%s/size_%s/scores_1_shot.csv" % (exp_name, exp), index=True)
a = pd.DataFrame(np.array(scores_5_shot).reshape(-1, len(snrs)))
a.columns = snrs
a.to_csv("tmp/%s/size_%s/scores_5_shot.csv" % (exp_name, exp), index=True)
a = pd.DataFrame(np.array(scores_5_shot_prod).reshape(-1, len(snrs)))
a.columns = snrs
a.to_csv("tmp/%s/size_%s/scores_5_shot_prod.csv" % (exp_name, exp), index=True)
a = pd.DataFrame(np.array(scores_wdcnn).reshape(-1, len(snrs)))
a.columns = snrs
a.to_csv("tmp/%s/size_%s/scores_wdcnn.csv" % (exp_name, exp), index=True)
EXPAB_train_and_test(exp_name, exps, is_training)
#
np.bincount([2,2,3,3,1])
#
def EXPAB_analysis(exp_name, exps):
scores_1_shot_all = pd.DataFrame()
scores_5_shot_all = pd.DataFrame()
scores_5_shot_prod_all = pd.DataFrame()
scores_wdcnn_all = pd.DataFrame()
for exp in exps:
file_path = "tmp/%s/size_%s" % (exp_name, exp)
tmp_data = pd.read_csv("%s/scores_1_shot.csv" % (file_path),
sep=',', index_col=0)
tmp_data['exp'] = exp
scores_1_shot_all = pd.concat([scores_1_shot_all, tmp_data], axis=0)
tmp_data = pd.read_csv("%s/scores_5_shot.csv" % (file_path),
sep=',', index_col=0)
tmp_data['exp'] = exp
scores_5_shot_all = pd.concat([scores_5_shot_all, tmp_data], axis=0)
tmp_data = pd.read_csv("%s/scores_5_shot_prod.csv" % (file_path),
sep=',', index_col=0)
tmp_data['exp'] = exp
scores_5_shot_prod_all = pd.concat([scores_5_shot_prod_all, tmp_data], axis=0)
tmp_data = pd.read_csv("%s/scores_wdcnn.csv" % (file_path),
sep=',', index_col=0)
tmp_data['exp'] = exp
scores_wdcnn_all = pd.concat([scores_wdcnn_all, tmp_data], axis=0)
scores_1_shot_all.to_csv("tmp/%s/scores_1_shot_all.csv" % (exp_name), float_format='%.6f', index=True)
scores_5_shot_all.to_csv("tmp/%s/scores_5_shot_all.csv" % (exp_name), float_format='%.6f', index=True)
scores_5_shot_prod_all.to_csv("tmp/%s/scores_5_shot_prob_all.csv" % (exp_name), float_format='%.6f', index=True)
scores_wdcnn_all.to_csv("tmp/%s/scores_wdcnn_all.csv" % (exp_name), float_format='%.6f', index=True)
scores_1_shot_all['model'] = 'One-shot'
scores_5_shot_all['model'] = 'Five-shot'
scores_5_shot_prod_all['model'] = 'Five-shot-prob'
scores_wdcnn_all['model'] = 'WDCNN'
scores_all = pd.concat([scores_1_shot_all, scores_5_shot_all, scores_5_shot_prod_all, scores_wdcnn_all], axis=0)
scores_all.to_csv("tmp/%s/scores_all.csv" % (exp_name), float_format='%.6f', index=True)
return scores_all
#
# analysis
scores_all = EXPAB_analysis(exp_name,exps)
scores_all_mean = scores_all.groupby(['model','exp']).mean()
scores_all_std = scores_all.groupby(['model','exp']).std()
scores_all_mean.to_csv("tmp/%s/scores_all_mean.csv" % (exp_name), float_format='%.2f', index=True)
scores_all_std.to_csv("tmp/%s/scores_all_std.csv" % (exp_name), float_format='%.2f', index=True)
scores_all_mean, scores_all_std
#
from sklearn.metrics import accuracy_score
import keras
num = 90
train_classes = sorted(list(set(data.y_train)))
train_indices = [np.where(data.y_train == i)[0] for i in train_classes]
train_idxs = []
val_idxs = []
for i, c in enumerate(train_classes):
select_idx = train_indices[i][np.random.choice(len(train_indices[i]), num, replace=False)]
split = int(0.6 * num)
train_idxs.extend(select_idx[:split])
val_idxs.extend(select_idx[split:])
X_train, y_train = data.X_train[train_idxs], data.y_train[train_idxs]
X_val, y_val = data.X_train[val_idxs], data.y_train[val_idxs]
siamese_loader = siamese.Siamese_Loader(X_train,
y_train,
data.X_test,
data.y_test)
siamese_net = models.load_siamese_net()
wdcnn_net = models.load_wdcnn_net()
settings["save_path"] = "tmp/%s/size_%s/time_%s/" % (exp_name, num, 0)
siamese_net.load_weights(settings["save_path"] + settings['save_weights_file'])
wdcnn_net.load_weights("%s/weights-best-10-cnn-low-data.hdf5" % (settings["save_path"]))
y_test = keras.utils.to_categorical(data.y_test, data.nclasses)
#
from keras import backend as K
import numpy as np
try: from sklearn.manifold import TSNE; HAS_SK = True
except: HAS_SK = False; print('Please install sklearn for layer visualization')
intermediate_tensor_function = K.function([siamese_net.layers[2].layers[0].input],
[siamese_net.layers[2].layers[-1].output])
plot_only = len(data.y_test)
intermediate_tensor = intermediate_tensor_function([data.X_test[0:plot_only]])[0]
# Visualization of trained flatten layer (T-SNE)
tsne = TSNE(perplexity=30, n_components=2, init='pca', n_iter=5000)
low_dim_embs = tsne.fit_transform(intermediate_tensor)
p_data = pd.DataFrame(columns=['x', 'y', 'label'])
p_data.x = low_dim_embs[:, 0]
p_data.y = low_dim_embs[:, 1]
p_data.label = data.y_test[0:plot_only]
utils.plot_with_labels(p_data)
plt.savefig("%s/90-tsne-one-shot.pdf" % (settings["save_path"]))
#
from keras import backend as K
import numpy as np
intermediate_tensor_function = K.function([wdcnn_net.layers[1].layers[0].input],
[wdcnn_net.layers[1].layers[-1].output])
plot_only = len(data.y_test)
intermediate_tensor = intermediate_tensor_function([data.X_test[0:plot_only]])[0]
# Visualization of trained flatten layer (T-SNE)
tsne = TSNE(perplexity=30, n_components=2, init='pca', n_iter=5000)
low_dim_embs = tsne.fit_transform(intermediate_tensor)
import pandas as pd
p_data = pd.DataFrame(columns=['x', 'y', 'label'])
p_data.x = low_dim_embs[:, 0]
p_data.y = low_dim_embs[:, 1]
p_data.label = data.y_test[0:plot_only]
utils.plot_with_labels(p_data)
plt.savefig("%s/90-tsne-wdcnn.pdf" % (settings["save_path"]))
#
from sklearn.metrics import f1_score,accuracy_score,confusion_matrix
s = 'val'
val_acc,preds, probs_all = siamese_loader.test_oneshot2(siamese_net,len(siamese_loader.classes[s]),len(siamese_loader.data[s]),verbose=False)
# utils.confusion_plot(preds[:,1],preds[:,0])
utils.plot_confusion_matrix(confusion_matrix(data.y_test,preds[:,1]), normalize=False, title=None)
plt.savefig("%s/90-cm-one-shot.pdf" % (settings["save_path"]))
#
pred = np.argmax(wdcnn_net.predict(data.X_test), axis=1).reshape(-1,1)
# utils.confusion_plot(pred,data.y_test)
utils.plot_confusion_matrix(confusion_matrix(data.y_test,pred), normalize=False,title=None)
plt.savefig("%s/90-cm-wdcnn.pdf" % (settings["save_path"]))