一种Dynamic ReLU：自适应参数化ReLU（调参记录2）

续前一篇：
一种Dynamic ReLU激活函数：自适应参数化ReLU（调参记录1）

自适应参数化ReLU是一种动态ReLU（Dynamic ReLU），于2019年5月3日投稿到IEEE Transactions on Industrial Electronics，于2020年1月24日（农历大年初一）录用，于2020年2月13日在IEEE官网公布。

本文依然是测试ResNet+自适应参数化ReLU，残差模块的个数增加到了27个，其他保持不变，继续测试它在Cifar10的分类表现。

自适应参数化ReLU是Parametric ReLU的一种动态化改进，原本是应用在基于振动信号的机械故障诊断，其基本原理如下：

具体Keras代码：

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Tue Apr 14 04:17:45 2020
Implemented using TensorFlow 1.10.0 and Keras 2.2.1

Minghang Zhao, Shisheng Zhong, Xuyun Fu, Baoping Tang, Shaojiang Dong, Michael Pecht,
Deep Residual Networks with Adaptively Parametric Rectifier Linear Units for Fault Diagnosis, 
IEEE Transactions on Industrial Electronics, 2020, DOI: 10.1109/TIE.2020.2972458
Date of Publication: 13 February 2020

@author: Minghang Zhao
"""

from __future__ import print_function
import keras
import numpy as np
from keras.datasets import cifar10
from keras.layers import Dense, Conv2D, BatchNormalization, Activation, Minimum
from keras.layers import AveragePooling2D, Input, GlobalAveragePooling2D, Concatenate, Reshape
from keras.regularizers import l2
from keras import backend as K
from keras.models import Model
from keras import optimizers
from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import LearningRateScheduler
K.set_learning_phase(1)

# The data, split between train and test sets
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.
x_test = x_test-np.mean(x_train)
x_train = x_train-np.mean(x_train)
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, 10)
y_test = keras.utils.to_categorical(y_test, 10)

# Schedule the learning rate, multiply 0.1 every 400 epoches
def scheduler(epoch):
    if epoch % 400 == 0 and epoch != 0:
        lr = K.get_value(model.optimizer.lr)
        K.set_value(model.optimizer.lr, lr * 0.1)
        print("lr changed to {}".format(lr * 0.1))
    return K.get_value(model.optimizer.lr)

# An adaptively parametric rectifier linear unit (APReLU)
def aprelu(inputs):
    # get the number of channels
    channels = inputs.get_shape().as_list()[-1]
    # get a zero feature map
    zeros_input = keras.layers.subtract([inputs, inputs])
    # get a feature map with only positive features
    pos_input = Activation('relu')(inputs)
    # get a feature map with only negative features
    neg_input = Minimum()([inputs,zeros_input])
    # define a network to obtain the scaling coefficients
    scales_p = GlobalAveragePooling2D()(pos_input)
    scales_n = GlobalAveragePooling2D()(neg_input)
    scales = Concatenate()([scales_n, scales_p])
    scales = Dense(channels//4, activation='linear', kernel_initializer='he_normal', kernel_regularizer=l2(1e-4))(scales)
    scales = BatchNormalization()(scales)
    scales = Activation('relu')(scales)
    scales = Dense(channels, activation='linear', kernel_initializer='he_normal', kernel_regularizer=l2(1e-4))(scales)
    scales = BatchNormalization()(scales)
    scales = Activation('sigmoid')(scales)
    scales = Reshape((1,1,channels))(scales)
    # apply a paramtetric relu
    neg_part = keras.layers.multiply([scales, neg_input])
    return keras.layers.add([pos_input, neg_part])

# Residual Block
def residual_block(incoming, nb_blocks, out_channels, downsample=False,
                   downsample_strides=2):
    
    residual = incoming
    in_channels = incoming.get_shape().as_list()[-1]
    
    for i in range(nb_blocks):
        
        identity = residual
        
        if not downsample:
            downsample_strides = 1
        
        residual = BatchNormalization()(residual)
        residual = aprelu(residual)
        residual = Conv2D(out_channels, 3, strides=(downsample_strides, downsample_strides), 
                          padding='same', kernel_initializer='he_normal', 
                          kernel_regularizer=l2(1e-4))(residual)
        
        residual = BatchNormalization()(residual)
        residual = aprelu(residual)
        residual = Conv2D(out_channels, 3, padding='same', kernel_initializer='he_normal', 
                          kernel_regularizer=l2(1e-4))(residual)
        
        # Downsampling
        if downsample_strides > 1:
            identity = AveragePooling2D(pool_size=(1,1), strides=(2,2))(identity)
            
        # Zero_padding to match channels
        if in_channels != out_channels:
            zeros_identity = keras.layers.subtract([identity, identity])
            identity = keras.layers.concatenate([identity, zeros_identity])
            in_channels = out_channels
        
        residual = keras.layers.add([residual, identity])
    
    return residual


# define and train a model
inputs = Input(shape=(32, 32, 3))
net = Conv2D(8, 3, padding='same', kernel_initializer='he_normal', kernel_regularizer=l2(1e-4))(inputs)
net = residual_block(net, 9, 8, downsample=False)
net = residual_block(net, 1, 16, downsample=True)
net = residual_block(net, 8, 16, downsample=False)
net = residual_block(net, 1, 32, downsample=True)
net = residual_block(net, 8, 32, downsample=False)
net = BatchNormalization()(net)
net = aprelu(net)
net = GlobalAveragePooling2D()(net)
outputs = Dense(10, activation='softmax', kernel_initializer='he_normal', kernel_regularizer=l2(1e-4))(net)
model = Model(inputs=inputs, outputs=outputs)
sgd = optimizers.SGD(lr=0.1, decay=0., momentum=0.9, nesterov=True)
model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])

# data augmentation
datagen = ImageDataGenerator(
    # randomly rotate images in the range (deg 0 to 180)
    rotation_range=30,
    # randomly flip images
    horizontal_flip=True,
    # randomly shift images horizontally
    width_shift_range=0.125,
    # randomly shift images vertically
    height_shift_range=0.125)

reduce_lr = LearningRateScheduler(scheduler)
# fit the model on the batches generated by datagen.flow().
model.fit_generator(datagen.flow(x_train, y_train, batch_size=100),
                    validation_data=(x_test, y_test), epochs=1000, 
                    verbose=1, callbacks=[reduce_lr], workers=4)

# get results
K.set_learning_phase(0)
DRSN_train_score1 = model.evaluate(x_train, y_train, batch_size=100, verbose=0)
print('Train loss:', DRSN_train_score1[0])
print('Train accuracy:', DRSN_train_score1[1])
DRSN_test_score1 = model.evaluate(x_test, y_test, batch_size=100, verbose=0)
print('Test loss:', DRSN_test_score1[0])
print('Test accuracy:', DRSN_test_score1[1])

实验结果如下:

Epoch 272/1000
53s 105ms/step - loss: 0.6071 - acc: 0.8711 - val_loss: 0.6295 - val_acc: 0.8667
Epoch 273/1000
53s 105ms/step - loss: 0.6078 - acc: 0.8705 - val_loss: 0.6373 - val_acc: 0.8678
Epoch 274/1000
53s 106ms/step - loss: 0.6043 - acc: 0.8714 - val_loss: 0.6245 - val_acc: 0.8686
Epoch 275/1000
52s 105ms/step - loss: 0.6056 - acc: 0.8720 - val_loss: 0.6228 - val_acc: 0.8713
Epoch 276/1000
52s 105ms/step - loss: 0.6059 - acc: 0.8730 - val_loss: 0.6104 - val_acc: 0.8730
Epoch 277/1000
52s 105ms/step - loss: 0.5980 - acc: 0.8756 - val_loss: 0.6265 - val_acc: 0.8671
Epoch 278/1000
52s 105ms/step - loss: 0.6093 - acc: 0.8716 - val_loss: 0.6363 - val_acc: 0.8617
Epoch 279/1000
53s 105ms/step - loss: 0.6051 - acc: 0.8716 - val_loss: 0.6355 - val_acc: 0.8650
Epoch 280/1000
53s 105ms/step - loss: 0.6062 - acc: 0.8725 - val_loss: 0.6227 - val_acc: 0.8669
Epoch 281/1000
52s 105ms/step - loss: 0.6025 - acc: 0.8731 - val_loss: 0.6156 - val_acc: 0.8723
Epoch 282/1000
53s 105ms/step - loss: 0.6031 - acc: 0.8725 - val_loss: 0.6450 - val_acc: 0.8630
Epoch 283/1000
52s 104ms/step - loss: 0.6030 - acc: 0.8745 - val_loss: 0.6282 - val_acc: 0.8688
Epoch 284/1000
52s 104ms/step - loss: 0.6049 - acc: 0.8717 - val_loss: 0.6213 - val_acc: 0.8693
Epoch 285/1000
53s 105ms/step - loss: 0.6005 - acc: 0.8709 - val_loss: 0.6208 - val_acc: 0.8682
Epoch 286/1000
52s 104ms/step - loss: 0.6049 - acc: 0.8718 - val_loss: 0.6420 - val_acc: 0.8647
Epoch 287/1000
53s 105ms/step - loss: 0.6040 - acc: 0.8728 - val_loss: 0.6188 - val_acc: 0.8694
Epoch 288/1000
53s 105ms/step - loss: 0.6011 - acc: 0.8741 - val_loss: 0.6548 - val_acc: 0.8577
Epoch 289/1000
53s 105ms/step - loss: 0.6060 - acc: 0.8731 - val_loss: 0.6163 - val_acc: 0.8717
Epoch 290/1000
53s 105ms/step - loss: 0.6047 - acc: 0.8717 - val_loss: 0.6172 - val_acc: 0.8733
Epoch 291/1000
53s 105ms/step - loss: 0.6029 - acc: 0.8728 - val_loss: 0.6319 - val_acc: 0.8639
Epoch 292/1000
52s 105ms/step - loss: 0.6011 - acc: 0.8742 - val_loss: 0.6237 - val_acc: 0.8664
Epoch 293/1000
53s 105ms/step - loss: 0.5998 - acc: 0.8741 - val_loss: 0.6410 - val_acc: 0.8646
Epoch 294/1000
52s 105ms/step - loss: 0.6001 - acc: 0.8736 - val_loss: 0.6435 - val_acc: 0.8644
Epoch 295/1000
53s 106ms/step - loss: 0.6022 - acc: 0.8730 - val_loss: 0.6233 - val_acc: 0.8657
Epoch 296/1000
53s 106ms/step - loss: 0.6015 - acc: 0.8746 - val_loss: 0.6224 - val_acc: 0.8665
Epoch 297/1000
52s 105ms/step - loss: 0.5995 - acc: 0.8750 - val_loss: 0.6471 - val_acc: 0.8613
Epoch 298/1000
53s 106ms/step - loss: 0.5992 - acc: 0.8735 - val_loss: 0.6436 - val_acc: 0.8635
Epoch 299/1000
53s 106ms/step - loss: 0.6040 - acc: 0.8716 - val_loss: 0.6273 - val_acc: 0.8674
Epoch 300/1000
52s 105ms/step - loss: 0.6008 - acc: 0.8736 - val_loss: 0.6543 - val_acc: 0.8603
Epoch 301/1000
52s 104ms/step - loss: 0.6023 - acc: 0.8732 - val_loss: 0.6420 - val_acc: 0.8633
Epoch 302/1000
52s 105ms/step - loss: 0.5992 - acc: 0.8747 - val_loss: 0.6125 - val_acc: 0.8712
Epoch 303/1000
52s 105ms/step - loss: 0.6016 - acc: 0.8743 - val_loss: 0.6402 - val_acc: 0.8660
Epoch 304/1000
53s 105ms/step - loss: 0.5998 - acc: 0.8742 - val_loss: 0.6256 - val_acc: 0.8663
Epoch 305/1000
53s 105ms/step - loss: 0.5998 - acc: 0.8736 - val_loss: 0.6193 - val_acc: 0.8713
Epoch 306/1000
52s 105ms/step - loss: 0.5977 - acc: 0.8760 - val_loss: 0.6219 - val_acc: 0.8686
Epoch 307/1000
52s 104ms/step - loss: 0.6000 - acc: 0.8743 - val_loss: 0.6643 - val_acc: 0.8539
Epoch 308/1000
53s 105ms/step - loss: 0.6022 - acc: 0.8740 - val_loss: 0.6308 - val_acc: 0.8671
Epoch 309/1000
52s 104ms/step - loss: 0.6083 - acc: 0.8737 - val_loss: 0.6168 - val_acc: 0.8730
Epoch 310/1000
52s 104ms/step - loss: 0.6008 - acc: 0.8727 - val_loss: 0.6165 - val_acc: 0.8751
Epoch 311/1000
53s 105ms/step - loss: 0.6046 - acc: 0.8731 - val_loss: 0.6369 - val_acc: 0.8639
Epoch 312/1000
53s 106ms/step - loss: 0.5976 - acc: 0.8753 - val_loss: 0.6246 - val_acc: 0.8695
Epoch 313/1000
53s 105ms/step - loss: 0.6037 - acc: 0.8738 - val_loss: 0.6266 - val_acc: 0.8691
Epoch 314/1000
52s 105ms/step - loss: 0.6007 - acc: 0.8732 - val_loss: 0.6520 - val_acc: 0.8631
Epoch 315/1000
52s 105ms/step - loss: 0.5993 - acc: 0.8751 - val_loss: 0.6436 - val_acc: 0.8632
Epoch 316/1000
52s 105ms/step - loss: 0.5996 - acc: 0.8750 - val_loss: 0.6413 - val_acc: 0.8589
Epoch 317/1000
53s 105ms/step - loss: 0.5998 - acc: 0.8740 - val_loss: 0.6406 - val_acc: 0.8621
Epoch 318/1000
52s 105ms/step - loss: 0.5992 - acc: 0.8753 - val_loss: 0.6364 - val_acc: 0.8614
Epoch 319/1000
53s 105ms/step - loss: 0.5983 - acc: 0.8748 - val_loss: 0.6275 - val_acc: 0.8650
Epoch 320/1000
53s 105ms/step - loss: 0.5987 - acc: 0.8766 - val_loss: 0.6207 - val_acc: 0.8724
Epoch 321/1000
52s 105ms/step - loss: 0.5979 - acc: 0.8756 - val_loss: 0.6266 - val_acc: 0.8711
Epoch 322/1000
52s 105ms/step - loss: 0.5981 - acc: 0.8748 - val_loss: 0.6461 - val_acc: 0.8627
Epoch 323/1000
53s 105ms/step - loss: 0.5966 - acc: 0.8757 - val_loss: 0.6235 - val_acc: 0.8696
Epoch 324/1000
52s 105ms/step - loss: 0.5940 - acc: 0.8758 - val_loss: 0.6141 - val_acc: 0.8750
Epoch 325/1000
52s 105ms/step - loss: 0.6007 - acc: 0.8757 - val_loss: 0.6513 - val_acc: 0.8610
Epoch 326/1000
53s 105ms/step - loss: 0.5988 - acc: 0.8760 - val_loss: 0.6219 - val_acc: 0.8724
Epoch 327/1000
52s 105ms/step - loss: 0.6003 - acc: 0.8744 - val_loss: 0.6115 - val_acc: 0.8693
Epoch 328/1000
53s 105ms/step - loss: 0.5942 - acc: 0.8762 - val_loss: 0.6358 - val_acc: 0.8660
Epoch 329/1000
52s 105ms/step - loss: 0.5923 - acc: 0.8769 - val_loss: 0.6340 - val_acc: 0.8672
Epoch 330/1000
53s 105ms/step - loss: 0.5954 - acc: 0.8781 - val_loss: 0.6246 - val_acc: 0.8688
Epoch 331/1000
52s 105ms/step - loss: 0.6015 - acc: 0.8747 - val_loss: 0.6194 - val_acc: 0.8710
Epoch 332/1000
52s 104ms/step - loss: 0.5980 - acc: 0.8764 - val_loss: 0.6311 - val_acc: 0.8685
Epoch 333/1000
52s 105ms/step - loss: 0.6019 - acc: 0.8748 - val_loss: 0.6095 - val_acc: 0.8733
Epoch 334/1000
53s 106ms/step - loss: 0.5964 - acc: 0.8760 - val_loss: 0.6515 - val_acc: 0.8623
Epoch 335/1000
53s 106ms/step - loss: 0.5973 - acc: 0.8765 - val_loss: 0.6300 - val_acc: 0.8702
Epoch 336/1000
53s 105ms/step - loss: 0.5953 - acc: 0.8776 - val_loss: 0.6297 - val_acc: 0.8656
Epoch 337/1000
53s 105ms/step - loss: 0.6005 - acc: 0.8752 - val_loss: 0.6252 - val_acc: 0.8711
Epoch 338/1000
53s 105ms/step - loss: 0.5949 - acc: 0.8778 - val_loss: 0.6175 - val_acc: 0.8693
Epoch 339/1000
52s 105ms/step - loss: 0.5996 - acc: 0.8749 - val_loss: 0.6215 - val_acc: 0.8688
Epoch 340/1000
52s 104ms/step - loss: 0.5921 - acc: 0.8777 - val_loss: 0.6239 - val_acc: 0.8713
Epoch 341/1000
52s 105ms/step - loss: 0.5910 - acc: 0.8776 - val_loss: 0.6327 - val_acc: 0.8684
Epoch 342/1000
52s 104ms/step - loss: 0.5952 - acc: 0.8778 - val_loss: 0.6083 - val_acc: 0.8767
Epoch 343/1000
53s 105ms/step - loss: 0.5965 - acc: 0.8763 - val_loss: 0.6312 - val_acc: 0.8696
Epoch 344/1000
53s 105ms/step - loss: 0.5965 - acc: 0.8771 - val_loss: 0.6204 - val_acc: 0.8707
Epoch 345/1000
52s 105ms/step - loss: 0.5932 - acc: 0.8764 - val_loss: 0.6211 - val_acc: 0.8709
Epoch 346/1000
53s 105ms/step - loss: 0.5900 - acc: 0.8785 - val_loss: 0.6422 - val_acc: 0.8663
Epoch 347/1000
53s 106ms/step - loss: 0.5919 - acc: 0.8775 - val_loss: 0.6437 - val_acc: 0.8646
Epoch 348/1000
53s 105ms/step - loss: 0.6001 - acc: 0.8753 - val_loss: 0.6184 - val_acc: 0.8709
Epoch 349/1000
53s 105ms/step - loss: 0.5952 - acc: 0.8778 - val_loss: 0.6410 - val_acc: 0.8626
Epoch 350/1000
53s 105ms/step - loss: 0.5946 - acc: 0.8768 - val_loss: 0.6321 - val_acc: 0.8660
Epoch 351/1000
53s 106ms/step - loss: 0.5931 - acc: 0.8770 - val_loss: 0.6444 - val_acc: 0.8655
Epoch 352/1000
52s 105ms/step - loss: 0.5969 - acc: 0.8757 - val_loss: 0.6205 - val_acc: 0.8710
Epoch 353/1000
53s 105ms/step - loss: 0.5978 - acc: 0.8754 - val_loss: 0.6287 - val_acc: 0.8672
Epoch 354/1000
53s 105ms/step - loss: 0.5925 - acc: 0.8778 - val_loss: 0.6314 - val_acc: 0.8664
Epoch 355/1000
53s 105ms/step - loss: 0.5942 - acc: 0.8765 - val_loss: 0.6392 - val_acc: 0.8658
Epoch 356/1000
52s 104ms/step - loss: 0.5961 - acc: 0.8786 - val_loss: 0.6316 - val_acc: 0.8675
Epoch 357/1000
52s 105ms/step - loss: 0.5945 - acc: 0.8766 - val_loss: 0.6536 - val_acc: 0.8619
Epoch 358/1000
53s 105ms/step - loss: 0.5957 - acc: 0.8769 - val_loss: 0.6112 - val_acc: 0.8748
Epoch 359/1000
52s 105ms/step - loss: 0.5992 - acc: 0.8750 - val_loss: 0.6291 - val_acc: 0.8677
Epoch 360/1000
53s 106ms/step - loss: 0.5935 - acc: 0.8778 - val_loss: 0.6283 - val_acc: 0.8691
Epoch 361/1000
53s 106ms/step - loss: 0.5886 - acc: 0.8795 - val_loss: 0.6396 - val_acc: 0.8654
Epoch 362/1000
53s 105ms/step - loss: 0.5900 - acc: 0.8774 - val_loss: 0.6273 - val_acc: 0.8699
Epoch 363/1000
52s 105ms/step - loss: 0.5952 - acc: 0.8769 - val_loss: 0.6017 - val_acc: 0.8798
Epoch 364/1000
52s 105ms/step - loss: 0.5928 - acc: 0.8771 - val_loss: 0.6156 - val_acc: 0.8729
Epoch 365/1000
52s 104ms/step - loss: 0.5997 - acc: 0.8761 - val_loss: 0.6384 - val_acc: 0.8662
Epoch 366/1000
52s 105ms/step - loss: 0.5946 - acc: 0.8771 - val_loss: 0.6245 - val_acc: 0.8714
Epoch 367/1000
53s 105ms/step - loss: 0.5958 - acc: 0.8769 - val_loss: 0.6280 - val_acc: 0.8660
Epoch 368/1000
52s 105ms/step - loss: 0.5917 - acc: 0.8786 - val_loss: 0.6152 - val_acc: 0.8727
Epoch 369/1000
53s 105ms/step - loss: 0.5895 - acc: 0.8784 - val_loss: 0.6376 - val_acc: 0.8654
Epoch 370/1000
53s 105ms/step - loss: 0.5948 - acc: 0.8779 - val_loss: 0.6222 - val_acc: 0.8692
Epoch 371/1000
52s 105ms/step - loss: 0.5895 - acc: 0.8788 - val_loss: 0.6430 - val_acc: 0.8652
Epoch 372/1000
52s 105ms/step - loss: 0.5891 - acc: 0.8801 - val_loss: 0.6184 - val_acc: 0.8750
Epoch 373/1000
52s 105ms/step - loss: 0.5912 - acc: 0.8784 - val_loss: 0.6222 - val_acc: 0.8687
Epoch 374/1000
52s 104ms/step - loss: 0.5899 - acc: 0.8784 - val_loss: 0.6184 - val_acc: 0.8711
Epoch 375/1000
53s 105ms/step - loss: 0.5921 - acc: 0.8778 - val_loss: 0.6091 - val_acc: 0.8736
Epoch 376/1000
53s 105ms/step - loss: 0.5927 - acc: 0.8778 - val_loss: 0.6492 - val_acc: 0.8604
Epoch 377/1000
53s 105ms/step - loss: 0.5969 - acc: 0.8762 - val_loss: 0.6185 - val_acc: 0.8708
Epoch 378/1000
53s 105ms/step - loss: 0.5901 - acc: 0.8778 - val_loss: 0.6314 - val_acc: 0.8681
Epoch 379/1000
52s 104ms/step - loss: 0.5936 - acc: 0.8767 - val_loss: 0.6159 - val_acc: 0.8733
Epoch 380/1000
52s 105ms/step - loss: 0.5941 - acc: 0.8771 - val_loss: 0.6361 - val_acc: 0.8674
Epoch 381/1000
52s 104ms/step - loss: 0.5910 - acc: 0.8778 - val_loss: 0.6542 - val_acc: 0.8600
Epoch 382/1000
52s 105ms/step - loss: 0.5915 - acc: 0.8785 - val_loss: 0.6324 - val_acc: 0.8675
Epoch 383/1000
53s 105ms/step - loss: 0.5905 - acc: 0.8770 - val_loss: 0.6428 - val_acc: 0.8629
Epoch 384/1000
53s 105ms/step - loss: 0.5887 - acc: 0.8786 - val_loss: 0.6285 - val_acc: 0.8663
Epoch 385/1000
52s 105ms/step - loss: 0.5908 - acc: 0.8779 - val_loss: 0.6417 - val_acc: 0.8616
Epoch 386/1000
52s 105ms/step - loss: 0.5887 - acc: 0.8790 - val_loss: 0.6283 - val_acc: 0.8680
Epoch 387/1000
52s 105ms/step - loss: 0.5864 - acc: 0.8783 - val_loss: 0.6315 - val_acc: 0.8660
Epoch 388/1000
52s 105ms/step - loss: 0.5842 - acc: 0.8793 - val_loss: 0.6250 - val_acc: 0.8676
Epoch 389/1000
52s 104ms/step - loss: 0.5876 - acc: 0.8796 - val_loss: 0.6333 - val_acc: 0.8685
Epoch 390/1000
52s 105ms/step - loss: 0.5907 - acc: 0.8784 - val_loss: 0.6327 - val_acc: 0.8655
Epoch 391/1000
53s 105ms/step - loss: 0.5887 - acc: 0.8790 - val_loss: 0.6402 - val_acc: 0.8676
Epoch 392/1000
52s 105ms/step - loss: 0.5937 - acc: 0.8767 - val_loss: 0.6210 - val_acc: 0.8708
Epoch 393/1000
52s 104ms/step - loss: 0.5870 - acc: 0.8801 - val_loss: 0.6186 - val_acc: 0.8750
Epoch 394/1000
52s 104ms/step - loss: 0.5937 - acc: 0.8774 - val_loss: 0.6369 - val_acc: 0.8652
Epoch 395/1000
52s 104ms/step - loss: 0.5891 - acc: 0.8805 - val_loss: 0.6279 - val_acc: 0.8700
Epoch 396/1000
52s 105ms/step - loss: 0.5955 - acc: 0.8776 - val_loss: 0.6179 - val_acc: 0.8702
Epoch 397/1000
52s 104ms/step - loss: 0.5877 - acc: 0.8793 - val_loss: 0.6340 - val_acc: 0.8660
Epoch 398/1000
52s 105ms/step - loss: 0.5899 - acc: 0.8787 - val_loss: 0.5990 - val_acc: 0.8802
Epoch 399/1000
52s 105ms/step - loss: 0.5899 - acc: 0.8802 - val_loss: 0.6270 - val_acc: 0.8694
Epoch 400/1000
53s 105ms/step - loss: 0.5942 - acc: 0.8774 - val_loss: 0.6336 - val_acc: 0.8639
Epoch 401/1000
lr changed to 0.010000000149011612
52s 105ms/step - loss: 0.5041 - acc: 0.9091 - val_loss: 0.5454 - val_acc: 0.8967
Epoch 402/1000
53s 105ms/step - loss: 0.4483 - acc: 0.9265 - val_loss: 0.5314 - val_acc: 0.8978
Epoch 403/1000
53s 105ms/step - loss: 0.4280 - acc: 0.9327 - val_loss: 0.5212 - val_acc: 0.9015
Epoch 404/1000
52s 105ms/step - loss: 0.4145 - acc: 0.9351 - val_loss: 0.5156 - val_acc: 0.9033
Epoch 405/1000
53s 105ms/step - loss: 0.4053 - acc: 0.9367 - val_loss: 0.5152 - val_acc: 0.9042
Epoch 406/1000
53s 105ms/step - loss: 0.3948 - acc: 0.9398 - val_loss: 0.5083 - val_acc: 0.9021
Epoch 407/1000
53s 105ms/step - loss: 0.3880 - acc: 0.9389 - val_loss: 0.5085 - val_acc: 0.9031
Epoch 408/1000
53s 105ms/step - loss: 0.3771 - acc: 0.9433 - val_loss: 0.5094 - val_acc: 0.8987
Epoch 409/1000
52s 105ms/step - loss: 0.3694 - acc: 0.9441 - val_loss: 0.5006 - val_acc: 0.9039
Epoch 410/1000
52s 105ms/step - loss: 0.3669 - acc: 0.9432 - val_loss: 0.4927 - val_acc: 0.9054
Epoch 411/1000
52s 105ms/step - loss: 0.3564 - acc: 0.9466 - val_loss: 0.4973 - val_acc: 0.9034
Epoch 412/1000
52s 104ms/step - loss: 0.3508 - acc: 0.9476 - val_loss: 0.4929 - val_acc: 0.9032
Epoch 413/1000
52s 104ms/step - loss: 0.3464 - acc: 0.9468 - val_loss: 0.4919 - val_acc: 0.9024
Epoch 414/1000
52s 105ms/step - loss: 0.3394 - acc: 0.9487 - val_loss: 0.4842 - val_acc: 0.9032
Epoch 415/1000
52s 105ms/step - loss: 0.3329 - acc: 0.9498 - val_loss: 0.4827 - val_acc: 0.9059
Epoch 416/1000
53s 105ms/step - loss: 0.3317 - acc: 0.9494 - val_loss: 0.4873 - val_acc: 0.9024
Epoch 417/1000
52s 105ms/step - loss: 0.3281 - acc: 0.9485 - val_loss: 0.4812 - val_acc: 0.9074
Epoch 418/1000
53s 105ms/step - loss: 0.3205 - acc: 0.9514 - val_loss: 0.4796 - val_acc: 0.9038
Epoch 419/1000
52s 105ms/step - loss: 0.3207 - acc: 0.9497 - val_loss: 0.4775 - val_acc: 0.9039
Epoch 420/1000
52s 104ms/step - loss: 0.3140 - acc: 0.9518 - val_loss: 0.4753 - val_acc: 0.9052
Epoch 421/1000
53s 105ms/step - loss: 0.3094 - acc: 0.9520 - val_loss: 0.4840 - val_acc: 0.9020
Epoch 422/1000
52s 104ms/step - loss: 0.3091 - acc: 0.9513 - val_loss: 0.4684 - val_acc: 0.9064
Epoch 423/1000
52s 105ms/step - loss: 0.3055 - acc: 0.9515 - val_loss: 0.4629 - val_acc: 0.9065
Epoch 424/1000
52s 104ms/step - loss: 0.2973 - acc: 0.9526 - val_loss: 0.4696 - val_acc: 0.9044
Epoch 425/1000
53s 105ms/step - loss: 0.2965 - acc: 0.9529 - val_loss: 0.4659 - val_acc: 0.9030
Epoch 426/1000
52s 105ms/step - loss: 0.2955 - acc: 0.9531 - val_loss: 0.4584 - val_acc: 0.9067
Epoch 427/1000
52s 105ms/step - loss: 0.2948 - acc: 0.9519 - val_loss: 0.4514 - val_acc: 0.9071
Epoch 428/1000
53s 105ms/step - loss: 0.2871 - acc: 0.9542 - val_loss: 0.4584 - val_acc: 0.9081
Epoch 429/1000
52s 105ms/step - loss: 0.2849 - acc: 0.9541 - val_loss: 0.4684 - val_acc: 0.9037
Epoch 430/1000
53s 105ms/step - loss: 0.2832 - acc: 0.9537 - val_loss: 0.4588 - val_acc: 0.9049
Epoch 431/1000
53s 106ms/step - loss: 0.2785 - acc: 0.9553 - val_loss: 0.4595 - val_acc: 0.9063
Epoch 432/1000
53s 106ms/step - loss: 0.2777 - acc: 0.9546 - val_loss: 0.4516 - val_acc: 0.9059
Epoch 433/1000
53s 106ms/step - loss: 0.2788 - acc: 0.9528 - val_loss: 0.4521 - val_acc: 0.9031
Epoch 434/1000
53s 107ms/step - loss: 0.2743 - acc: 0.9555 - val_loss: 0.4679 - val_acc: 0.9015
Epoch 435/1000
53s 106ms/step - loss: 0.2739 - acc: 0.9540 - val_loss: 0.4512 - val_acc: 0.9053
Epoch 436/1000
53s 105ms/step - loss: 0.2701 - acc: 0.9555 - val_loss: 0.4622 - val_acc: 0.9034
Epoch 437/1000
53s 105ms/step - loss: 0.2697 - acc: 0.9547 - val_loss: 0.4585 - val_acc: 0.9015
Epoch 438/1000
53s 105ms/step - loss: 0.2663 - acc: 0.9552 - val_loss: 0.4556 - val_acc: 0.9027
Epoch 439/1000
53s 106ms/step - loss: 0.2641 - acc: 0.9553 - val_loss: 0.4538 - val_acc: 0.9023
Epoch 440/1000
53s 106ms/step - loss: 0.2649 - acc: 0.9548 - val_loss: 0.4458 - val_acc: 0.9047
Epoch 441/1000
53s 106ms/step - loss: 0.2601 - acc: 0.9561 - val_loss: 0.4499 - val_acc: 0.9032
Epoch 442/1000
53s 106ms/step - loss: 0.2610 - acc: 0.9549 - val_loss: 0.4533 - val_acc: 0.9042
Epoch 443/1000
53s 105ms/step - loss: 0.2608 - acc: 0.9543 - val_loss: 0.4542 - val_acc: 0.9054
Epoch 444/1000
53s 105ms/step - loss: 0.2606 - acc: 0.9547 - val_loss: 0.4585 - val_acc: 0.9003
Epoch 445/1000
53s 105ms/step - loss: 0.2567 - acc: 0.9554 - val_loss: 0.4549 - val_acc: 0.8993
Epoch 446/1000
53s 105ms/step - loss: 0.2551 - acc: 0.9555 - val_loss: 0.4653 - val_acc: 0.8983
Epoch 447/1000
53s 106ms/step - loss: 0.2554 - acc: 0.9558 - val_loss: 0.4561 - val_acc: 0.9000
Epoch 448/1000
53s 106ms/step - loss: 0.2565 - acc: 0.9540 - val_loss: 0.4562 - val_acc: 0.9002
Epoch 449/1000
53s 106ms/step - loss: 0.2528 - acc: 0.9551 - val_loss: 0.4515 - val_acc: 0.8996
Epoch 450/1000
53s 106ms/step - loss: 0.2545 - acc: 0.9545 - val_loss: 0.4475 - val_acc: 0.9015
Epoch 451/1000
53s 106ms/step - loss: 0.2554 - acc: 0.9543 - val_loss: 0.4460 - val_acc: 0.9059
Epoch 452/1000
53s 105ms/step - loss: 0.2506 - acc: 0.9545 - val_loss: 0.4526 - val_acc: 0.8997
Epoch 453/1000
53s 106ms/step - loss: 0.2517 - acc: 0.9542 - val_loss: 0.4442 - val_acc: 0.8999
Epoch 454/1000
53s 105ms/step - loss: 0.2517 - acc: 0.9543 - val_loss: 0.4523 - val_acc: 0.9001
Epoch 455/1000
53s 106ms/step - loss: 0.2458 - acc: 0.9560 - val_loss: 0.4329 - val_acc: 0.9029
Epoch 456/1000
53s 106ms/step - loss: 0.2495 - acc: 0.9546 - val_loss: 0.4407 - val_acc: 0.9026
Epoch 457/1000
53s 105ms/step - loss: 0.2451 - acc: 0.9553 - val_loss: 0.4378 - val_acc: 0.9025
Epoch 458/1000
53s 106ms/step - loss: 0.2472 - acc: 0.9543 - val_loss: 0.4403 - val_acc: 0.9026
Epoch 459/1000
53s 106ms/step - loss: 0.2461 - acc: 0.9550 - val_loss: 0.4359 - val_acc: 0.9041
Epoch 460/1000
53s 105ms/step - loss: 0.2475 - acc: 0.9531 - val_loss: 0.4423 - val_acc: 0.9021
Epoch 461/1000
53s 106ms/step - loss: 0.2450 - acc: 0.9537 - val_loss: 0.4392 - val_acc: 0.9019
Epoch 462/1000
53s 105ms/step - loss: 0.2452 - acc: 0.9543 - val_loss: 0.4408 - val_acc: 0.8996
Epoch 463/1000
53s 106ms/step - loss: 0.2441 - acc: 0.9545 - val_loss: 0.4495 - val_acc: 0.8999
Epoch 464/1000
53s 106ms/step - loss: 0.2439 - acc: 0.9539 - val_loss: 0.4413 - val_acc: 0.9029
Epoch 465/1000
53s 105ms/step - loss: 0.2406 - acc: 0.9555 - val_loss: 0.4503 - val_acc: 0.8977
Epoch 466/1000
53s 106ms/step - loss: 0.2445 - acc: 0.9541 - val_loss: 0.4388 - val_acc: 0.9025
Epoch 467/1000
53s 105ms/step - loss: 0.2402 - acc: 0.9547 - val_loss: 0.4306 - val_acc: 0.9027
Epoch 468/1000
53s 105ms/step - loss: 0.2402 - acc: 0.9565 - val_loss: 0.4391 - val_acc: 0.9040
Epoch 469/1000
53s 105ms/step - loss: 0.2419 - acc: 0.9546 - val_loss: 0.4442 - val_acc: 0.8987
Epoch 470/1000
53s 105ms/step - loss: 0.2409 - acc: 0.9537 - val_loss: 0.4414 - val_acc: 0.9007
Epoch 471/1000
53s 106ms/step - loss: 0.2446 - acc: 0.9527 - val_loss: 0.4478 - val_acc: 0.8971
Epoch 472/1000
53s 106ms/step - loss: 0.2374 - acc: 0.9553 - val_loss: 0.4522 - val_acc: 0.8967
Epoch 473/1000
53s 106ms/step - loss: 0.2418 - acc: 0.9528 - val_loss: 0.4440 - val_acc: 0.8983
Epoch 474/1000
53s 106ms/step - loss: 0.2394 - acc: 0.9552 - val_loss: 0.4418 - val_acc: 0.9000
Epoch 475/1000
52s 105ms/step - loss: 0.2403 - acc: 0.9539 - val_loss: 0.4379 - val_acc: 0.9031
Epoch 476/1000
52s 105ms/step - loss: 0.2379 - acc: 0.9543 - val_loss: 0.4358 - val_acc: 0.8999
Epoch 477/1000
52s 105ms/step - loss: 0.2409 - acc: 0.9529 - val_loss: 0.4433 - val_acc: 0.9006
Epoch 478/1000
53s 106ms/step - loss: 0.2389 - acc: 0.9533 - val_loss: 0.4410 - val_acc: 0.9009
Epoch 479/1000
53s 105ms/step - loss: 0.2385 - acc: 0.9544 - val_loss: 0.4427 - val_acc: 0.9007
Epoch 480/1000
53s 106ms/step - loss: 0.2399 - acc: 0.9538 - val_loss: 0.4248 - val_acc: 0.9018
Epoch 481/1000
53s 106ms/step - loss: 0.2367 - acc: 0.9540 - val_loss: 0.4425 - val_acc: 0.9005
Epoch 482/1000
53s 106ms/step - loss: 0.2376 - acc: 0.9544 - val_loss: 0.4424 - val_acc: 0.9010
Epoch 483/1000
53s 105ms/step - loss: 0.2400 - acc: 0.9537 - val_loss: 0.4414 - val_acc: 0.8987
Epoch 484/1000
52s 105ms/step - loss: 0.2367 - acc: 0.9539 - val_loss: 0.4423 - val_acc: 0.8994
Epoch 485/1000
53s 105ms/step - loss: 0.2356 - acc: 0.9547 - val_loss: 0.4297 - val_acc: 0.9013
Epoch 486/1000
53s 105ms/step - loss: 0.2378 - acc: 0.9543 - val_loss: 0.4286 - val_acc: 0.9039
Epoch 487/1000
53s 107ms/step - loss: 0.2347 - acc: 0.9551 - val_loss: 0.4304 - val_acc: 0.9018
Epoch 488/1000
 87/500 [====>.........................] - ETA: 42s - loss: 0.2317 - acc: 0.9568Traceback (most recent call last):

KeyboardInterrupt

无意中按了Ctrl+C，把程序给中断了，没有运行完。本来设置跑1000个epoch，只跑到第488个。验证集上的准确率已经达到了90%。

Minghang Zhao, Shisheng Zhong, Xuyun Fu, Baoping Tang, Shaojiang Dong, Michael Pecht, Deep Residual Networks with Adaptively Parametric Rectifier Linear Units for Fault Diagnosis, IEEE Transactions on Industrial Electronics, 2020, DOI: 10.1109/TIE.2020.2972458, Date of Publication: 13 February 2020

https://ieeexplore.ieee.org/d...
————————————————
版权声明：本文为CSDN博主「dangqing1988」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/dangqin...