Keras实现卷积神经网络

来源:互联网 发布:mysql synonym 编辑:程序博客网 时间:2024/05/16 12:40

在安装过Tensorflow后,在在安装Keras默认将TF作为后端,Keras实现卷积网络的代码十分简洁,而且keras中的callback类提供对模型训练过程中变量的检测方法,能够根据检测变量的情况及时的调整模型的学习效率和一些参数.
下面的例子,MNIST数据作为测试

import pandas as pdimport numpy as npimport matplotlib.pyplot as pltimport matplotlib.image as pimgimport seaborn as sb         # 一个构建在matplotlib上的绘画模块,支持numpy,pandas等数据结构%matplotlib inlinefrom sklearn.model_selection import train_test_splitfrom sklearn.metrics import confusion_matrix     # 混淆矩阵import itertools#  kerasfrom keras.utils import to_categorical         #数字标签转化成one-hot编码from keras.models import Sequentialfrom keras.layers import Dense,Dropout,Flatten,Conv2D,MaxPool2Dfrom keras.optimizers import RMSpropfrom keras.preprocessing.image import ImageDataGeneratorfrom keras.callbacks import ReduceLROnPlateau
Using TensorFlow backend.
# 设置绘画风格sb.set(style='white', context='notebook', palette='deep')
# 加载数据train_data = pd.read_csv('data/train.csv')test_data = pd.read_csv('data/test.csv')#train_x = train_data.drop(labels=['label'],axis=1)  # 去掉标签列train_x = train_data.iloc[:,1:]train_y = train_data.iloc[:,0]del  train_data   # 释放一下内存
# 观察一下训练数据的分布情况g = sb.countplot(train_y)train_y.value_counts()
1    46847    44013    43519    41882    41776    41370    41324    40728    40635    3795Name: label, dtype: int64

这里写图片描述

train_x.isnull().describe() # 检查是否存在确实值
train_x.isnull().any().describe()
count       784unique        1top       Falsefreq        784dtype: object
test_data.isnull().any().describe()
count       784unique        1top       Falsefreq        784dtype: object
# 归一化train_x =  train_x/255.0test_x = test_data/255.0
del test_data

转换数据的shape

# reshape trian_x, test_x#train_x = train_x.values.reshape(-1, 28, 28, 1)#test_x = test_x.values.reshape(-1, 28, 28, 1)train_x = train_x.as_matrix().reshape(-1, 28, 28, 1)test_x = test_x.as_matrix().reshape(-1, 28, 28, 1)
# 吧标签列转化为one-hot 编码格式train_y = to_categorical(train_y, num_classes = 10)

从数据中分离出验证数据

#从训练数据中分出十分之一的数据作为验证数据random_seed = 3train_x , val_x , train_y, val_y = train_test_split(train_x, train_y, test_size=0.1, random_state=random_seed)

一个训练样本

plt.imshow(train_x[0][:,:,0])

这里写图片描述

使用Keras搭建CNN

model = Sequential()# 第一个卷积层,32个卷积核,大小5x5,卷积模式SAME,激活函数relu,输入张量的大小model.add(Conv2D(filters= 32, kernel_size=(5,5), padding='Same', activation='relu',input_shape=(28,28,1)))model.add(Conv2D(filters= 32, kernel_size=(5,5), padding='Same', activation='relu'))# 池化层,池化核大小2x2model.add(MaxPool2D(pool_size=(2,2)))# 随机丢弃四分之一的网络连接,防止过拟合model.add(Dropout(0.25))  model.add(Conv2D(filters= 64, kernel_size=(3,3), padding='Same', activation='relu'))model.add(Conv2D(filters= 64, kernel_size=(3,3), padding='Same', activation='relu'))model.add(MaxPool2D(pool_size=(2,2), strides=(2,2)))model.add(Dropout(0.25))# 全连接层,展开操作,model.add(Flatten())# 添加隐藏层神经元的数量和激活函数model.add(Dense(256, activation='relu'))    model.add(Dropout(0.25))# 输出层model.add(Dense(10, activation='softmax'))  
# 设置优化器# lr :学习效率, decay :lr的衰减值optimizer = RMSprop(lr = 0.001, decay=0.0)
# 编译模型# loss:损失函数,metrics:对应性能评估函数model.compile(optimizer=optimizer, loss = 'categorical_crossentropy',metrics=['accuracy'])

创建一个callback类的实例

# keras的callback类提供了可以跟踪目标值,和动态调整学习效率# moitor : 要监测的量,这里是验证准确率# matience: 当经过3轮的迭代,监测的目标量,仍没有变化,就会调整学习效率# verbose : 信息展示模式,去0或1# factor : 每次减少学习率的因子,学习率将以lr = lr*factor的形式被减少# mode:‘auto’,‘min’,‘max’之一,在min模式下,如果检测值触发学习率减少。在max模式下,当检测值不再上升则触发学习率减少。# epsilon:阈值,用来确定是否进入检测值的“平原区”# cooldown:学习率减少后,会经过cooldown个epoch才重新进行正常操作# min_lr:学习率的下限learning_rate_reduction = ReduceLROnPlateau(monitor = 'val_acc', patience = 3,                                            verbose = 1, factor=0.5, min_lr = 0.00001)
epochs = 40batch_size = 100

数据增强处理

# 数据增强处理,提升模型的泛化能力,也可以有效的避免模型的过拟合# rotation_range : 旋转的角度# zoom_range : 随机缩放图像# width_shift_range : 水平移动占图像宽度的比例# height_shift_range # horizontal_filp : 水平反转# vertical_filp : 纵轴方向上反转data_augment = ImageDataGenerator(rotation_range= 10,zoom_range= 0.1,                                  width_shift_range = 0.1,height_shift_range = 0.1,                                  horizontal_flip = False, vertical_flip = False)

训练模型

history = model.fit_generator(data_augment.flow(train_x, train_y, batch_size=batch_size),                             epochs= epochs, validation_data = (val_x,val_y),                             verbose =2, steps_per_epoch=train_x.shape[0]//batch_size,                             callbacks=[learning_rate_reduction])
Epoch 1/40359s - loss: 0.4529 - acc: 0.8498 - val_loss: 0.0658 - val_acc: 0.9793Epoch 2/40375s - loss: 0.1188 - acc: 0.9637 - val_loss: 0.0456 - val_acc: 0.9848Epoch 3/40374s - loss: 0.0880 - acc: 0.9734 - val_loss: 0.0502 - val_acc: 0.9845Epoch 4/40375s - loss: 0.0750 - acc: 0.9767 - val_loss: 0.0318 - val_acc: 0.9902Epoch 5/40374s - loss: 0.0680 - acc: 0.9800 - val_loss: 0.0379 - val_acc: 0.9888Epoch 6/40369s - loss: 0.0584 - acc: 0.9823 - val_loss: 0.0267 - val_acc: 0.9910Epoch 7/40381s - loss: 0.0556 - acc: 0.9832 - val_loss: 0.0505 - val_acc: 0.9824Epoch 8/40381s - loss: 0.0531 - acc: 0.9842 - val_loss: 0.0236 - val_acc: 0.9912Epoch 9/40376s - loss: 0.0534 - acc: 0.9839 - val_loss: 0.0310 - val_acc: 0.9910Epoch 10/40379s - loss: 0.0537 - acc: 0.9848 - val_loss: 0.0274 - val_acc: 0.9917Epoch 11/40375s - loss: 0.0501 - acc: 0.9856 - val_loss: 0.0254 - val_acc: 0.9931Epoch 12/40382s - loss: 0.0492 - acc: 0.9860 - val_loss: 0.0212 - val_acc: 0.9924Epoch 13/40380s - loss: 0.0482 - acc: 0.9864 - val_loss: 0.0259 - val_acc: 0.9919Epoch 14/40373s - loss: 0.0488 - acc: 0.9858 - val_loss: 0.0305 - val_acc: 0.9905Epoch 15/40Epoch 00014: reducing learning rate to 0.000500000023749.370s - loss: 0.0493 - acc: 0.9853 - val_loss: 0.0259 - val_acc: 0.9919Epoch 16/40367s - loss: 0.0382 - acc: 0.9888 - val_loss: 0.0176 - val_acc: 0.9936Epoch 17/40376s - loss: 0.0376 - acc: 0.9891 - val_loss: 0.0187 - val_acc: 0.9945Epoch 18/40376s - loss: 0.0410 - acc: 0.9885 - val_loss: 0.0220 - val_acc: 0.9926Epoch 19/40371s - loss: 0.0385 - acc: 0.9886 - val_loss: 0.0194 - val_acc: 0.9933Epoch 20/40372s - loss: 0.0345 - acc: 0.9894 - val_loss: 0.0186 - val_acc: 0.9938Epoch 21/40Epoch 00020: reducing learning rate to 0.000250000011874.375s - loss: 0.0395 - acc: 0.9888 - val_loss: 0.0233 - val_acc: 0.9945Epoch 22/40369s - loss: 0.0313 - acc: 0.9907 - val_loss: 0.0141 - val_acc: 0.9955Epoch 23/40376s - loss: 0.0308 - acc: 0.9910 - val_loss: 0.0187 - val_acc: 0.9945Epoch 24/40374s - loss: 0.0331 - acc: 0.9908 - val_loss: 0.0170 - val_acc: 0.9940Epoch 25/40372s - loss: 0.0325 - acc: 0.9904 - val_loss: 0.0166 - val_acc: 0.9948Epoch 26/40Epoch 00025: reducing learning rate to 0.000125000005937.373s - loss: 0.0319 - acc: 0.9904 - val_loss: 0.0167 - val_acc: 0.9943Epoch 27/40372s - loss: 0.0285 - acc: 0.9915 - val_loss: 0.0138 - val_acc: 0.9950Epoch 28/40375s - loss: 0.0280 - acc: 0.9913 - val_loss: 0.0150 - val_acc: 0.9950Epoch 29/40Epoch 00028: reducing learning rate to 6.25000029686e-05.377s - loss: 0.0281 - acc: 0.9924 - val_loss: 0.0158 - val_acc: 0.9948Epoch 30/40374s - loss: 0.0265 - acc: 0.9920 - val_loss: 0.0134 - val_acc: 0.9952Epoch 31/40378s - loss: 0.0270 - acc: 0.9922 - val_loss: 0.0128 - val_acc: 0.9957Epoch 32/40372s - loss: 0.0237 - acc: 0.9930 - val_loss: 0.0133 - val_acc: 0.9957Epoch 33/40375s - loss: 0.0237 - acc: 0.9931 - val_loss: 0.0138 - val_acc: 0.9955Epoch 34/40371s - loss: 0.0276 - acc: 0.9920 - val_loss: 0.0135 - val_acc: 0.9962Epoch 35/40373s - loss: 0.0259 - acc: 0.9920 - val_loss: 0.0136 - val_acc: 0.9952Epoch 36/40369s - loss: 0.0249 - acc: 0.9924 - val_loss: 0.0126 - val_acc: 0.9952Epoch 37/40370s - loss: 0.0257 - acc: 0.9923 - val_loss: 0.0130 - val_acc: 0.9960Epoch 38/40Epoch 00037: reducing learning rate to 3.12500014843e-05.374s - loss: 0.0252 - acc: 0.9926 - val_loss: 0.0136 - val_acc: 0.9950Epoch 39/40372s - loss: 0.0246 - acc: 0.9927 - val_loss: 0.0134 - val_acc: 0.9957Epoch 40/40371s - loss: 0.0247 - acc: 0.9929 - val_loss: 0.0139 - val_acc: 0.9950

在训练过程当中,有几次触发学习效率衰减的条件,每当val_acc连续3轮没有增长,就会把学习效率调整为当前的一半,调整之后,val_acc都有明显的增长,但是在最后几轮,模型可能已经收敛.

# learning curvesfig,ax = plt.subplots(2,1,figsize=(10,10))ax[0].plot(history.history['loss'], color='r', label='Training Loss')ax[0].plot(history.history['val_loss'], color='g', label='Validation Loss')ax[0].legend(loc='best',shadow=True)ax[0].grid(True)ax[1].plot(history.history['acc'], color='r', label='Training Accuracy')ax[1].plot(history.history['val_acc'], color='g', label='Validation Accuracy')ax[1].legend(loc='best',shadow=True)ax[1].grid(True)

这里写图片描述

# 混淆矩阵def plot_sonfusion_matrix(cm, classes, normalize=False, title='Confusion matrix',cmap=plt.cm.Blues):    plt.imshow(cm, interpolation='nearest', cmap=cmap)    plt.title(title)    plt.colorbar()    tick_marks = np.arange(len(classes))    plt.xticks(tick_marks, classes, rotation=45)    plt.yticks(tick_marks, classes)    if normalize:        cm = cm.astype('float')/cm.sum(axis=1)[:,np.newaxis]    thresh = cm.max()/2.0    for i,j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):        plt.text(j,i,cm[i,j], horizontalalignment='center',color='white' if cm[i,j] > thresh else 'black')    plt.tight_layout()    plt.ylabel('True label')    plt.xlabel('Predict label')

验证数据的混淆举证

pred_y = model.predict(val_x)pred_label = np.argmax(pred_y, axis=1)true_label = np.argmax(val_y, axis=1)confusion_mat = confusion_matrix(true_label, pred_label)plot_sonfusion_matrix(confusion_mat, classes = range(10))

这里写图片描述

原创粉丝点击