Keras 系列（六） CNN 分类及fit_generator函数

 验证集生成器一个形如（x_var,y_val）的元组
一个形如(x_val, y_val，sample_weights)的元组

7 workers: 最大进程数。在使用多线程时，启动进程最大数量（process-based threading）。未特别指定时，默认为1。如果指定为0，则执行主线程.8 use_multiprocessing: 布尔值。True:使用基于过程的线程

#数据下载地址：http://download.tensorflow.org/example_images/flower_photos.tgz
#加载相关模块
from skimage import io,transform
from pandas import Series, DataFrame 
import glob
import os
import numpy as np
from keras.models import Sequential
from keras.layers.core import Flatten,Dense,Dropout
from keras.layers.convolutional import Convolution2D,MaxPooling2D,ZeroPadding2D
from keras.optimizers import SGD,Adadelta,Adagrad
from keras.utils import np_utils,generic_utils
from keras.layers.advanced_activations import PReLU
from keras.layers.core import Flatten, Dense, Dropout 
from keras.layers.core import Dense, Dropout, Activation, Flatten 
from six.moves import range

path='/home/CNN_Classification/flower_photos' #自己修改路径
w=182
h=182
c=3

#读取图片
def read_img(path):
    cate=[path+'/'+x for x in os.listdir(path) if os.path.isdir(path+'/'+x)]
    print(cate)
    imgs=[]
    labels=[]
    n=
    for idx,folder in enumerate(cate):
        for im in glob.glob(folder+'/*.jpg'):
            print('reading the images:%s'%(im))
            img=io.imread(im)
            print('before resize:',img.shape)  # 
            img=transform.resize(img,(w,h))
            print('after:',img.shape)
            imgs.append(img)
            labels.append(idx)
            n=n+
           
    return np.asarray(imgs,np.float32),np.asarray(labels,np.int32)
#调用函数
data,label=read_img(path)
print('叠加之后的形状：',data.shape) #

reading the images:/home/CNN_Classification/flower_photos/daisy/517054467_d82d323c33_m.jpg
before resize: (240, 172, 3)
after: (182, 182, 3)
reading the images:/home/CNN_Classification/flower_photos/daisy/7454630692_ab2d67dd18_m.jpg
before resize: (240, 159, 3)
after: (182, 182, 3)
reading the images:/home/CNN_Classification/flower_photos/daisy/5684911529_88a7ae32ba_n.jpg
before resize: (213, 320, 3)
after: (182, 182, 3)
reading the images:/home/CNN_Classification/flower_photos/daisy/4268817944_cdbdb226ae.jpg

#打乱顺序,将标签转为二进制独热形式（0和1组成）
num_example=data.shape[0]
arr=np.arange(num_example)
np.random.shuffle(arr)
data=data[arr]
label=label[arr]
from keras.utils.np_utils import to_categorical
labels_5= to_categorical(label,num_classes=5)
print(label_5)

array([[0., 1., 0., 0., 0.],
       [0., 1., 0., 0., 0.],
       [1., 0., 0., 0., 0.],
       ...,
       [0., 0., 0., 0., 1.],
       [1., 0., 0., 0., 0.],
       [0., 0., 1., 0., 0.]], dtype=float32)

def minibatches(inputs=None, targets=None, batch_size=None, shuffle=False):
    while 1:  # 要无限循环
        assert len(inputs) == len(targets)
        if shuffle:
            indices = np.arange(len(inputs))
            np.random.shuffle(indices)
        for start_idx in range(, len(inputs) - batch_size + , batch_size):
            if shuffle:
                excerpt = indices[start_idx:start_idx + batch_size]
            else:
                excerpt = slice(start_idx, start_idx + batch_size)
            yield inputs[excerpt], targets[excerpt]

for x,y  in minibatches(inputs=data, targets=labels_5, batch_size=6, shuffle=False):
    
    print('y的一个批次:',y)
    print('x批次:',x)

#注意函数内部是一个while 1死循环，单独测试该函数时要手动停止
y的一个批次：
 [0. 0. 0. 0. 1.]
 [1. 0. 0. 0. 0.]
 [0. 1. 0. 0. 0.]
 [0. 0. 0. 1. 0.]
 [0. 0. 0. 1. 0.]]
x批次: [[[[8.31834912e-01 3.92123640e-01 5.50473154e-01]
   [8.52876544e-01 4.23938811e-01 5.76879978e-01]
   [8.67920816e-01 4.76173371e-01 6.31592453e-01]
   ...
   [9.12868023e-01 5.83456278e-01 7.16789603e-01]
   [8.96439314e-01 5.55262804e-01 6.91900969e-01]
   [8.46583009e-01 5.25663435e-01 6.52079284e-01]]

  [[8.17657828e-01 3.77946556e-01 5.36296070e-01]
   [8.52876544e-01 4.23938811e-01 5.76879978e-01]
   [8.67076039e-01 4.75328594e-01 6.30747676e-01]
   ...
   [9.27464306e-01 5.98052561e-01 7.30427027e-01]
   [9.08384621e-01 5.67208111e-01 7.02976286e-01]
   [8.69190335e-01 5.41265309e-01 6.70181513e-01]]

  [[8.14695120e-01 3.74983847e-01 5.33333361e-01]
   [8.50441694e-01 4.21503991e-01 5.74445188e-01]
   [8.67076039e-01 4.75328594e-01 6.30747676e-01]
   ...
   [9.42060530e-01 6.08727217e-01 7.38138974e-01]
   [9.19397473e-01 5.77890158e-01 7.11333752e-01]
   [8.89581978e-01 5.56199253e-01 6.85865104e-01]]

  ...
第五步  建模、预测

model = Sequential() #第一个卷积层，4个卷积核，每个卷积核大小5*5。 
#激活函数用tanh #你还可以在model.add(Activation('tanh'))后加上dropout的技巧: model.add(Dropout(0.5)) 
model.add(Convolution2D(4, 5, 5,input_shape=(w, h,3))) 
model.add(Activation('relu')) 
model.add(MaxPooling2D(pool_size=(2, 2))) #第二个卷积层，8个卷积核，每个卷积核大小3*3。
#激活函数用tanh #采用maxpooling，poolsize为(2,2) 
model.add(Convolution2D(8, 3, 3)) 
model.add(Activation('relu')) 
model.add(MaxPooling2D(pool_size=(2, 2))) 
#第三个卷积层，16个卷积核，每个卷积核大小3*3 #激活函数用tanh 
#采用maxpooling，poolsize为(2,2) 
model.add(Convolution2D(16, 3, 3)) 
model.add(Activation('relu')) 
model.add(MaxPooling2D(pool_size=(2, 2)))
#全连接层，先将前一层输出的二维特征图flatten为一维的。
model.add(Flatten())
model.add(Dense(128))
model.add(Activation('relu'))
model.add(Dropout(0.5))
#多分类  
model.add(Dense(5)) # 共有5个类别
model.add(Activation('softmax'))
#print(model.summary())
model.compile(loss='categorical_crossentropy',optimizer='adam')
#model.fit(data,labels_5,epochs=6,batch_size=2,verbose=2)#旧方法不再适用
history=model.fit_generator(minibatches(data,labels_5,batch_size=6,shuffle=False),
                            steps_per_epoch=len(data)//6,
                            epochs=6)
#model.train_on_batch(minibatches(data, labels_5, batch_size=6, shuffle=False))
pyplot.plot(history.history['loss'],label='train')
#这里只是简单地预测训练集中的目标值，想要预测验证集换一下数据即可。
print('预测结果为：',model.predict_classes(data[-3:]))

Epoch 1/6
611/611 [==============================] - 70s 115ms/step - loss: 1.2815
Epoch 2/6
611/611 [==============================] - 69s 113ms/step - loss: 1.0301
Epoch 3/6
611/611 [==============================] - 70s 114ms/step - loss: 0.8700
Epoch 4/6
611/611 [==============================] - 70s 114ms/step - loss: 0.7510