Tensorflow2.0学习笔记--Day2

Tensorflow2.0学习笔记–Day2

这次主要是入门了一波基本的全连接神经网络，以及卷积神经网络模型的构建
接着上次的来看
输出单张图片函数：

def show_single_image(img_arr):
    plt.imshow(img_arr, cmap="binary")#cmap颜色图谱默认RGB，黑白用的binary
    plt.show()

show_single_image(x_train[0])

输出多张图的函数：

def show_imgs(n_rows,n_cols,x_data,y_data,class_names):
    #输出多张图，图摆成n行n列，x代表图片，y代表类别名，cn类别名本身
    assert len(x_data)==len(y_data)#图片和名字数要相同
    assert n_rows * n_cols<len(x_data)
    plt.figure(figsize=(n_cols*1.4,n_rows*1.6))
    for row in range(n_rows):
        for col in range(n_cols):
            index=n_cols*row+col#当前位置放置图片的索引，从0开始
            plt.subplot(n_rows,n_cols,index+1)
            plt.imshow(x_data[index],cmap="binary",interpolation='nearest')#用最近的像素点
            plt.axis('off')#关闭坐标轴
            plt.title(class_names[y_data[index]])
    plt.show()

class_names = ['T-shirt', 'Trouser', 'Pullover', 'Dress',
               'Coat', 'Sandal', 'Shirt', 'Sneaker',
               'Bag', 'Ankle boot']#所有类别的名字
show_imgs(3, 5, x_train, y_train, class_names)#输出10*10=100张图片以及他们对应的类别

构建网络模型（全连接网络模型）：

#构建神经网络模型
import tensorflow as tf
model=tf.keras.models.Sequential()#也可以用列表的形式把层写进去
model.add(tf.keras.layers.Flatten(input_shape=[28,28]))#将28*28的图像展平成一维向量
model.add(tf.keras.layers.Dense(300,activation="relu"))#全连接层
#model.add(tf.keras.layers.Dense(300,activation="relu"))
#model.add(tf.keras.layers.Dense(300,activation="relu"))
model.add(tf.keras.layers.Dense(100,activation="relu"))
model.add(tf.keras.layers.Dense(10,activation="softmax"))
#relu输出是max（0，x）
#softmax将向量变成概率分布x=[x1,x2,x3],e^x/sum
#当y是一个数的时候用带sparse的，y是one hot->[]，不用带
model.compile(loss="sparse_categorical_crossentropy",optimizer=tf.keras.optimizers.RMSprop(learning_rate=0.0004),metrics=["accuracy"])
#optimizer="sgd",metrics=["accuracy"])#训练结果与优化器和学习率都有关系

优化器两种都用到了，效果都不是太好，还需要进一步的调参

model.layers#看模型中的层

[<tensorflow.python.keras.layers.core.Flatten at 0x21f9163cf88>,
 <tensorflow.python.keras.layers.core.Dense at 0x21f92a91f48>,
 <tensorflow.python.keras.layers.core.Dense at 0x21f92c28408>,
 <tensorflow.python.keras.layers.core.Dense at 0x21f92c36f88>]

model.summary()#看网络内部结构

Model: "sequential_14"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
flatten_14 (Flatten)         (None, 784)               0         
_________________________________________________________________
dense_57 (Dense)             (None, 300)               235500    
_________________________________________________________________
dense_58 (Dense)             (None, 100)               30100     
_________________________________________________________________
dense_59 (Dense)             (None, 10)                1010      
=================================================================
Total params: 266,610
Trainable params: 266,610
Non-trainable params: 0
_________________________________________________________________

原始数据的归一化，x=（x-平均值）/方差，来提高训练效果

#归一化：x=(x-u)/std
from sklearn.preprocessing import StandardScaler
scaler=StandardScaler()#初始化对象
#x_train:[None,28,28]->[None,784]
x_train_scaled=scaler.fit_transform(x_train.astype(np.float32).reshape(-1,1)).reshape(-1,28,28)#要做除法就先化成了float，transform只能计算二维数据，所以先化成二维再化回三维
#fit功能是记录训练集的均值和方差
x_valid_scaled=scaler.transform(x_valid.astype(np.float32).reshape(-1,1)).reshape(-1,28,28)
x_test_scaled=scaler.transform(x_test.astype(np.float32).reshape(-1,1)).reshape(-1,28,28)

print(np.max(x_train),np.min(x_train))
print(np.max(x_train_scaled),np.min(x_train_scaled))

255 0
2.0218322 -0.8112618#归一化之后最大值和最小值相差较小

开始训练：

logdir = os.path.join('cnn-selu-callbacks')#将模型文件保存在该目录，win系统中要加上os.path.join(),否则会报错找不到路径
if not os.path.exists(logdir):
    os.mkdir(logdir)
output_model_file = os.path.join(logdir,"fashion_mnist_model.h5")
callbacks = [
    keras.callbacks.TensorBoard(logdir),
    keras.callbacks.ModelCheckpoint(output_model_file,save_best_only = True),
    keras.callbacks.EarlyStopping(patience=5, min_delta=1e-3),
]
history = model.fit(x_train_scaled, y_train, epochs=10,
                    validation_data=(x_valid_scaled, y_valid),
                    callbacks = callbacks)#开始训练，训练10轮，每个一段时间就会对验证集进行验证,model是可以返回值的，history是它的运行结果,callback类型的

这波训练效果挺差的，参数还没调整好，先放卷积神经网络的那波训练吧，后面的画图也以那个为例

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(filters=32, kernel_size=3,
                              padding='same',
                              activation='relu',
                              input_shape=(28, 28, 1)))
model.add(tf.keras.layers.Conv2D(filters=32, kernel_size=3,
                              padding='same',
                              activation='relu'))
model.add(tf.keras.layers.MaxPool2D(pool_size=2))
model.add(tf.keras.layers.Conv2D(filters=64, kernel_size=3,
                              padding='same',
                              activation='relu'))
model.add(tf.keras.layers.Conv2D(filters=64, kernel_size=3,
                              padding='same',
                              activation='relu'))
model.add(tf.keras.layers.MaxPool2D(pool_size=2))
model.add(tf.keras.layers.Conv2D(filters=128, kernel_size=3,
                              padding='same',
                              activation='relu'))
model.add(tf.keras.layers.Conv2D(filters=128, kernel_size=3,
                              padding='same',
                              activation='relu'))
model.add(tf.keras.layers.MaxPool2D(pool_size=2))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(128, activation='relu'))
model.add(tf.keras.layers.Dense(10, activation="softmax"))

model.compile(loss="sparse_categorical_crossentropy",
              optimizer = "sgd",
              metrics = ["accuracy"])

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 28, 28, 32)        320       
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 28, 28, 32)        9248      
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 14, 14, 32)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 14, 14, 64)        18496     
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 14, 14, 64)        36928     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 7, 7, 64)          0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 7, 7, 128)         73856     
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 7, 7, 128)         147584    
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 3, 3, 128)         0         
_________________________________________________________________
flatten (Flatten)            (None, 1152)              0         
_________________________________________________________________
dense (Dense)                (None, 128)               147584    
_________________________________________________________________
dense_1 (Dense)              (None, 10)                1290      
=================================================================
Total params: 435,306
Trainable params: 435,306
Non-trainable params: 0
_________________________________________________________________

然后开始训练：

Train on 55000 samples, validate on 5000 samples
Epoch 1/10
55000/55000 [==============================] - 16s 294us/sample - loss: 0.7778 - accuracy: 0.7171 - val_loss: 0.4607 - val_accuracy: 0.8262
Epoch 2/10
55000/55000 [==============================] - 12s 211us/sample - loss: 0.4154 - accuracy: 0.8484 - val_loss: 0.3743 - val_accuracy: 0.8628
Epoch 3/10
55000/55000 [==============================] - 12s 219us/sample - loss: 0.3506 - accuracy: 0.8709 - val_loss: 0.3373 - val_accuracy: 0.8734
Epoch 4/10
55000/55000 [==============================] - 12s 214us/sample - loss: 0.3174 - accuracy: 0.8840 - val_loss: 0.3139 - val_accuracy: 0.8828
Epoch 5/10
55000/55000 [==============================] - 12s 220us/sample - loss: 0.2928 - accuracy: 0.8939 - val_loss: 0.2972 - val_accuracy: 0.8930
Epoch 6/10
55000/55000 [==============================] - 12s 219us/sample - loss: 0.2725 - accuracy: 0.9002 - val_loss: 0.2774 - val_accuracy: 0.8982
Epoch 7/10
55000/55000 [==============================] - 12s 212us/sample - loss: 0.2548 - accuracy: 0.9056 - val_loss: 0.2724 - val_accuracy: 0.8978
Epoch 8/10
55000/55000 [==============================] - 12s 219us/sample - loss: 0.2405 - accuracy: 0.9122 - val_loss: 0.2441 - val_accuracy: 0.9100
Epoch 9/10
55000/55000 [==============================] - 12s 219us/sample - loss: 0.2261 - accuracy: 0.9163 - val_loss: 0.2442 - val_accuracy: 0.9124
Epoch 10/10
55000/55000 [==============================] - 12s 221us/sample - loss: 0.2130 - accuracy: 0.9219 - val_loss: 0.2409 - val_accuracy: 0.9136

history.history#可以查看训练过程中的各个变量值的变化

在这里插入代码片
```{'loss': [0.777842640356584,
  0.41535070024837145,
  0.35062143627513537,
  0.3174463068506934,
  0.29282925888841804,
  0.2725241752191023,
  0.2547639598553831,
  0.24052352876554836,
  0.22614784826365383,
  0.21300285703594035],
 'accuracy': [0.71712726,
  0.8484182,
  0.87087274,
  0.88403636,
  0.89385456,
  0.9002,
  0.90563637,
  0.9121636,
  0.91629094,
  0.92185456],
 'val_loss': [0.4606949928283691,
  0.3742639167308807,
  0.3373405999660492,
  0.31385451159477235,
  0.29724889080524447,
  0.27741495609283445,
  0.2723794868707657,
  0.24407010363340378,
  0.24418159279823304,
  0.2408748551428318],
 'val_accuracy': [0.8262,
  0.8628,
  0.8734,
  0.8828,
  0.893,
  0.8982,
  0.8978,
  0.91,
  0.9124,
  0.9136]}

定义一个画图的函数，可以画出训练过程中的各个变量值的变化

def plot_learning_curve(history):
    #传的是Tensorflow的callback类型
    pd.DataFrame(history.history).plot(figsize=(8,5))#把所有的Loss，accuracy导入DataFrame，然后画图
    plt.grid(True)#显示为网格
    plt.gca().set_ylim(0,1)#设置坐标轴（y轴）的范围
    plt.show()
plot_learning_curves(history)

model.evaluate(x_test_scaled, y_test)#在测试集上进行测试

最终的效果还不错，本次学习还是有很多问题值得去完善，现在还有一个坑尚未解决，就是训练是有时候会出现如下报错情况：

接下来继续加油！