mnist 自定義CNN模型與lenet

import numpy as np
import matplotlib.pyplot as plt
from tensorflow import keras
from keras.layers import Input, Dense, Dropout, Activation,Conv2D,MaxPool2D,Flatten
from keras.datasets import mnist
from keras.models import Model
from tensorflow.python.keras.utils.np_utils import to_categorical
from keras.callbacks import TensorBoard
 

%load_ext tensorboard
%tensorboard --logdir './log/train'

# 載入一次後，如果要重新載入，就需要使用reload方法
%reload_ext tensorboard
%tensorboard --logdir './log/train'

(x_train,y_train),(x_test,y_test) = mnist.load_data()

# 資料視覺化
def image(data,row,col):
  images=data[:col*row]
  fig,axs=plt.subplots(row,col)
  idx = 0
  for i in range(row):
      for j in range(col):
          axs[i, j].imshow(images[idx, :, :], cmap='gray')
          axs[i, j].axis('off')
          idx += 1
  plt.show()
 

image(x_train,5,5)

temp=np.unique(y_train)
y_train_numbers = []
for v in temp:
  y_train_numbers.append(np.sum(y_train==v))
plt.title("The number of each digit in the train data")
plt.bar(temp,y_train_numbers) 
plt.xlabel('digits', color='b')
plt.ylabel('numbers of digits', color='b')
for x, y in zip(range(len(y_train_numbers)), y_train_numbers):
    plt.text(x + 0.05, y + 0.05, '%.2d' % y, ha='center', va='bottom')
plt.show()

 

# 資料標準化
def normalize(x):
    x = x.astype(np.float32) / 255.0
    return x
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1).astype('float32')
x_test = x_test.reshape(x_test.shape[0], 28, 28, 1).astype('float32')
x_train = normalize(x_train)
x_test = normalize(x_test)

# 建立模型

(60000, 28, 28)

import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dense, Dropout, Flatten
%load_ext tensorboard  
#使用tensorboard 擴充套件
%tensorboard --logdir logs 
#定位tensorboard讀取的檔案目錄
# logs是存放tensorboard檔案的目錄

def model_1():
  # 建立模型，輸入28*28*1個神經元，輸出10個神經元
  model = Sequential()
  model.add(Conv2D(filters=8, kernel_size=(5, 5), padding='same',
                  input_shape=(28, 28, 1), activation='relu'))
  # 第一個池化層
  model.add(MaxPooling2D(pool_size=(2, 2)))  # 出來的特徵圖為14*14大小
  # 第二個卷積層 16個濾波器（卷積核），卷積視窗大小為5*5
  # 經過第二個卷積層後，有16個feature map，每個feature map為14*14
  model.add(Conv2D(filters=16, kernel_size=(5,5), padding='same',
                  activation='relu'))
  model.add(MaxPooling2D(pool_size=(2, 2)))  # 出來的特徵圖為7*7大小

  # 平坦層，把第二個池化層的輸出扁平化為1維 16*7*7
  model.add(Flatten())
  # 第一個全連線層
  model.add(Dense(100, activation='relu'))
  # Dropout，用來放棄一些權值，防止過擬合
  model.add(Dropout(0.25))
  # 第二個全連線層，由於是輸出層，所以使用softmax做啟用函式
  model.add(Dense(10, activation='softmax'))
  # 列印模型
  return model

model_demo1=model_1()
model_demo1.summary()

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_2 (Conv2D)            (None, 28, 28, 8)         208       
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 14, 14, 8)         0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 14, 14, 16)        3216      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 7, 7, 16)          0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 784)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 100)               78500     
_________________________________________________________________
dropout_1 (Dropout)          (None, 100)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 10)                1010      
=================================================================
Total params: 82,934
Trainable params: 82,934
Non-trainable params: 0
_________________________________________________________________

batch_size=200

# 使用tensorboard視覺化訓練過程
# logs是存放tensorboard檔案的目錄
model_demo1.compile(loss='sparse_categorical_crossentropy',
              optimizer='adam', metrics=['acc'])
from tensorflow.keras.callbacks import TensorBoard
tbCallBack = TensorBoard(log_dir='./log', histogram_freq=1,
                         write_graph=True,
                         write_grads=True,
                         batch_size=batch_size,
                         write_images=True)

WARNING:tensorflow:`write_grads` will be ignored in TensorFlow 2.0 for the `TensorBoard` Callback.
WARNING:tensorflow:`batch_size` is no longer needed in the `TensorBoard` Callback and will be ignored in TensorFlow 2.0.

# 開始訓練
history = model_demo1.fit(x=x_train, y=y_train, validation_split=0.2,
                    epochs=1, batch_size=200, verbose=1
                    ,validation_data=(x_test, y_test)
            
                    ,callbacks=[tbCallBack])

240/240 [==============================] - 32s 132ms/step - loss: 0.4900 - acc: 0.8523 - val_loss: 0.1351 - val_acc: 0.9604

%load_ext tensorboard
%tensorboard --logdir './log' 
#定位tensorboard讀取的檔案目錄
# logs是存放tensorboard檔案的目錄

The tensorboard extension is already loaded. To reload it, use:
  %reload_ext tensorboard



Reusing TensorBoard on port 6008 (pid 975), started 0:13:42 ago. (Use '!kill 975' to kill it.)



<IPython.core.display.Javascript object>

x_train.shape

(60000, 28, 28, 1)

import tensorflow as tf

# 使用LeNet網路
def model_LeNet():
  model = tf.keras.Sequential()
  #28*28->14*14
  
  model.add(tf.keras.layers.MaxPool2D((2,2),input_shape=(28,28,1)))
  #卷積層：16個卷積核，尺寸為5*5，引數個數16*（5*5*6+1）=2416,維度變為（10，10，16）
  model.add(tf.keras.layers.Conv2D(16,(5,5),activation='relu'))
  #池化層，維度變化為（5，5，16）
  model.add(tf.keras.layers.MaxPool2D((2,2)))
  #拉直，維度（400，）
  model.add(tf.keras.layers.Flatten())
  #全連線層,引數量400*120+120=48120
  model.add(tf.keras.layers.Dense(120,activation='relu'))
  #全連線層,引數個數84*120+84=10164
  model.add(tf.keras.layers.Dense(84,activation='relu'))
  #全連線層,引數個數10*120+10=130,
  model.add(tf.keras.layers.Dense(10,activation='softmax'))
  return model

model_demo2=model_LeNet()
model_demo2.compile(loss='sparse_categorical_crossentropy',
              optimizer='adam', metrics=['acc'])
from tensorflow.keras.callbacks import TensorBoard
tbCallBack = TensorBoard(log_dir='./log', histogram_freq=1,
                         write_graph=True,
                         write_grads=True,
                         batch_size=batch_size,
                         write_images=True)

# 開始訓練
history = model_demo2.fit(x=x_train, y=y_train, validation_split=0.2,
                    epochs=1, batch_size=200, verbose=1
                    ,validation_data=(x_test, y_test)
            
                    ,callbacks=[tbCallBack])

WARNING:tensorflow:`write_grads` will be ignored in TensorFlow 2.0 for the `TensorBoard` Callback.
WARNING:tensorflow:`batch_size` is no longer needed in the `TensorBoard` Callback and will be ignored in TensorFlow 2.0.
240/240 [==============================] - 6s 24ms/step - loss: 0.4880 - acc: 0.8655 - val_loss: 0.1780 - val_acc: 0.9462

%load_ext tensorboard
%tensorboard --logdir './log' 
#定位tensorboard讀取的檔案目錄
# logs是存放tensorboard檔案的目錄

The tensorboard extension is already loaded. To reload it, use:
  %reload_ext tensorboard



Reusing TensorBoard on port 6008 (pid 975), started 0:28:04 ago. (Use '!kill 975' to kill it.)



<IPython.core.display.Javascript object>

x_train[0].shape

(28, 28, 1)

def lenet():
  # 輸入需要為32*32
  model=keras.Sequential()
  model.add(keras.layers.Conv2D(6,(1,1),activation='relu',input_shape=(28,28,1)))
  # 28*28*6
  #池化層,沒有引數，維度變化為（14，14，6）
  model.add(tf.keras.layers.MaxPool2D((2,2)))
  #卷積層：16個卷積核，尺寸為5*5，引數個數16*（5*5*6+1）=2416,維度變為（10，10，16）
  model.add(tf.keras.layers.Conv2D(16,(5,5),activation='relu'))
  #池化層，維度變化為（5，5，16）
  model.add(tf.keras.layers.MaxPool2D((2,2)))
  #拉直，維度（400，）
  model.add(tf.keras.layers.Flatten())
  #全連線層,引數量400*120+120=48120
  model.add(tf.keras.layers.Dense(120,activation='relu'))
  #全連線層,引數個數84*120+84=10164
  model.add(tf.keras.layers.Dense(84,activation='relu'))
  #全連線層,引數個數10*120+10=130,
  model.add(tf.keras.layers.Dense(10,activation='softmax'))
  model.summary()
  return model