由淺入深學(xué)習(xí)TensorFlow MNIST 數(shù)據(jù)集
發(fā)布日期：2022-01-19 08:53 | 文章來源：站長之家
MNIST 數(shù)據(jù)集介紹
MNIST 包含 0~9 的手寫數(shù)字, 共有 60000 個訓(xùn)練集和 10000 個測試集. 數(shù)據(jù)的格式為單通道 28*28 的灰度圖.
LeNet 模型介紹
LeNet 網(wǎng)絡(luò)最早由紐約大學(xué)的 Yann LeCun 等人于 1998 年提出, 也稱 LeNet5. LeNet 是神經(jīng)網(wǎng)絡(luò)的鼻祖, 被譽為卷積神經(jīng)網(wǎng)絡(luò)的 “Hello World”.
卷積
池化 (下采樣)
激活函數(shù) (ReLU)
LeNet 逐層分析
1. 第一個卷積層
2. 第一個池化層
3. 第二個卷積層
4. 第二個池化層
5. 全連接卷積層
6. 全連接層
7. 全連接層 (輸出層)
代碼實現(xiàn)
導(dǎo)包
from tensorflow.keras.datasets import mnist from matplotlib import pyplot as plt import numpy as np import tensorflow as tf
讀取 & 查看數(shù)據(jù)
# ------------------1. 讀取 & 查看數(shù)據(jù)------------------ # 讀取數(shù)據(jù) (X_train, y_train), (X_test, y_test) = mnist.load_data() # 數(shù)據(jù)集查看 print(X_train.shape) # (60000, 28, 28) print(y_train.shape) # (60000,) print(X_test.shape) # (10000, 28, 28) print(y_test.shape) # (10000,) print(type(X_train)) # <class 'numpy.ndarray'> # 圖片顯示 plt.imshow(X_train[0], cmap="Greys") # 查看第一張圖片 plt.show()
數(shù)據(jù)預(yù)處理
# ------------------2. 數(shù)據(jù)預(yù)處理------------------ # 格式轉(zhuǎn)換 (將圖片從28*28擴(kuò)充為32*32) X_train = np.pad(X_train, ((0, 0), (2, 2), (2, 2)), "constant", constant_values=0) X_test = np.pad(X_test, ((0, 0), (2, 2), (2, 2)), "constant", constant_values=0) print(X_train.shape) # (60000, 32, 32) print(X_test.shape) # (10000, 32, 32) # 數(shù)據(jù)集格式變換 X_train = X_train.astype(np.float32) X_test = X_test.astype(np.float32) # 數(shù)據(jù)正則化 X_train /= 255 X_test /= 255 # 數(shù)據(jù)維度轉(zhuǎn)換 X_train = np.expand_dims(X_train, axis=-1) X_test = np.expand_dims(X_test, axis=-1) print(X_train.shape) # (60000, 32, 32, 1) print(X_test.shape) # (10000, 32, 32, 1)
模型建立
# 第一個卷積層 conv_layer_1 = tf.keras.layers.Conv2D(filters=6, kernel_size=(5, 5), padding="valid", activation=tf.nn.relu) # 第一個池化層 pool_layer_1 = tf.keras.layers.MaxPool2D(pool_size=(2, 2), padding="same") # 第二個卷積層 conv_layer_2 = tf.keras.layers.Conv2D(filters=16, kernel_size=(5, 5), padding="valid", activation=tf.nn.relu) # 第二個池化層 pool_layer_2 = tf.keras.layers.MaxPool2D(padding="same") # 扁平化 flatten = tf.keras.layers.Flatten() # 第一個全連接層 fc_layer_1 = tf.keras.layers.Dense(units=120, activation=tf.nn.relu) # 第二個全連接層 fc_layer_2 = tf.keras.layers.Dense(units=84, activation=tf.nn.softmax) # 輸出層 output_layer = tf.keras.layers.Dense(units=10, activation=tf.nn.softmax)
卷積 Conv2D 的用法:
- filters: 卷積核個數(shù)
- kernel_size: 卷積核大小
- strides = (1, 1): 步長
- padding = “vaild”: valid 為舍棄, same 為補齊
- activation = tf.nn.relu: 激活函數(shù)
- data_format = None: 默認(rèn) channels_last
池化 AveragePooling2D 的用法:
- pool_size: 池的大小
- strides = (1, 1): 步長
- padding = “vaild”: valid 為舍棄, same 為補齊
- activation = tf.nn.relu: 激活函數(shù)
- data_format = None: 默認(rèn) channels_last
全連接 Dense 的用法:
- units: 輸出的維度
- activation: 激活函數(shù)
- strides = (1, 1): 步長
- padding = “vaild”: valid 為舍棄, same 為補齊
- activation = tf.nn.relu: 激活函數(shù)
- data_format = None: 默認(rèn) channels_last
# 模型實例化 model = tf.keras.models.Sequential([ tf.keras.layers.Conv2D(filters=6, kernel_size=(5, 5), padding='valid', activation=tf.nn.relu,input_shape=(32, 32, 1)), # relu tf.keras.layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2), padding='same'), tf.keras.layers.Conv2D(filters=16, kernel_size=(5, 5), padding='valid', activation=tf.nn.relu), tf.keras.layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2), padding='same'), tf.keras.layers.Flatten(), tf.keras.layers.Dense(units=120, activation=tf.nn.relu), tf.keras.layers.Dense(units=84, activation=tf.nn.relu), tf.keras.layers.Dense(units=10, activation=tf.nn.softmax) ]) # 模型展示 model.summary()
輸出結(jié)果:
訓(xùn)練模型
# ------------------4. 訓(xùn)練模型------------------ # 設(shè)置超參數(shù) num_epochs = 10 # 訓(xùn)練輪數(shù) batch_size = 1000 # 批次大小 learning_rate = 0.001 # 學(xué)習(xí)率
# 定義優(yōu)化器 adam_optimizer = tf.keras.optimizers.Adam(learning_rate) model.compile(optimizer=adam_optimizer,loss=tf.keras.losses.sparse_categorical_crossentropy,metrics=['accuracy'])
complie 的用法:
- optimizer: 優(yōu)化器
- loss: 損失函數(shù)
- metrics: 評價
with tf.Session() as sess: # 初始化所有變量 init = tf.global_variables_initializer() sess.run(init) model.fit(x=X_train,y=y_train,batch_size=batch_size,epochs=num_epochs) # 評估指標(biāo) print(model.evaluate(X_test, y_test)) # loss value & metrics values
輸出結(jié)果:
fit 的用法:
- x: 訓(xùn)練集
- y: 測試集
- batch_size: 批次大小
- enpochs: 訓(xùn)練遍數(shù)
保存模型
# ------------------5. 保存模型------------------
model.save('lenet_model.h5')
流程總結(jié)
完整代碼
from tensorflow.keras.datasets import mnist
from matplotlib import pyplot as plt
import numpy as np
import tensorflow as tf
# ------------------1. 讀取 & 查看數(shù)據(jù)------------------
# 讀取數(shù)據(jù)
(X_train, y_train), (X_test, y_test) = mnist.load_data()
# 數(shù)據(jù)集查看
print(X_train.shape) # (60000, 28, 28)
print(y_train.shape) # (60000,)
print(X_test.shape) # (10000, 28, 28)
print(y_test.shape) # (10000,)
print(type(X_train)) # <class 'numpy.ndarray'>
# 圖片顯示
plt.imshow(X_train[0], cmap="Greys") # 查看第一張圖片
plt.show()
# ------------------2. 數(shù)據(jù)預(yù)處理------------------
# 格式轉(zhuǎn)換 (將圖片從28*28擴(kuò)充為32*32)
X_train = np.pad(X_train, ((0, 0), (2, 2), (2, 2)), "constant", constant_values=0)
X_test = np.pad(X_test, ((0, 0), (2, 2), (2, 2)), "constant", constant_values=0)
print(X_train.shape) # (60000, 32, 32)
print(X_test.shape) # (10000, 32, 32)
# 數(shù)據(jù)集格式變換
X_train = X_train.astype(np.float32)
X_test = X_test.astype(np.float32)
# 數(shù)據(jù)正則化
X_train /= 255
X_test /= 255
# 數(shù)據(jù)維度轉(zhuǎn)換
X_train = np.expand_dims(X_train, axis=-1)
X_test = np.expand_dims(X_test, axis=-1)
print(X_train.shape) # (60000, 32, 32, 1)
print(X_test.shape) # (10000, 32, 32, 1)
# ------------------3. 模型建立------------------
# 第一個卷積層
conv_layer_1 = tf.keras.layers.Conv2D(filters=6, kernel_size=(5, 5), padding="valid", activation=tf.nn.relu)
# 第一個池化層
pool_layer_1 = tf.keras.layers.MaxPool2D(pool_size=(2, 2), padding="same")
# 第二個卷積層
conv_layer_2 = tf.keras.layers.Conv2D(filters=16, kernel_size=(5, 5), padding="valid", activation=tf.nn.relu)
# 第二個池化層
pool_layer_2 = tf.keras.layers.MaxPool2D(padding="same")
# 扁平化
flatten = tf.keras.layers.Flatten()
# 第一個全連接層
fc_layer_1 = tf.keras.layers.Dense(units=120, activation=tf.nn.relu)
# 第二個全連接層
fc_layer_2 = tf.keras.layers.Dense(units=84, activation=tf.nn.softmax)
# 輸出層
output_layer = tf.keras.layers.Dense(units=10, activation=tf.nn.softmax)
# 模型實例化
model = tf.keras.models.Sequential([
tf.keras.layers.Conv2D(filters=6, kernel_size=(5, 5), padding='valid', activation=tf.nn.relu,input_shape=(32, 32, 1)),
# relu
tf.keras.layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2), padding='same'),
tf.keras.layers.Conv2D(filters=16, kernel_size=(5, 5), padding='valid', activation=tf.nn.relu),
tf.keras.layers.AveragePooling2D(pool_size=(2, 2), strides=(2, 2), padding='same'),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(units=120, activation=tf.nn.relu),
tf.keras.layers.Dense(units=84, activation=tf.nn.relu),
tf.keras.layers.Dense(units=10, activation=tf.nn.softmax)
])
# 模型展示
model.summary()
# ------------------4. 訓(xùn)練模型------------------
# 設(shè)置超參數(shù)
num_epochs = 10 # 訓(xùn)練輪數(shù)
batch_size = 1000 # 批次大小
learning_rate = 0.001 # 學(xué)習(xí)率
# 定義優(yōu)化器
adam_optimizer = tf.keras.optimizers.Adam(learning_rate)
model.compile(optimizer=adam_optimizer,loss=tf.keras.losses.sparse_categorical_crossentropy,metrics=['accuracy'])
with tf.Session() as sess:
# 初始化所有變量
init = tf.global_variables_initializer()
sess.run(init)
model.fit(x=X_train,y=y_train,batch_size=batch_size,epochs=num_epochs)
# 評估指標(biāo)
print(model.evaluate(X_test, y_test)) # loss value & metrics values
# ------------------5. 保存模型------------------
model.save('lenet_model.h5')
到此這篇關(guān)于由淺入深學(xué)習(xí)TensorFlow MNIST 數(shù)據(jù)集的文章就介紹到這了,更多相關(guān)TensorFlow MNIST 數(shù)據(jù)集內(nèi)容請搜索本站以前的文章或繼續(xù)瀏覽下面的相關(guān)文章希望大家以后多多支持本站！
版權(quán)聲明：本站文章來源標(biāo)注為YINGSOO的內(nèi)容版權(quán)均為本站所有，歡迎引用、轉(zhuǎn)載，請保持原文完整并注明來源及原文鏈接。禁止復(fù)制或仿造本網(wǎng)站，禁止在非maisonbaluchon.cn所屬的服務(wù)器上建立鏡像，否則將依法追究法律責(zé)任。本站部分內(nèi)容來源于網(wǎng)友推薦、互聯(lián)網(wǎng)收集整理而來，僅供學(xué)習(xí)參考，不代表本站立場，如有內(nèi)容涉嫌侵權(quán)，請聯(lián)系alex-e#qq.com處理。
相關(guān)文章
關(guān)注官方微信