TensorFlow-2.x-09-LeNet-5的實現

本章節我們進入卷積神經網絡的實現，LeNet-5可以說是進入卷積神經網絡的入門級別，包含卷積層塊和全連接層塊兩個部分。
利用TF-2.x可以很方便的構建模型，主要構建模型的方法如下：

def model():
    net=tf.keras.Sequential()
    net.add(tf.keras.layers.Conv2D(filters=6,kernel_size=5,
                                   activation='sigmoid',input_shape=(28,28,1)))
    net.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
    net.add(tf.keras.layers.Conv2D(filters=16,kernel_size=5,activation='sigmoid'))
    net.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
    net.add(tf.keras.layers.Flatten())
    net.add(tf.keras.layers.Dense(120,activation='sigmoid'))
    net.add(tf.keras.layers.Dense(84,activation='sigmoid'))
    net.add(tf.keras.layers.Dense(10,activation='sigmoid'))
    return net

本章主要使用兩種訓練方法來實現LeNet-5的調參訓練：
1）自定義方法

import tensorflow as tf
from tensorflow.keras.datasets import fashion_mnist
from tensorflow import data as tfdata
import numpy as np

# 1、讀取數據集
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
train_images = tf.reshape(x_train, (x_train.shape[0],x_train.shape[1],x_train.shape[2], 1))
test_images = tf.reshape(x_test, (x_test.shape[0],x_test.shape[1],x_test.shape[2], 1))
#2、讀取模型
batch_size = 64
train_dataset = tfdata.Dataset.from_tensor_slices((train_images, y_train)).shuffle(train_images.shape[0])
test_dataset = tfdata.Dataset.from_tensor_slices((test_images, y_test))
train_iter = train_dataset.batch(batch_size)
test_iter = test_dataset.batch(batch_size)

for x,y in train_iter:
    print(x.shape,y.shape)
    break


# 定義模型
def model():
    net=tf.keras.Sequential()
    net.add(tf.keras.layers.Conv2D(filters=6,kernel_size=5,
                                   activation='sigmoid',input_shape=(28,28,1)))
    net.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
    net.add(tf.keras.layers.Conv2D(filters=16,kernel_size=5,activation='sigmoid'))
    net.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
    net.add(tf.keras.layers.Flatten())
    net.add(tf.keras.layers.Dense(120,activation='sigmoid'))
    net.add(tf.keras.layers.Dense(84,activation='sigmoid'))
    net.add(tf.keras.layers.Dense(10,activation='sigmoid'))
    return net
net=model()
net.summary()
'''-------第一種------'''

#定義損失函數
def loss_fu(y_pred, y_true):
    loss = tf.losses.sparse_categorical_crossentropy(y_true, y_pred)
    return loss
#定義優化器
trainer = tf.keras.optimizers.SGD(learning_rate=0.03)
# 計算準確率
def accuracy(y_hat, y):
    return np.mean((tf.cast(tf.argmax(y_hat, axis=1), tf.int32) == tf.cast(y, tf.int32)))
# SGD優化器
def sgd(params, lr, batch_size, grads):
    for i, param in enumerate(params):
        print(param)
        param.assign_sub(lr * grads[i] / batch_size)
# 計算測試集準確率
def evaluate_accuracy(test_iter,net):
    acc_sum, n = 0.0, 0
    for X, y in test_iter:
        X=tf.cast(X,tf.float32)
        n += 1
        y = tf.cast(y, dtype=tf.int32)
        output = net(X)
        acc_sum += accuracy(output, y)
    return acc_sum / n

#訓練
'''
通過調用tf.GradientTape記錄動態圖梯度，執行tape.gradient獲得動態圖中各變量梯度。
通過 model.trainable_variables 找到需要更新的變量，並用 trainer.apply_gradients 更新權重，
完成一步訓練。
'''
params=net.trainable_variables #獲得模型結構中需要更新的變量參數
trainer = tf.keras.optimizers.SGD(learning_rate=0.9)
num_epochs = 50
for epoch in range(0, num_epochs ):
    train_acc_sum,train_loss, num_count =0.0, 0, 0
    for X, y in train_iter:
        X=tf.cast(X,dtype=tf.float32)
        y = tf.cast(y, dtype=tf.int32)
        num_count += 1  # 一個epoch取多少次
        with tf.GradientTape() as t:
            t.watch(params)
            l = tf.reduce_mean(loss_fu(net(X,training=True), y))

        grads = t.gradient(l, params)  # 通過調用反向函數t.gradients計算小批量隨機梯度，並調用優化算法sgd迭代模型參數
        # sgd(params, 0.01, batch_size, grads)
        trainer.apply_gradients(zip(grads, params))
        train_acc_sum += accuracy(net(X), y)
        train_loss += l.numpy()
    test_acc = evaluate_accuracy(test_iter,net)
    print("epoch %d, train loss:%.6f, train acc:%.5f, test acc:%.5f" % (
    epoch, train_loss / num_count, train_acc_sum / num_count, test_acc))

訓練結果：

2）Keras高階API
這種方法就比較簡單了，幾行代碼就能解決：

import tensorflow as tf
from tensorflow.keras.datasets import fashion_mnist
from tensorflow import data as tfdata
import numpy as np

# 1、讀取數據集
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
train_images = tf.reshape(x_train, (x_train.shape[0],x_train.shape[1],x_train.shape[2], 1))
test_images = tf.reshape(x_test, (x_test.shape[0],x_test.shape[1],x_test.shape[2], 1))
#2、讀取模型
batch_size = 64
train_dataset = tfdata.Dataset.from_tensor_slices((train_images, y_train)).shuffle(train_images.shape[0])
test_dataset = tfdata.Dataset.from_tensor_slices((test_images, y_test))
train_iter = train_dataset.batch(batch_size)
test_iter = test_dataset.batch(batch_size)
for x,y in train_iter:
    print(x.shape,y.shape)
    break


# 定義模型
def model():
    net=tf.keras.Sequential()
    net.add(tf.keras.layers.Conv2D(filters=6,kernel_size=5,
                                   activation='sigmoid',input_shape=(28,28,1)))
    net.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
    net.add(tf.keras.layers.Conv2D(filters=16,kernel_size=5,activation='sigmoid'))
    net.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
    net.add(tf.keras.layers.Flatten())
    net.add(tf.keras.layers.Dense(120,activation='sigmoid'))
    net.add(tf.keras.layers.Dense(84,activation='sigmoid'))
    net.add(tf.keras.layers.Dense(10,activation='sigmoid'))
    return net
net=model()
net.summary()

optimizer = tf.keras.optimizers.SGD(learning_rate=0.9, momentum=0.0, nesterov=False)

net.compile(optimizer=optimizer,
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
net.fit(train_images, y_train, epochs=10, validation_data=(test_images,y_test))

訓練結果：

附上本章節所有代碼：

import tensorflow as tf
from tensorflow.keras.datasets import fashion_mnist
from tensorflow import data as tfdata
import numpy as np

# 1、讀取數據集
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
train_images = tf.reshape(x_train, (x_train.shape[0],x_train.shape[1],x_train.shape[2], 1))
test_images = tf.reshape(x_test, (x_test.shape[0],x_test.shape[1],x_test.shape[2], 1))
#2、讀取模型
batch_size = 64
train_dataset = tfdata.Dataset.from_tensor_slices((train_images, y_train)).shuffle(train_images.shape[0])
test_dataset = tfdata.Dataset.from_tensor_slices((test_images, y_test))
train_iter = train_dataset.batch(batch_size)
test_iter = test_dataset.batch(batch_size)

for x,y in train_iter:
    print(x.shape,y.shape)
    break


# 定義模型
def model():
    net=tf.keras.Sequential()
    net.add(tf.keras.layers.Conv2D(filters=6,kernel_size=5,
                                   activation='sigmoid',input_shape=(28,28,1)))
    net.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
    net.add(tf.keras.layers.Conv2D(filters=16,kernel_size=5,activation='sigmoid'))
    net.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
    net.add(tf.keras.layers.Flatten())
    net.add(tf.keras.layers.Dense(120,activation='sigmoid'))
    net.add(tf.keras.layers.Dense(84,activation='sigmoid'))
    net.add(tf.keras.layers.Dense(10,activation='sigmoid'))
    return net
net=model()
net.summary()
# '''-------第一種------'''
#
# #定義損失函數
# def loss_fu(y_pred, y_true):
#     loss = tf.losses.sparse_categorical_crossentropy(y_true, y_pred)
#     return loss
# #定義優化器
# trainer = tf.keras.optimizers.SGD(learning_rate=0.03)
# # 計算準確率
# def accuracy(y_hat, y):
#     return np.mean((tf.cast(tf.argmax(y_hat, axis=1), tf.int32) == tf.cast(y, tf.int32)))
# # SGD優化器
# def sgd(params, lr, batch_size, grads):
#     for i, param in enumerate(params):
#         print(param)
#         param.assign_sub(lr * grads[i] / batch_size)
# # 計算測試集準確率
# def evaluate_accuracy(test_iter,net):
#     acc_sum, n = 0.0, 0
#     for X, y in test_iter:
#         X=tf.cast(X,tf.float32)
#         n += 1
#         y = tf.cast(y, dtype=tf.int32)
#         output = net(X)
#         acc_sum += accuracy(output, y)
#     return acc_sum / n
#
# #訓練
# '''
# 通過調用tf.GradientTape記錄動態圖梯度，執行tape.gradient獲得動態圖中各變量梯度。
# 通過 model.trainable_variables 找到需要更新的變量，並用 trainer.apply_gradients 更新權重，
# 完成一步訓練。
# '''
# params=net.trainable_variables #獲得模型結構中需要更新的變量參數
# trainer = tf.keras.optimizers.SGD(learning_rate=0.9)
# num_epochs = 50
# for epoch in range(0, num_epochs ):
#     train_acc_sum,train_loss, num_count =0.0, 0, 0
#     for X, y in train_iter:
#         X=tf.cast(X,dtype=tf.float32)
#         y = tf.cast(y, dtype=tf.int32)
#         num_count += 1  # 一個epoch取多少次
#         with tf.GradientTape() as t:
#             t.watch(params)
#             l = tf.reduce_mean(loss_fu(net(X,training=True), y))
#
#         grads = t.gradient(l, params)  # 通過調用反向函數t.gradients計算小批量隨機梯度，並調用優化算法sgd迭代模型參數
#         # sgd(params, 0.01, batch_size, grads)
#         trainer.apply_gradients(zip(grads, params))
#         train_acc_sum += accuracy(net(X), y)
#         train_loss += l.numpy()
#     test_acc = evaluate_accuracy(test_iter,net)
#     print("epoch %d, train loss:%.6f, train acc:%.5f, test acc:%.5f" % (
#     epoch, train_loss / num_count, train_acc_sum / num_count, test_acc))

'''------第二種------'''
optimizer = tf.keras.optimizers.SGD(learning_rate=0.9, momentum=0.0, nesterov=False)

net.compile(optimizer=optimizer,
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
net.fit(train_images, y_train, epochs=10, validation_data=(test_images,y_test))

TensorFlow-2.x-09-LeNet-5的實現

記一次 .NET某工業設計軟件崩潰分析

創建 Vue3 項目

TS + Webpack 整合 Jest

分享5款.NET開源免費的Redis客戶端組件庫

安卓手機如何登錄抖音境外版

golang開發 gorilla websocket的使用

面試官：如果不允許線程池丟棄任務，應該選擇哪個拒絕策略？

Mac卸載 Node npm，升級 Node

嵌入式汽車電子學習路線

uni.showModel內容換行

OpenCV-C++-CUDA-06-形態學二值化操作

OpenCV-C++-CUDA-07-基於顏色的物體跟蹤

OpenCV-C++-CUDA-08-角點（特徵點）檢測

MXNET深度學習框架-06-使用gluon實現邏輯迴歸

OpenCV-C++-CUDA-05-高斯雙邊加速，實時美顏

https://yachay.unat.edu.pe/blog/index.php?comment_area=format_blog&comment_component=blog&comment_co

linux以太網驅動總結