［tensorflow认证考试］01 构建和训练一个神经网络模型
01 构建和训练一个神经网络模型(Build and train neura
network models using TensorFlow 2.x)
本期文章是一个系列文章的开始，本文是第一篇，训练一个基础的神经网络，想学习tensorflow的可以关注一波，后续作者将以通过tensorflow认证考试为目标的系列学习笔记．
(1)Build and train neural network models using TensorFlow 2.x
(2)Image classification
(3)Natural language processing(NLP)
(4)Time series, sequences and predictions
Build and train neural network models using TensorFlow 2.x
使用tensorflow构建和训练一个神经网络模型
环境
tensorflow认证是基于tensorflow 2.x进行的，所以复习的时候一定要用这个版本的tensorflow进行复习
训练神经网络模型的步骤
准备数据集
本次使用mnist数据集
(x_train,y_train),(x_test,y_test)=tf.keras.datasets.mnist.load_data()print(x_train.shape)print(y_train.shape)
输出
(60000, 28, 28)
(60000, )
数据预处理
因为图像是 0~255 之间的整形，需要进行归一化到 0~1 区间，这个目的为了使模型精度更高．01区间收敛速度更快．具体这块可以单独展开讲．
x_train,x_test=x_train/255.0,x_test/255.0
搭建模型
模型这块使用的是顺序模型
第一层是平坦化，将二维变成一维，作为输入
第二层是普通的Dense层, 激活函数为relu
第三层是一个Drop层，防止过拟合的, drop的比例为0.2
第四层是一个输出层，激活函数为softmax，输出多个分类的概率值
model=tf.keras.models.Sequential( [ tf.keras.layers.Flatten(input_shape=(28,28)), tf.keras.layers.Dense(128,activation="relu"), tf.keras.layers.Dropout(0.2), tf.keras.layers.Dense(10,activation="softmax") ] )model.compile(optimizer="adam",loss="sparse_categorical_crossentropy",metrics=["accuracy"])
训练
history=model.fit(x_train,y_train,epochs=10)
评估
model.evaluate(x_test,y_test)
输出
Epoch1/101875/1875[==============================]-2s 1ms/step - loss: 0.2974 - accuracy:0.9141Epoch2/101875/1875[==============================]-3s 2ms/step - loss: 0.1439 - accuracy:0.9574Epoch3/101875/1875[==============================]-2s 1ms/step - loss: 0.1076 - accuracy:0.9684Epoch4/101875/1875[==============================]-3s 2ms/step - loss: 0.0858 - accuracy:0.9736Epoch5/101875/1875[==============================]-3s 1ms/step - loss: 0.0738 - accuracy:0.9764Epoch6/101875/1875[==============================]-2s 1ms/step - loss: 0.0644 - accuracy:0.9793Epoch7/101875/1875[==============================]-2s 1ms/step - loss: 0.0559 - accuracy:0.9820Epoch8/101875/1875[==============================]-3s 1ms/step - loss: 0.0532 - accuracy:0.9826Epoch9/101875/1875[==============================]-4s 2ms/step - loss: 0.0483 - accuracy:0.9839Epoch10/101875/1875[==============================]-2s 1ms/step - loss: 0.0420 - accuracy:0.98622020-12-06 22:38:52.602650:Wtensorflow/core/framework/cpu_allocator_impl.cc:81]Allocationof31360000exceeds10%offreesystemmemory.313/313[==============================]-0s 968us/step - loss: 0.0680 - accuracy:0.9797{'loss':[0.2974238991737366,0.14386004209518433,0.10755111277103424,0.08577871322631836,0.07378625124692917,0.06444691121578217,0.055944059044122696,0.05322827026247978,0.04829448089003563,0.04199640452861786],'accuracy':[0.9140833616256714,0.9573666453361511,0.9684333205223083,0.973633348941803,0.9763833284378052,0.9793499708175659,0.9820166826248169,0.9825666546821594,0.9839333295822144,0.9861666560173035]}
总结
主要是熟悉了基于tensorflow 2.x基于自带数据集如何搭建一个简单的机器学习模型．
上面的是一个手写数字识别的数据集，应该是一个图像识别分类任务．
从最终结果来看，在测试集中的准确率低于训练集，有一定程度上的过拟合．
主要掌握就是顺序模型的搭建，至于损失函数，参数归一化，优化器，学习率，dropout, pooling，激活函数，反向传播，正向传播等知识，需要单独去学习．
完整代码
https://github.com/xcrossed/blog/blob/master/code/tf/01.py
