Neural network implements iris classification
We need three parts to realize the classification of iris using neural network
-
Prepare data,
including data set reading, data set disorder, generating train and test (that is, training set and test set that never meet), and matching the data in the training set and test set into input feature and label pairs -
Build the network
Define all the trainable parameters in the neural network -
Optimize trainable parameters
Use nested loop iteration and with structure to update parameters (in the with structure, obtain the partial derivative of the loss function for each trainable parameter, update these trainable parameters), display the current close -
Test effect In
order to check the effect, the program can add a data set every traversal to display the current accuracy rate, and also draw the change curve of accuracy rate acc and loss function loss.
Let's take a look specifically:
First, we must read in the data set, and use datasets_load_iris to read in the features and tags respectively.
When human beings understand the world, information flows into the brain in a disorderly manner. So the data set fed to the neural network also needs to be shuffled. Using the same random seed is out of order for the data set.
Using the same seed, the input features and labels are also one-to-one correspondence
Take out the first 120 data in the scrambled data set as the training set, and the last 30 data as the test set.
In order to judge the training set and the test set fairly, it is required that there is no intersection between the training set and the test set.
Use from_tensor_slices to pack the input features and label pairs of the training set. Each 32 sets of input feature label pairs are packed into a batch
and fed to the neural network in batches.
Analysis of each part of the code (use of functions):
# 随机打乱数据（因为原始数据是顺序的，顺序不打乱会影响准确率）
# seed: 随机数种子，是一个整数，当设置之后，每次生成的随机数都一样（为方便教学，以保每位同学结果一致）
np.random.seed(116) # 使用相同的seed，保证输入特征和标签一一对应
np.random.shuffle(x_data) # 打乱顺序函数
np.random.seed(116) # 使用相同的seed，保证输入特征和标签一一对应
np.random.shuffle(y_data)
tf.random.set_seed(116)
[Data processing] Usage of Numpy.random.seed() : setting the number in seed() is equivalent to setting a "cornucopia" filled with random numbers, a number represents a "cornucopia", when we are in seed() Set the same seed in the parentheses, the "cornucopia" is the same, so of course the random number taken out will be the same every time (don't think that it is randomly taking numbers from it, as long as the set seeds are the same and taking out the random numbers is the same). If the seed is not set, a different random number will be generated each time. (Note: The values in the seed brackets can basically be set arbitrarily)
numpy.random.shuffle shuffle order function
: This function only shuffles the array along the first index of a multi-dimensional array (in a multi-dimensional matrix, only the first dimension (row) is shuffled)
# 将打乱后的数据集分割为训练集和测试集，训练集为前120行，测试集为后30行
x_train = x_data[:-30]
y_train = y_data[:-30]
x_test = x_data[-30:]
y_test = y_data[-30:]
Sharding operation is nothing to talk about
# 转换x的数据类型，否则后面矩阵相乘时会因数据类型不一致报错
x_train = tf.cast(x_train, tf.float32)
x_test = tf.cast(x_test, tf.float32)
tf.cast() data type conversion
# from_tensor_slices函数使输入特征和标签值一一对应。（把数据集分批次，每个批次batch组数据）
train_db = tf.data.Dataset.from_tensor_slices((x_train, y_train)).batch(32)
test_db = tf.data.Dataset.from_tensor_slices((x_test, y_test)).batch(32)
Usage of tf.data.Dataset.from_tensor_slices() and batch() under tensorFlow2.1
# 生成神经网络的参数，4个输入特征故，输入层为4个输入节点；因为3分类，故输出层为3个神经元
# 用tf.Variable()标记
# 参数可训练
# 使用seed使每次生成的随机数相同（方便教学，使大家结果都一致，在现实使用时不写seed）
w1 = tf.Variable(tf.random.truncated_normal([4, 3], stddev=0.1, seed=1))
b1 = tf.Variable(tf.random.truncated_normal([3], stddev=0.1, seed=1)) # b1是偏差量
#学习率、用于画图的每轮loss、用于画图的每轮test_acc、epoch、loss_all每轮损失和
lr = 0.1 # 学习率为0.1
train_loss_results = [] # 将每轮的loss记录在此列表中，为后续画loss曲线提供数据
test_acc = [] # 将每轮的acc记录在此列表中，为后续画acc曲线提供数据
epoch = 500 # 循环500轮
loss_all = 0 # 每轮分4个step，loss_all记录四个step生成的4个loss的和
tf.truncated_normal understand
# 训练部分
for epoch in range(epoch): # 数据集级别的循环，每个epoch循环一次数据集
for step, (x_train, y_train) in enumerate(train_db): # batch级别的循环 ，每个step循环一个batch
with tf.GradientTape() as tape: # with结构记录梯度信息 梯度下降 with GradienTape() as tape:
y = tf.matmul(x_train, w1) + b1 # 神经网络乘加运算 其中求出y为预测值
y = tf.nn.softmax(y) # 使输出y符合概率分布（此操作后与独热码同量级，可相减求loss） softmax归一化预测值y
y_ = tf.one_hot(y_train, depth=3) # 将标签值转换为独热码格式，方便计算loss和accuracy 将1x1的y_train训练集标签转换为1x3的y_
loss = tf.reduce_mean(tf.square(y_ - y)) # 采用均方误差损失函数mse = mean(sum(y-out)^2)
loss_all += loss.numpy() # 将每个step计算出的loss累加，为后续求loss平均值提供数据，这样计算的loss更准确
# 计算loss对各个参数的梯度
grads = tape.gradient(loss, [w1, b1])
# 实现梯度更新 w1 = w1 - lr * w1_grad b = b - lr * b_grad
w1.assign_sub(lr * grads[0]) # 参数w1自更新
b1.assign_sub(lr * grads[1]) # 参数b自更新
# 每个epoch，打印loss信息
print("Epoch {}, loss: {}".format(epoch, loss_all/4))
train_loss_results.append(loss_all / 4) # 将4个step的loss求平均记录在此变量中
loss_all = 0 # loss_all归零，为记录下一个epoch的loss做准备
# 测试部分
# total_correct为预测对的样本个数, total_number为测试的总样本数，将这两个变量都初始化为0
total_correct, total_number = 0, 0
for x_test, y_test in test_db:
# 使用更新后的参数进行预测
y = tf.matmul(x_test, w1) + b1
y = tf.nn.softmax(y)
pred = tf.argmax(y, axis=1) # 返回y中最大值的索引，即预测的分类
# 将pred转换为y_test的数据类型
pred = tf.cast(pred, dtype=y_test.dtype)
# 若分类正确，则correct=1，否则为0，将bool型的结果转换为int型
correct = tf.cast(tf.equal(pred, y_test), dtype=tf.int32)
# 将每个batch的correct数加起来
correct = tf.reduce_sum(correct)
# 将所有batch中的correct数加起来
total_correct += int(correct)
# total_number为测试的总样本数，也就是x_test的行数，shape[0]返回变量的行数
total_number += x_test.shape[0]
# 总的准确率等于total_correct/total_number
acc = total_correct / total_number
test_acc.append(acc)
print("Test_acc:", acc)
print("--------------------------")
# 绘制 loss 曲线
plt.title('Loss Function Curve') # 图片标题
plt.xlabel('Epoch') # x轴变量名称
plt.ylabel('Loss') # y轴变量名称
plt.plot(train_loss_results, label="$Loss$") # 逐点画出trian_loss_results值并连线，连线图标是Loss
plt.legend() # 画出曲线图标
plt.show() # 画出图像
# 绘制 Accuracy 曲线
plt.title('Acc Curve') # 图片标题
plt.xlabel('Epoch') # x轴变量名称
plt.ylabel('Acc') # y轴变量名称
plt.plot(test_acc, label="$Accuracy$") # 逐点画出test_acc值并连线，连线图标是Accuracy
plt.legend()
plt.show()
The complete code is as follows:
# -*- coding: UTF-8 -*-
# 利用鸢尾花数据集，实现前向传播、反向传播，可视化loss曲线
# 导入所需模块
import tensorflow as tf
from sklearn import datasets
from matplotlib import pyplot as plt
import numpy as np
# 导入数据，分别为输入特征和标签
x_data = datasets.load_iris().data
y_data = datasets.load_iris().target
# 随机打乱数据（因为原始数据是顺序的，顺序不打乱会影响准确率）
# seed: 随机数种子，是一个整数，当设置之后，每次生成的随机数都一样（为方便教学，以保每位同学结果一致）
np.random.seed(116) # 使用相同的seed，保证输入特征和标签一一对应
np.random.shuffle(x_data) # 打乱顺序函数
np.random.seed(116) # 使用相同的seed，保证输入特征和标签一一对应
np.random.shuffle(y_data)
tf.random.set_seed(116)
# 将打乱后的数据集分割为训练集和测试集，训练集为前120行，测试集为后30行
x_train = x_data[:-30]
y_train = y_data[:-30]
x_test = x_data[-30:]
y_test = y_data[-30:]
# 转换x的数据类型，否则后面矩阵相乘时会因数据类型不一致报错
x_train = tf.cast(x_train, tf.float32)
x_test = tf.cast(x_test, tf.float32)
# from_tensor_slices函数使输入特征和标签值一一对应。（把数据集分批次，每个批次batch组数据）
train_db = tf.data.Dataset.from_tensor_slices((x_train, y_train)).batch(32)
test_db = tf.data.Dataset.from_tensor_slices((x_test, y_test)).batch(32)
# 生成神经网络的参数，4个输入特征故，输入层为4个输入节点；因为3分类，故输出层为3个神经元
# 用tf.Variable()标记参数可训练
# 使用seed使每次生成的随机数相同（方便教学，使大家结果都一致，在现实使用时不写seed）
w1 = tf.Variable(tf.random.truncated_normal([4, 3], stddev=0.1, seed=1))
b1 = tf.Variable(tf.random.truncated_normal([3], stddev=0.1, seed=1))
lr = 0.1 # 学习率为0.1
train_loss_results = [] # 将每轮的loss记录在此列表中，为后续画loss曲线提供数据
test_acc = [] # 将每轮的acc记录在此列表中，为后续画acc曲线提供数据
epoch = 500 # 循环500轮
loss_all = 0 # 每轮分4个step，loss_all记录四个step生成的4个loss的和
# 训练部分
for epoch in range(epoch): #数据集级别的循环，每个epoch循环一次数据集
for step, (x_train, y_train) in enumerate(train_db): #batch级别的循环 ，每个step循环一个batch
with tf.GradientTape() as tape: # with结构记录梯度信息
y = tf.matmul(x_train, w1) + b1 # 神经网络乘加运算
y = tf.nn.softmax(y) # 使输出y符合概率分布（此操作后与独热码同量级，可相减求loss）
y_ = tf.one_hot(y_train, depth=3) # 将标签值转换为独热码格式，方便计算loss和accuracy
loss = tf.reduce_mean(tf.square(y_ - y)) # 采用均方误差损失函数mse = mean(sum(y-out)^2)
loss_all += loss.numpy() # 将每个step计算出的loss累加，为后续求loss平均值提供数据，这样计算的loss更准确
# 计算loss对各个参数的梯度
grads = tape.gradient(loss, [w1, b1])
# 实现梯度更新 w1 = w1 - lr * w1_grad b = b - lr * b_grad
w1.assign_sub(lr * grads[0]) # 参数w1自更新
b1.assign_sub(lr * grads[1]) # 参数b自更新
# 每个epoch，打印loss信息
print("Epoch {}, loss: {}".format(epoch, loss_all/4))
train_loss_results.append(loss_all / 4) # 将4个step的loss求平均记录在此变量中
loss_all = 0 # loss_all归零，为记录下一个epoch的loss做准备
# 测试部分
# total_correct为预测对的样本个数, total_number为测试的总样本数，将这两个变量都初始化为0
total_correct, total_number = 0, 0
for x_test, y_test in test_db:
# 使用更新后的参数进行预测
y = tf.matmul(x_test, w1) + b1
y = tf.nn.softmax(y)
pred = tf.argmax(y, axis=1) # 返回y中最大值的索引，即预测的分类
# 将pred转换为y_test的数据类型
pred = tf.cast(pred, dtype=y_test.dtype)
# 若分类正确，则correct=1，否则为0，将bool型的结果转换为int型
correct = tf.cast(tf.equal(pred, y_test), dtype=tf.int32)
# 将每个batch的correct数加起来
correct = tf.reduce_sum(correct)
# 将所有batch中的correct数加起来
total_correct += int(correct)
# total_number为测试的总样本数，也就是x_test的行数，shape[0]返回变量的行数
total_number += x_test.shape[0]
# 总的准确率等于total_correct/total_number
acc = total_correct / total_number
test_acc.append(acc)
print("Test_acc:", acc)
print("--------------------------")
# 绘制 loss 曲线
plt.title('Loss Function Curve') # 图片标题
plt.xlabel('Epoch') # x轴变量名称
plt.ylabel('Loss') # y轴变量名称
plt.plot(train_loss_results, label="$Loss$") # 逐点画出trian_loss_results值并连线，连线图标是Loss
plt.legend() # 画出曲线图标
plt.show() # 画出图像
# 绘制 Accuracy 曲线
plt.title('Acc Curve') # 图片标题
plt.xlabel('Epoch') # x轴变量名称
plt.ylabel('Acc') # y轴变量名称
plt.plot(test_acc, label="$Accuracy$") # 逐点画出test_acc值并连线，连线图标是Accuracy
plt.legend()
plt.show()
Operation result:
the accuracy rate is very low at the beginning 
after 500 rounds, the accuracy rate is 1
