1、数据获取

• 从sklearn中获取鸢尾花数据，并合并处理

from sklearn.datasets import load_iris
import pandas as pd

x_data = load_iris().data
y_data = load_iris().target

x_data = pd.DataFrame(x_data, columns=['花萼长度','花萼宽度','花瓣长度','花瓣宽度'])
pd.set_option('display.unicode.east_asian_width', True)

x_data['类别'] = y_data
x_data

2、数据集构建

• 数据集构建包括：
  • 数据读取
  • 数据打乱
  • 数据划分
  • 小批量迭代器生成

import tensorflow as tf
import numpy as np
from sklearn.datasets import load_iris

# 1、从sklearn包中datasets读取数据集
x_data = load_iris().data
y_data = load_iris().target

# 2、数据打乱
np.random.seed(1)   # 使用相同的seed，使输入特征/标签一一对应
np.random.shuffle(x_data)
np.random.seed(1)
np.random.shuffle(y_data)
tf.random.set_seed(1)

# 3、训练集、测试集划分
x_train, x_test = x_data[:-30], x_data[-30:]
y_train, y_test = y_data[:-30], y_data[-30:]

# 4、小批量数据
train_db = tf.data.Dataset.from_tensor_slices((x_train, y_train)).batch(32)
train_db = tf.data.Dataset.from_tensor_slices((x_test, y_test)).batch(32)

3、模型的训练验证

# 定义超参数，预设变量
lr = 0.1
loss_all = 0
Epoch = 500
train_loss_list = []
test_acc = []

# 定义神经网络的可训练参数
w = tf.Variable(tf.random.truncated_normal([4,3], stddev=0.1, seed=1))
b = tf.Variable(tf.random.truncated_normal([3], stddev=0.1, seed=1))

# 循环迭代，训练参数
for epoch in range(Epoch):
    for step, (x_, y_) in enumerate(train_db):
        with tf.GradientTape() as tape:
            x_ = tf.cast(x_, tf.float32)
            y_pre = tf.matmul(x_, w) + b
            y_pre = tf.nn.softmax(y_pre)

            y_lab = tf.one_hot(y_, depth=3)
            loss = tf.reduce_mean(tf.square(y_lab - y_pre))
            loss_all += loss.numpy()

        grads = tape.gradient(loss, [w,b])
        w.assign_sub(lr * grads[0])
        b.assign_sub(lr * grads[1])
    print(f'Epoch: {epoch}, loss: {loss_all/4}')
    train_loss_list.append(loss_all/4)
    loss_all = 0

    # 测试部分
    total_correct, total_number = 0, 0
    for x_,y_ in test_db:
        x_ = tf.cast(x_, tf.float32)
        y_pre = tf.matmul(x_, w) + b
        y_pre = tf.nn.softmax(y_pre)
        y_p = tf.argmax(y_pre, axis=1)
        y_p = tf.cast(y_p, dtype=y_.dtype)

        correct = tf.cast(tf.equal(y_p, y_), dtype=tf.int32)
        correct = tf.reduce_sum(correct)
        total_correct += int(correct) 
        total_number += x_.shape[0]
    acc = total_correct / total_number
    test_acc.append(acc)
    print("Test_acc:", acc)
    print("-"*30)

可视化训练过程

# 绘制测试Acc曲线和训练loss曲线
import matplotlib.pyplot as plt
fig, ax = plt.subplots()
ax.plot(train_loss_list,'b-')
ax.set_xlabel('Epoch')
ax.set_ylabel('Loss')

ax1 = ax.twinx()
ax1.plot(test_acc,'r-')
ax1.set_ylabel('Acc')

ax1.spines['left'].set_color('blue')
ax1.spines['right'].set_color('red')