导入需要的模块

import numpy as np
from matplotlib import pyplot as plt
import tensorflow as tf

定义需要的参数类

class Config:def __init__(self):self.save_path = './model_sin/sin_cos'  # 训练结果保存的路径self.lr = 0.001  # 学习步长self.epoches = 1000  # 迭代的次数self.samples = 200  # 样本的数量self.hidden_units = 200  # 隐藏层神经元的数量

定义样本类

class Sample:def __init__(self,samples):self.xs = np.random.uniform(-np.pi, np.pi, samples)  # 使用Numpy随机生成点self.xs = sorted(self.xs)  # 对生成的点进行排序（否则会乱）self.ys = np.sin(self.xs), np.cos(self.xs), np.square(self.xs)  # 定义需要拟合的三个函数 

定义张量类

class Tensors:def __init__(self,config):self.x = tf.placeholder(tf.float32,[None],'x')  # 定义输入样本点x的值self.y = tf.placeholder(tf.float32,[3,None],'y')  # 定义输入样本点y的值x = tf.reshape(self.x,[-1,1])  # 将所输入的标量转换为向量x = tf.layers.dense(x, config.hidden_units, tf.nn.relu)  # 使用全连接进行函数self.y_predict = tf.layers.dense(x,3)  # 全连接获取预测的y值y = tf.transpose(self.y)  # 对输入的y值进行转置，方便进行后续操作self.loss = tf.reduce_mean(tf.square(self.y_predict - y)) # 获取损失函数的值self.lr = tf.placeholder(tf.float32, [], 'lr')  # 定义学习步长opt = tf.train.AdamOptimizer(config.lr)  # 梯度下降优化器，进行拟合self.train_op = opt.minimize(self.loss)  # 获取dertx的值self.loss = tf.sqrt(self.loss) # 对获取的损失值进行开平方操作，使误差减少

定义应用类
应用类应该包含：

• 样本的训练方法
• 样本的预测方法

class SinApp:def __init__(self,config):self.config = configself.ts = Tensors(config)self.session = tf.Session()self.saver = tf.train.Saver()try:self.saver.restore(self.session, config.save_path)print(f"restore model from {config.save_path} successfully")except:print(f"fail to restore the model from {config.save_path}")self.session.run(tf.global_variables_initializer())def train(self):  # 训练方法samples = Sample(self.config.samples)cfg = self.configts = self.tsfor _ in range(cfg.epoches):self.session.run(ts.train_op,{ts.x:samples.xs, ts.y : samples.ys, ts.lr:cfg.lr})self.save()return samples.xs,samples.ysdef predict(self):  # 预测方法samples = Sample(400)ys = self.session.run(self.ts.y_predict,{self.ts.x:samples.xs})return samples.xs, ysdef save(self):self.saver.save(self.session, self.config.save_path)    def close(self):self.session.close()

定义主方法打印拟合的图像

if __name__ == "__main__":app = SinApp(Config())xs_train, ys_train = app.train()xs_predict, ys_presict = app.predict()ys_train = np.transpose(ys_train)plt.plot(xs_train, ys_train)plt.plot(xs_predict, ys_presict)plt.legend(['sin', 'cos', 'square', 'predict_sin', 'predict_cos', 'predict_square'])plt.show()

运行结果为

样本的训练结果