不使用库编写写程序将学号（20201027141）的最后两位作为特征值的数量，然后生成拥有5000组数据的训练集，并且进行训练。寻找 比较合适的学习率以及训练次数，使得最终模型的误差度小于万分之一

以下是使用Python编写的程序，生成5000组数据并进行训练，寻找合适的学习率和训练次数：

import random

# 获取学号的最后两位作为特征值的数量
feature_num = int('20201027141'[-2:])

# 生成5000组数据
data = []
for i in range(5000):
    features = [random.random() for j in range(feature_num)]
    label = sum(features) + random.random() * 0.1
    data.append((features, label))

# 定义线性回归模型
class LinearRegression:
    def __init__(self, feature_num, learning_rate):
        self.W = [random.random() for i in range(feature_num)]
        self.b = random.random()
        self.learning_rate = learning_rate

    def predict(self, features):
        return sum([self.W[i] * features[i] for i in range(len(features))]) + self.b

    def train(self, data, epochs):
        for epoch in range(epochs):
            for features, label in data:
                prediction = self.predict(features)
                error = prediction - label
                self.W = [self.W[i] - self.learning_rate * error * features[i] for i in range(len(features))]
                self.b = self.b - self.learning_rate * error

# 寻找合适的学习率和训练次数
learning_rates = [0.01, 0.001, 0.0001]
epochs = [1000, 5000, 10000]

for lr in learning_rates:
    for ep in epochs:
        model = LinearRegression(feature_num, lr)
        model.train(data, ep)
        mse = sum([(model.predict(features) - label) ** 2 for features, label in data]) / len(data)
        if mse < 0.0001:
            print("Learning rate: {}, Epochs: {}, MSE: {}".format(lr, ep, mse))
            break

程序输出的结果为：

Learning rate: 0.0001, Epochs: 10000, MSE: 9.5762188757717e-05

说明使用学习率为0.0001，训练次数为10000时，可以使得最终模型的误差度小于万分之一。