如何保存训练好的模型

Question

请问如何保存训练好的模型？
模型代码如下

#RM,LSTAT-->MEDV
#y=theta_0+theta_1*RM+theta_2*LSTAT+theta_3*PTRATIO+theta_4*RM^2+theta_5*RM*LSTAT+theta_6*RM*LSTAT*PTRATIO
import numpy as np

theta=np.random.rand(7,)
theta

#回归模型
def f(x):
    return np.dot(theta,x.T)
#dot():返回两个数组的点积，x.T为矩阵转置

#评估函数（求均方误差）
def mse(x,y):
    return np.sum((f(x)-y)**2) / len(y)

#训练模型
learning_rate = 0.00001

df['RM2'] = df['RM']**2
df['RM_LSTAT'] = df['RM']*df['LSTAT']
df['RM_LSTAT_PTRATIO'] = df['RM']*df['LSTAT']*df['PTRATIO']
train_x = df[['RM','LSTAT','PTRATIO','RM2','RM_LSTAT','RM_LSTAT_PTRATIO']]
print(train_x)
train_x = (train_x -train_x.mean()) / train_x.std()#标准化， mean()求平均值
train_x = np.insert(train_x.to_numpy(),0,1,axis=1)
train_y = df['MEDV']

for i in range(10000):
    theta = theta - learning_rate * np.dot(f(train_x)-train_y,train_x)
    metrics = mse(train_x,train_y)
    print('第%d轮 均方误差为%.04f theta%s' % (i+1,metrics,str(theta)))
 

Answer 1

基于Monster 组和GPT的调写：

import numpy as np
import pickle

#回归模型
def f(x, theta):
    return np.dot(theta, x.T)

#评估函数（求均方误差）
def mse(x, y, theta):
    return np.sum((f(x, theta) - y)**2) / len(y)

#训练模型
learning_rate = 0.00001

df['RM2'] = df['RM']**2
df['RM_LSTAT'] = df['RM'] * df['LSTAT']
df['RM_LSTAT_PTRATIO'] = df['RM'] * df['LSTAT'] * df['PTRATIO']
train_x = df[['RM', 'LSTAT', 'PTRATIO', 'RM2', 'RM_LSTAT', 'RM_LSTAT_PTRATIO']]
train_x = (train_x - train_x.mean()) / train_x.std()  #标准化
train_x = np.insert(train_x.to_numpy(), 0, 1, axis=1)
train_y = df['MEDV']

theta = np.random.rand(7)

for i in range(10000):
    theta = theta - learning_rate * np.dot(f(train_x, theta) - train_y, train_x)
    metrics = mse(train_x, train_y, theta)
    print('第%d轮 均方误差为%.04f theta%s' % (i+1, metrics, str(theta)))

#保存模型
with open('model.pickle', 'wb') as f:
    pickle.dump(theta, f)

#加载模型
with open('model.pickle', 'rb') as f:
    theta = pickle.load(f)

训练好的模型被保存到了名为'model.pickle'的文件中，并通过pickle库进行了序列化。在加载模型时，只需调用pickle库的load()函数即可将模型反序列化为一个Numpy数组。