（机器学习）岭回归 模型的保存与加载

岭回归

过拟合处理：正则化

• 将过拟合的曲线的凹凸幅度减少即可将过拟合曲线趋近于拟合曲线

• 正则化可以通过不断尝试发现高次项的特征，然后将这些特征的权重w调小到0，则高次项特征消失，凹凸幅度便减少，趋近于拟合曲线

L2正则化：

• 使用带有正则化算法的回归模型（Ridge岭回归）处理过拟合的问题

Ridge岭回归：具备L2正则化的线性回归模型

• API: from sklearn.linear_model import Ridge

• Ridge(alpha=1.0):

  • alpha:正则化力度，力度越大，则表示高次项的权重w越接近于0，导致过拟合曲线的凹凸幅度越小

    • 取值：0-1小数或1-10整数

  • coef_：回归系数

• 岭回归的优点：

  • 获取的回归系数更符合实际，更可靠

  • 在病态数据（异常值多的数据）偏多的研究中有更大的存在意义

• 使用

  from sklearn.preprocessing import PolynomialFeatures
  from sklearn.linear_model import LinearRegression
  from sklearn.linear_model import Ridge
  import numpy as np
  import matplotlib.pyplot as plt
  import joblib
  ​
  ​
  # 岭回归
  # 样本的训练数据，特征和目标值
  x_train = [[6], [8], [10], [14], [18]]  # 大小
  y_train = [[7], [9], [13], [17.5], [18]]  # 大小
  poly3 = PolynomialFeatures(degree=3)
  x_train_poly3 = poly3.fit_transform(x_train)
  #  建立模型预测
  regressor_poly3 = LinearRegression()
  regressor_poly3.fit(x_train_poly3, y_train)
  print(regressor_poly3.coef_)
  ​
  print("alpha = 0.1")
  # 使用岭回归可以通过控制正则化力度参数alpha降低高次项特征的权重
  regressor_poly3 = Ridge(alpha=0.1)
  regressor_poly3.fit(x_train_poly3, y_train)
  print(regressor_poly3.coef_)
  print("alpha = 0.3")
  regressor_poly3 = Ridge(alpha=0.3)
  regressor_poly3.fit(x_train_poly3, y_train)
  print(regressor_poly3.coef_)
  print("alpha = 0.5")
  regressor_poly3 = Ridge(alpha=0.5)
  regressor_poly3.fit(x_train_poly3, y_train)
  print(regressor_poly3.coef_)
  print("alpha = 0.7")
  regressor_poly3 = Ridge(alpha=0.7)
  regressor_poly3.fit(x_train_poly3, y_train)
  print(regressor_poly3.coef_)
  ​
  ​
  [[ 0.         -1.42626096  0.31320489 -0.01103344]]
  alpha = 0.1
  [[ 0.         -0.54302902  0.23512062 -0.00888958]]
  alpha = 0.3
  [[ 0.         -0.23508271  0.20785367 -0.00813998]]
  alpha = 0.5
  [[ 0.         -0.14579637  0.19991159 -0.00792083]]
  alpha = 0.7
  [[ 0.         -0.10329945  0.19610472 -0.00781518]]
  ​
  Process finished with exit code 0

模型的保存和加载

import joblib

• joblib.dump(knn, './123.m')

• knn = joblib.load('./123.m')

# 建立并保存模型
​
iris = datasets.load_iris()
feature = iris['data']
target = iris['target']
x_train, x_test, y_train, y_test = train_test_split(feature, target, test_size=0.2, random_state=2020)
knn = KNeighborsClassifier(n_neighbors=3)
knn = knn.fit(x_train, y_train)
joblib.dump(knn, './knn.m')
y_pred = knn.predict(x_test)
y_true = y_test
print("模型分类结果", y_pred)
print("真实分类结果", y_true)
# 输入测试特征与测试标签
score = knn.score(x_test, y_test)
print(score)
​
# 使用模型
iris = datasets.load_iris()
feature = iris['data']
target = iris['target']
x_train, x_test, y_train, y_test = train_test_split(feature, target, test_size=0.2, random_state=2020)
# 加载模型
knn = joblib.load('./knn.m')
y_pred = knn.predict(x_test)
y_true = y_test
print("模型分类结果", y_pred)
print("真实分类结果", y_true)
# 输入测试特征与测试标签
score = knn.score(x_test, y_test)
print(score)