多元线性回归模型的等值线图

Question

我必须获得等高线图才能使用以下变量获得一系列最佳值：

X axis = SiO2/Al2O3
Y axis = Precursor/Aggregate
Z axis = Compressive Strength

我的代码如下

import numpy as np
import matplotlib as mlt
import matplotlib.pyplot as plt
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression
from sklearn.metrics import r2_score

dataset = pd.read_csv('Data.csv')
X = dataset.iloc[:, :-1].values
y = dataset.iloc[:, -1].values

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)

regressor = LinearRegression()
regressor.fit(X_train, y_train)
y_predict = regressor.predict(X_test)

feature_x = X_test[:, 1]
feature_y = X_test[:, 3]

[X, Y] = np.meshgrid(feature_x, feature_y)  
Z = y_predict    

ax.contourf(X, Y, Z)  
ax.set_title('Filled Contour Plot')
ax.set_xlabel('SiO2/Al2O3')
ax.set_ylabel('Precursor/Aggregate')
plt.show()

但它给出了这个错误

TypeError: Input z must be 2D, not 1D

我想我在 Z 轴输入上出错了。

数据可用at this link。

预期输出：

Answer 1

您的代码不起作用，您需要为您的预测变量值创建一个网格，首先我们读入您的数据并进行拟合：

dataset = pd.read_csv('Data.csv')
X = dataset.iloc[:, :-1].values
y = dataset.iloc[:, -1].values

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)

regressor = LinearRegression()
regressor.fit(X_train, y_train)

然后您需要为您感兴趣的特征创建一个网格：

feature_x = np.linspace(X_test[:, 1].min(),X_test[:, 1].max(),100)
feature_y = np.linspace(X_test[:, 3].min(),X_test[:, 3].max(),100)

网格：

dim1, dim2 = np.meshgrid(feature_x, feature_y)

现在你的模型有 6 个你需要提供的其他预测变量来拟合。一种方法是将这些其他变量保持在它们的平均值，然后我们在网格中插入：

mesh_df = np.array([X_test.mean(axis=0) for i in range(dim1.size)])
mesh_df[:,1] = dim1.ravel()
mesh_df[:,2] = dim2.ravel()

现在预测、重塑和绘图：

Z = regressor.predict(mesh_df).reshape(dim1.shape)  

fig, ax = plt.subplots()

ax.contourf(dim1, dim2, Z)  
ax.set_title('Filled Contour Plot')
ax.set_xlabel('SiO2/Al2O3')
ax.set_ylabel('Precursor/Aggregate')
plt.show()

看起来像这样，因为您使用的是线性回归，值将随您的变量线性增加或减少：

多元线性回归模型的等值线图

Contour plot for multi linear regression model

python

pandas

matplotlib

scikit-learn

regression