PCA图 (主成分 biplot & 三维图) 绘制与解释 | python

问题：如何绘制二维和三维的主成分图，以及如何解释？

Code: 二维图

import matplotlib.pyplot as plt
import numpy as np

def biplot(reduced_data, labels, pc, variable):
    """plot compositional biplot for two principle components
    
    :param reduced_data: data processed by PCA
    :param labels: labels of original dataset 
    :param pc: all the principle components
    :param variable: the name of the variables of the data set
    """
    plt.figure(1, figsize=(14, 10))
    
    legend = []  # 
    classes = np.unique(labels)  # label type
    n = pc.shape[1]
    # colors = ['g', 'r', 'y']
    # markers = ['o', '^', 'x']
    
    x = reduced_data[:, 0]  # variable contributions for PC1
    y = reduced_data[:, 1]  # variable contributions for PC2
    scalex = 1.0/(x.max() - x.min())
    scaley = 1.0/(y.max() - y.min())
    
    # Draw a data point projection plot that is projected to 
    # a two-dimensional plane using normal PCA
    for i, label in enumerate(classes):
        plt.scatter(x[labels==label] * scalex,
                    y[labels==label] * scaley,
                    linewidth=0.01) 
        # hyperparameter in plt.scatter(): c=colors[i], marker=markers[i]
        legend.append("Label: {}".format(label))
     
    plt.legend(legend)
    
    # plot arrows as the variable contribution,
    # each variable has a score for PC1 and for PC2 respectively
    for i in range(n):
        plt.arrow(0, 0, pc[0, i], pc[1, i], color='k', alpha=0.7,
                  linewidth=1, )
        plt.text(pc[0, i]*1.01, pc[1, i]*1.01, variable[i],
                 ha='center', va='center', color='k', fontsize=12)
        
    plt.xlabel("$PC1$")
    plt.ylabel("$PC2$")
    plt.title("Compositional biplot")
    plt.grid()
    save_fig("Compositional biplot")

解释：

1. 能够清楚的看到哪些特征对主成分（PC）的贡献值大，一般来说，是分别分析哪些特征对PC1较为重要，哪些特征对PC2较为重要；
2. 对于PC1来说，元素的横坐标的绝对值越大，越重要
3. 对于PC2来说，元素的纵坐标的绝对值越大，越重要

Code：三维图

import matplotlib.pyplot as plt
import numpy as np

def componential_plot_static_3d(reduced_data, labels, pc, variable):
    """draw a static componential plot in 3d for three principle components
    
    :param reduced_data: data processed by PCA
    :param labels: labels of original dataset 
    :param pc: all the principle components
    :param variable: the name of the variables of the data set
    """
    
    fig = plt.figure(1, figsize=(14, 12))
    ax = plt.axes(projection='3d')
    
    legend = []  # 
    classes = np.unique(labels)  # label type
    n = pc.shape[1]
    # colors = ['g', 'r', 'y']
    # markers = ['o', '^', 'x']
    
    x = reduced_data[:, 0]  # variable contributions for PC1
    y = reduced_data[:, 1]  # variable contributions for PC2 
    z = reduced_data[:, 2]  # variable contributions for PC3
    scalex = 1.0/(x.max() - x.min())
    scaley = 1.0/(y.max() - y.min())
    scalez = 1.0/(z.max() - z.min())
    
    # Draw a data point projection plot that is projected to 
    # a three-dimensional space using normal PCA
    for i, label in enumerate(classes):
        ax.scatter3D(x[labels==label] * scalex,
                     y[labels==label] * scaley,
                     z[labels==label] * scalez,
                     linewidth=0.01) 
        # hyperparameter in plt.scatter(): c=colors[i], marker=markers[i]
        legend.append("Label: {}".format(label))
    
    ax.legend(legend)
    
    # the initial angle to draw the 3d plot
    azim = -60  # azimuth
    elev = 30   # elevation
    ax.view_init(elev, azim)  # set the angles
    
    # plot arrows as the variable contribution,
    # each variable has a score for PC1, for PC2 and for PC3 respectively
    for i in range(n):
        ax.quiver(0, 0, 0, pc[0, i], pc[1, i], pc[2, i], color='k', alpha=0.7,
                  linewidth=1, arrow_length_ratio=0.05)
        ax.text(pc[0, i]*1.1, pc[1, i]*1.1, pc[2, i]*1.1, variable[i],
                ha='center', va='center', color='k', fontsize=12)
        
    ax.set_xlabel("$PC1$")
    ax.set_ylabel("$PC2$")
    ax.set_zlabel("$PC3$")
    plt.title("Componential Plot in 3 Dimension")
    plt.grid()
    save_fig("Componential_Plot_in_3_Dimension")

• elevation 仰角；azimuth方位角，ax.view_init(elev, azim)设置视角
• 当elevation=0时，视角为沿x1负方向看，当elevation=90时，视角沿x3负方向看。
• 当azimuth=0时，视角为沿x1负方向看，当azimuth=90时，视角沿x2负方向看。
• 随着azimuth的增加，从x3负方向看，x1x2平面是顺时针旋转的。
• 逆时针旋转，能把x1,x2的大小顺序调整为常规平面坐标系。

进一步学习【如何绘制三维PCA动图】

参考资料：
PCA clearly explained —When, Why, How to use it and feature importance: A guide in Python