import csv
import numpy as np
import matplotlib.pyplot as plt


def readData(filename):
    """
    讀取資料
    :param filename: csv格式資料集
    :return: X:list with shape[N,d],  X1:shape[8,2], X2:shape[9,2]
             y:list with shape[N,],   y1:shape[8,1],  y2:shape[9,1]
    """
    X1, X2, y1, y2 = [], [], [], []
    density, sugar = [], []
    with open(filename) as f:
        reader = csv.reader(f)
        head_row = next(reader)
        for line in reader:
            if line[9] == '是':
                X1.append([float(line[7]), float(line[8])])
                y1.append([float(line[10])])
                density.append(float(line[7]))
                sugar.append(float(line[8]))
            if line[9] == "否":
                X2.append([float(line[7]), float(line[8])])
                y2.append([float(line[10])])
                density.append(float(line[7]))
                sugar.append(float(line[8]))
    return X1, X2, y1, y2, density, sugar


def LDA(X1, X2):
    """
    線性判別分析
    :param X1: np.array with shape[8,2]
    :param X2: np.array with shape[9,2]
    :return: omega: np.array with shape[2,1], LDA最優化引數
    """
    mean1 = np.mean(X1, axis=0, keepdims=True)  # shape[1,d]
    mean2 = np.mean(X2, axis=0, keepdims=True)
    Sw = (X1 - mean1).T.dot(X1 - mean1) + (X2 - mean2).T.dot(X2 - mean2)  # shape[d,d]
    omega = np.linalg.inv(Sw).dot((mean1 - mean2).T)   # shape[d,1]
    return omega


if __name__ == '__main__':
    dataset = "C:\\Users\\14399\\Desktop\\西瓜3.0.csv"
    X1, X2, y1, y2, density, sugar = readData(dataset)
    # 視覺化
    plt.plot(density[:8], sugar[:8], 'r+')
    plt.plot(density[8:], sugar[8:], 'bo')

    # LDA
    X1 = np.array(X1)
    X2 = np.array(X2)
    y1 = np.array(y1)
    y2 = np.array(y2)
    omega = LDA(X1, X2)
    # 畫圖
    lda_left = 0
    lda_right = -(omega[0, 0] * 0.9) / omega[1, 0]
    plt.plot([0, 0.9], [lda_left, lda_right], 'g')

    plt.xlabel('density')
    plt.ylabel('sugar')
    plt.title('LDA')
    plt.show()
 

結果：

西瓜3.0資料集：連結：https://pan.baidu.com/s/1RXTUG9gP1Jn9HKFCiEzOlA         密碼：3h6n

參考資料：https://blog.csdn.net/victoriaw/article/details/77989610