PCA与LDA降维前后分类精度比较

PCA与LDA降维前后分类精度比较

1. 题目
用sklearn.datasets 里的load_digits（手写数字1797张），先对数据集进行可视化，接着对比并可视化PCA、LDA的降维后用KNN进行分类的效果。

2. 实验结果

3. 代码实现

from sklearn.datasets import load_digits
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
from sklearn.decomposition import PCA
from sklearn.neighbors import KNeighborsClassifier
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.metrics import accuracy_score
from sklearn.metrics import normalized_mutual_info_score
from sklearn.metrics import adjusted_rand_score
import math
import numpy as np
#聚类精度模板、三大指标模板
from sklearn import metrics
from scipy.optimize import linear_sum_assignment
def cluster_acc(y_true, y_pred):
    y_true = np.array(y_true).astype(np.int64)
    assert y_pred.size == y_true.size
    D = max(y_pred.max(), y_true.max()) + 1
    w = np.zeros((D, D), dtype=np.int64)
    for i in range(y_pred.size):
        w[y_pred[i], y_true[i]] += 1
    ind = linear_sum_assignment(w.max() - w)
    ind = np.asarray(ind)
    ind = np.transpose(ind)
    return sum([w[i, j] for i, j in ind]) * 1.0 / y_pred.size
def clusteringMetrics(trueLabel, predictiveLabel):
    # Clustering accuracy
    ACC = cluster_acc(trueLabel, predictiveLabel)
    # Normalized mutual information
    NMI = metrics.normalized_mutual_info_score(trueLabel, predictiveLabel)
    # Adjusted rand index
    ARI = metrics.adjusted_rand_score(trueLabel, predictiveLabel)
    return ACC, NMI, ARI

# 加载 `digits` 数据集
digits = load_digits()
X = digits.data
labels = digits.target

X_train, X_test, y_train, y_test = \
    train_test_split(X, labels, test_size=0.2, random_state=22)

# 设置图形大小(宽、高)以英寸为单位
fig = plt.figure(figsize=(6, 6))

# 设置子图形布局，如间隔之类...
fig.subplots_adjust(bottom=0,top=0.4,wspace =0, hspace =0)

fig.tight_layout()

# 对于64幅图像中的每一幅
for i in range(200):
    # 初始化子图:在10×20的网格中，在第i+1个位置添加一个子图
    ax = fig.add_subplot(10, 20, i + 1, xticks=[], yticks=[])
    # 在第i个位置显示图像
    ax.imshow(digits.images[i])



A_PCA = []
A_LDA = []
for i in range(1, 10):
    # PCA + KNN
    pca = PCA(n_components=i).fit(X_train)  # pca模型训练
    # 将输入数据投影到特征面正交基上
    X_train_pca = pca.transform(X_train)
    X_test_pca = pca.transform(X_test)
    knn = KNeighborsClassifier()
    knn.fit(X_train_pca, y_train)
    y_sample = knn.predict(X_test_pca)
    ACC_PCA= cluster_acc(y_test, y_sample)
    A_PCA.append(ACC_PCA)
    # LDA + KNN
    lda = LinearDiscriminantAnalysis(n_components=i).fit(X_train, y_train)  # lda模型训练
    #n_components：即我们进行LDA降维时降到的维数。在降维时需要输入这个参数。
    # 注意只能为[1,类别数-1)范围之间的整数。如果我们不是用于降维，则这个值可以用默认的None。
    # 将输入数据投影到特征面正交基上
    X_train_lda = lda.transform(X_train)
    X_test_lda = lda.transform(X_test)
    knn = KNeighborsClassifier()
    knn.fit(X_train_lda, y_train)
    y_sample = knn.predict(X_test_lda)
    ACC_LDA= cluster_acc(y_test, y_sample)
    #ACC_LDA, y,z=clusteringMetrics(y_test, y_sample)
    A_LDA.append(ACC_LDA)

# 画柱状图
fig, ax = plt.subplots()
bar_width = 0.35
opacity = 0.6  # 不透明度
index = np.arange(9)

ax.set_xticks(index + bar_width / 2)
cylinder1 = ax.bar(index, A_PCA, bar_width,\
                   alpha=opacity, color='b', label='PCA')

cylinder2 = ax.bar(index + bar_width, A_LDA, bar_width, \
                   alpha=opacity, color='g', label='LDA')


label = []  # 横坐标标签
for j in range(1, 10):
    label.append(j)
ax.set_xticklabels(label)
ax.legend()  # 图例标签
# 显示图形
plt.show()