5-1 无监督特征抽取

第5章 特征抽取

5.1 无监督特征抽取

请参考《数据准备和特征工程》中的相关章节，调试如下代码。

---

本节视频课程：无监督特征抽取

5.1.1 主成分分析

基础知识

from sklearn import datasets

iris = datasets.load_iris()
X = iris.data
X[: 4]

array([[5.1, 3.5, 1.4, 0.2],
       [4.9, 3. , 1.4, 0.2],
       [4.7, 3.2, 1.3, 0.2],
       [4.6, 3.1, 1.5, 0.2]])

from sklearn.decomposition import PCA
import numpy as np

# PCA算法可通过协方差实现，默认使用奇异值SVD分解
pca = PCA()    # 创建PCA(主成分分析)模型
X_pca = pca.fit_transform(X)    # 对X进行转换
np.round(X_pca[: 4], 2)    # 对所有数据取两位小数

array([[-2.68,  0.32, -0.03, -0.  ],
       [-2.71, -0.18, -0.21, -0.1 ],
       [-2.89, -0.14,  0.02, -0.02],
       [-2.75, -0.32,  0.03,  0.08]])

# 获取各个特征的可解释方差比，比例越高的特征越重要
pca.explained_variance_ratio_

array([0.92461872, 0.05306648, 0.01710261, 0.00521218])

# 保留两个特征
pca = PCA(n_components=2)    
X_pca = pca.fit_transform(X)
X_pca[: 4]

array([[-2.68412563,  0.31939725],
       [-2.71414169, -0.17700123],
       [-2.88899057, -0.14494943],
       [-2.74534286, -0.31829898]])

# 计算保留下来的两个特征的可解释方差比例之和
pca.explained_variance_ratio_.sum()

0.977685206318795

from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

# 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(X, iris.target,
                                                   test_size=0.3, 
                                                    random_state=0)

# 创建决策树分类模型clf，并通过训练好的模型来生成预测结果
clf = DecisionTreeClassifier()
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)

# 根据预测结果进行评分accuracy_score
accuracy = accuracy_score(y_test, y_pred)

X_train_pca, X_test_pca, y_train_pca, y_test_pca = train_test_split(
                                                X_pca, iris.target,
                                                test_size=0.3, 
                                                random_state=0)
clf2 = DecisionTreeClassifier()
clf2.fit(X_train_pca, y_train_pca)
y_pred_pca = clf2.predict(X_test_pca)
accuracy2 = accuracy_score(y_test_pca, y_pred_pca)

print("含4个特征的数据集的预测准确率: ", np.round(accuracy, 2))
print("含2个特征的数据集的预测准确率: ", np.round(accuracy2, 2))

含4个特征的数据集的预测准确率:  0.98
含2个特征的数据集的预测准确率:  0.91

项目案例

from scipy.io import loadmat

mnist = loadmat("data/data20537/mnist-original.mat")
mnist

{'__header__': b'MATLAB 5.0 MAT-file Platform: posix, Created on: Sun Mar 30 03:19:02 2014',
 '__version__': '1.0',
 '__globals__': [],
 'mldata_descr_ordering': array([[array(['label'], dtype='<U5'), array(['data'], dtype='<U4')]],
       dtype=object),
 'data': array([[0, 0, 0, ..., 0, 0, 0],
        [0, 0, 0, ..., 0, 0, 0],
        [0, 0, 0, ..., 0, 0, 0],
        ...,
        [0, 0, 0, ..., 0, 0, 0],
        [0, 0, 0, ..., 0, 0, 0],
        [0, 0, 0, ..., 0, 0, 0]], dtype=uint8),
 'label': array([[0., 0., 0., ..., 9., 9., 9.]])}

mnist_data = mnist["data"].T
mnist_label = mnist["label"][0]
mnist_data.shape

(70000, 784)

# 保留可解释方差比例大于0.95的特征
pca = PCA(.95)    
lower_dimensional_data = pca.fit_transform(mnist_data)
lower_dimensional_data.shape

(70000, 154)

# 复制文件到matplotlib字体路径
!cp simhei.ttf /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/matplotlib/mpl-data/fonts/ttf/
# 删除matplotlib的缓冲目录
!rm -rf .cache/matplotlib
# 重启内核

%matplotlib inline
import matplotlib.pyplot as plt

# 设置显示中文
plt.rcParams['font.sans-serif'] = ['SimHei'] # 指定默认字体
plt.rcParams['axes.unicode_minus'] = False   # 解决保存图像是负号'-'显示为方块的问题}

plt.figure(figsize=(8,4));

# 原图，plt.imshow：把数值展示成热图
#gray：黑-白   interpolation：显示图像的插值方法。
plt.subplot(1, 2, 1);
plt.imshow(mnist_data[1].reshape(28,28),
              cmap = plt.cm.gray,   
              interpolation='nearest', #
              clim=(0, 255));
plt.xlabel('784 components', fontsize = 14)
plt.title('原图', fontsize = 20);

# 154 principal components
iverse_data = pca.inverse_transform(lower_dimensional_data)
plt.subplot(1, 2, 2);
plt.imshow(iverse_data[1].reshape(28, 28),
              cmap = plt.cm.gray, 
              interpolation='nearest',
              clim=(0, 255));
plt.xlabel('154 components', fontsize = 14)
plt.title('特征抽取后的图', fontsize = 20)

Text(0.5,1,'特征抽取后的图')

import time
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
from sklearn.decomposition import PCA
from sklearn.metrics import accuracy_score
from scipy.io import loadmat
import pandas as pd

mnist = loadmat("data/data20537/mnist-original.mat")
mnist_data = mnist["data"].T
mnist_label = mnist["label"][0]

# 划分训练集和测试集，并对训练集进行特征标准化
train_img, test_img, train_lbl, test_lbl = train_test_split(
                                            mnist_data, 
                                            mnist_label, 
                                            test_size=1/7.0, 
                                            random_state=0)
scaler = StandardScaler()
scaler.fit(train_img)
train_img = scaler.transform(train_img)
test_img = scaler.transform(test_img)

# logistic_reg：对数几率回归，exp_var为保留比例
def logistic_reg(exp_var): 
    pca = PCA(exp_var)
    pca.fit(train_img)

    # 有监督特征提取，计算两个标签集之间的差异
    lr = LogisticRegression(solver = 'lbfgs')
    lr.fit(pca.transform(train_img), train_lbl)
    lbl_pred = lr.predict(pca.transform(test_img))
    acc = accuracy_score(test_lbl, lbl_pred)

    # 返回特征数量，和评分结果 
    return pca.n_components_, acc

v, n, a, t = [], [], [], []
for i in [None, 0.99, 0.95, 0.90, 0.85]:
    start = time.time()
    components, accuracy = logistic_reg(i)
    stop = time.time()
    deltat = stop - start
    v.append(i)
    n.append(components)
    a.append(accuracy)
    t.append(deltat)

df = pd.DataFrame({"Var_ratio":v, 
                   "N_components":n, 
                   "Accuracy": a,
                   "Times": t})
df

	Var_ratio	N_components	Accuracy	Times
0	NaN	784	0.9164	67.382504
1	0.99	541	0.9198	46.709717
2	0.95	330	0.9215	41.099790
3	0.90	236	0.9216	40.692168
4	0.85	184	0.9204	41.502007

动手练习

import pandas as pd

# 练习通过URL获得数据的方法
url = "https://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data"    

# 从url加载数据集
df = pd.read_csv(url, names=['sepal length','sepal width','petal length','petal width','target'])

from sklearn.preprocessing import StandardScaler

features = ['sepal length', 'sepal width', 'petal length', 'petal width']

# 划分特征集x和标签集y，并对x进行标准化
x = df.loc[:, features].values
y = df.loc[:,['target']].values
x = StandardScaler().fit_transform(x)

from sklearn.decomposition import PCA

# 对x进行特征抽取，得到最后只有2个主成分的集合principalDf
pca = PCA(n_components=2)
principalComponents = pca.fit_transform(x)
principalDf = pd.DataFrame(data = principalComponents
             , columns = ['主成分1', '主成分2'])

finalDf = pd.concat([principalDf, df[['target']]], axis = 1)

# ax.scatter: 绘制散点图
fig = plt.figure(figsize = (8,8))
ax = fig.add_subplot(1,1,1) 
ax.set_xlabel('主成分1', fontsize = 15)
ax.set_ylabel('主成分2', fontsize = 15)
ax.set_title('PCA的2个主成分', fontsize = 20)

targets = ['Iris-setosa', 'Iris-versicolor', 'Iris-virginica']
colors = ['r', 'g', 'b']
for target, color in zip(targets,colors):

    # 定位到要保留的索引：indicesToKeep
    indicesToKeep = finalDf['target'] == target
    ax.scatter(finalDf.loc[indicesToKeep, '主成分1']
               , finalDf.loc[indicesToKeep, '主成分2']
               , c = color
               , s = 50)
ax.legend(targets)
ax.grid()

pca.explained_variance_ratio_

array([0.72770452, 0.23030523])

5.1.2 因子分析

基础知识

from sklearn.decomposition import FactorAnalysis

# FactorAnalysis: 因子分析
fa = FactorAnalysis()    
iris_fa = fa.fit(iris.data)

# 展示了训练后的各个数据与潜在因子的方差
fa.components_    

array([[ 0.70698856, -0.15800499,  1.65423609,  0.70084996],
       [ 0.115161  ,  0.15963548, -0.04432109, -0.01403039],
       [-0.        ,  0.        ,  0.        ,  0.        ],
       [-0.        ,  0.        ,  0.        , -0.        ]])

# 设置特征抽取的数量为2
fa = FactorAnalysis(n_components=2)
iris_two = fa.fit_transform(iris.data)
iris_two[: 4]

array([[-1.32761727, -0.56131076],
       [-1.33763854, -0.00279765],
       [-1.40281483,  0.30634949],
       [-1.30104274,  0.71882683]])

%matplotlib inline
import matplotlib.pyplot as plt

# 绘制FA的散点图
f = plt.figure(figsize=(5, 5))
ax = f.add_subplot(111)
ax.scatter(iris_two[:,0], iris_two[:, 1], c=iris.target)
ax.set_title("包含2个因子的FA")

Text(0.5,1,'包含2个因子的FA')

# 绘制PCA的散点图
f = plt.figure(figsize=(5, 5))
ax = f.add_subplot(111)
ax.scatter(X_pca[:,0], X_pca[:, 1], c=iris.target)
ax.set_title("包含2个主成分的PCA")

Text(0.5,1,'包含2个主成分的PCA')

项目案例

import pandas as pd

df = pd.read_csv("/home/aistudio/data/data20537/bfi.csv")
df.columns

Index(['Unnamed: 0', 'A1', 'A2', 'A3', 'A4', 'A5', 'C1', 'C2', 'C3', 'C4',
       'C5', 'E1', 'E2', 'E3', 'E4', 'E5', 'N1', 'N2', 'N3', 'N4', 'N5', 'O1',
       'O2', 'O3', 'O4', 'O5', 'gender', 'education', 'age'],
      dtype='object')

# 删除无关数据
df.drop(['Unnamed: 0', 'gender', 'education', 'age'],axis=1,inplace=True)
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 2800 entries, 0 to 2799
Data columns (total 25 columns):
 #   Column  Non-Null Count  Dtype  
---  ------  --------------  -----  
 0   A1      2784 non-null   float64
 1   A2      2773 non-null   float64
 2   A3      2774 non-null   float64
 3   A4      2781 non-null   float64
 4   A5      2784 non-null   float64
 5   C1      2779 non-null   float64
 6   C2      2776 non-null   float64
 7   C3      2780 non-null   float64
 8   C4      2774 non-null   float64
 9   C5      2784 non-null   float64
 10  E1      2777 non-null   float64
 11  E2      2784 non-null   float64
 12  E3      2775 non-null   float64
 13  E4      2791 non-null   float64
 14  E5      2779 non-null   float64
 15  N1      2778 non-null   float64
 16  N2      2779 non-null   float64
 17  N3      2789 non-null   float64
 18  N4      2764 non-null   float64
 19  N5      2771 non-null   float64
 20  O1      2778 non-null   float64
 21  O2      2800 non-null   int64  
 22  O3      2772 non-null   float64
 23  O4      2786 non-null   float64
 24  O5      2780 non-null   float64
dtypes: float64(24), int64(1)
memory usage: 547.0 KB

df.dropna(inplace=True)

# !mkdir /home/aistudio/external-libraries
# !pip install -i https://pypi.tuna.tsinghua.edu.cn/simple factor-analyzer -t /home/aistudio/external-libraries

import sys
sys.path.append('/home/aistudio/external-libraries')

from factor_analyzer.factor_analyzer import calculate_bartlett_sphericity

# bartlett球形检验，p_value为0说明拒绝原假设，可以进行因子分析
# 特征相关性检验的方法还有很多，如：KMO检验，目的是为了确认各个特征的背后是否存在某种联系
chi_square_value,p_value=calculate_bartlett_sphericity(df)
chi_square_value, p_value

(18170.9663508689, 0.0)

from factor_analyzer.factor_analyzer import calculate_kmo

# KMO检验，结果值在0~1之间，越接近于1，变量间的相关性越强，因子分析的效果越好。
kmo_all,kmo_model=calculate_kmo(df)
kmo_model

0.8485397221949231

from factor_analyzer import FactorAnalyzer

# 创建因子分析模型：fa,训练抽取其中的25个特征
# rotation:因子旋转的方法,None就是不进行因子旋转，'varimax’为方差最大旋转法
fa = FactorAnalyzer(rotation=None)     
fa.fit(df, 25)    

# 获取这25个特征的本征值，计算因子相关矩阵ev和特征值v
ev, v = fa.get_eigenvalues()    
ev

array([5.13431118, 2.75188667, 2.14270195, 1.85232761, 1.54816285,
       1.07358247, 0.83953893, 0.79920618, 0.71898919, 0.68808879,
       0.67637336, 0.65179984, 0.62325295, 0.59656284, 0.56309083,
       0.54330533, 0.51451752, 0.49450315, 0.48263952, 0.448921  ,
       0.42336611, 0.40067145, 0.38780448, 0.38185679, 0.26253902])

# 绘制各特征因子与本征值的分布图
plt.scatter(range(1,df.shape[1]+1),ev)
plt.plot(range(1,df.shape[1]+1),ev)
plt.title('特征因子与本征值的分布图')
plt.xlabel('因子')
plt.ylabel('本征值')
plt.grid()

动手练习

# rotation="varimax"：因子旋转的方法为：方差最大旋转法
fa = FactorAnalyzer(rotation="varimax", n_factors=5)
fa.fit(df)
fa.transform(df)

array([[-0.43983041,  0.10389656, -1.21671279, -0.69621532, -1.51944901],
       [ 0.08436922,  0.55931085, -0.60152092, -0.18934143, -0.24838425],
       [ 0.5205342 ,  0.3182346 ,  0.02564572, -0.75113554,  0.20450298],
       ...,
       [-0.22326681,  0.72116924,  0.78710981, -1.05847976,  0.92510304],
       [ 0.92052049,  0.7493374 ,  0.07095309, -2.13514505,  0.67346015],
       [-1.68998315, -1.35240274, -0.08143349, -1.85488924, -0.16387988]])