Unsupervised Learning Model-Reducing Dimension

Unsupervised Learning Model-Reducing Dimension

Author: Xie Zhong-zhao

1. PCA

import numpy as np

#初始化一个2*2的线性相关矩阵
M = np.array([[1,2],[2,4]])
#计算2*2线性相关矩阵的秩
print(np.linalg.matrix_rank(M,tol=None))

import pandas as pd

#使用pandas分别读取训练数据和测试数据集
digit_train = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/optdigits/optdigits.tra',header=None)
digit_test = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/optdigits/optdigits.tes',header=None)

#分割训练数据的特征向量和标记
X_digits = digit_train[np.arange(64)]
y_digits = digit_train[64]

# print(X_digits)
# print(y_digits)

'''
从sklearn.decomposition导入PCA
'''
from sklearn.decomposition import PCA

#初始化一个可以将高维特征向量64维压缩到二个维度的PCA
estimator = PCA(n_components=2)
X_pca = estimator.fit_transform(X_digits)
print(X_pca)

print(X_pca[:,0])
print([y_digits.as_matrix() == 1])
print(X_pca[:,0][y_digits.as_matrix() == 1])
'''
显示10类手写数字图片经PCA压缩后的2维空间分布
'''
from matplotlib import pyplot as plt

def plot_pca_scatter():
    colors = ['black','blue','purple','yellow','white','red','lime','cyan','orange','gray']
    for i in range(len(colors)):
        px = X_pca[:,0][y_digits.as_matrix() == i]
        py = X_pca[:,1][y_digits.as_matrix() == i]
        plt.scatter(px,py,c=colors[i])

    plt.legend(np.arange(0,10).astype(str))
    plt.xlabel('First Principal Component')
    plt.ylabel('Second Principal Component')
    plt.show()

plot_pca_scatter()

1
[[-12.44580352   4.71301334]
 [-16.52024584  13.07467011]
 [ 18.84671433 -10.71445017]
 ..., 
 [-24.11867087  -5.78704883]
 [-29.35115567   3.41363267]
 [ 24.993826   -11.79116904]]
[-12.44580352 -16.52024584  18.84671433 ..., -24.11867087 -29.35115567
  24.993826  ]
[array([False, False, False, ..., False, False, False], dtype=bool)]
[  1.03802829e+01  -1.04379656e+01   1.43618578e+01   1.16339451e+01
  -3.30200846e+00   1.22332485e+01   9.20407222e+00   8.30492835e+00
   8.56969878e+00   4.23669716e+00   6.60284625e+00   9.01959870e+00
   1.15801627e+01   8.49019062e+00  -3.78521322e-01  -1.68102735e+01
  -2.19399331e+00   1.40600870e+01   4.16898313e+00   2.94525323e+00
   8.72491087e+00   6.37000716e+00   1.03965755e+01   1.06175330e+01
   9.78565432e+00   7.63470042e+00   1.06497944e+01   8.14431578e+00
   1.20032089e+01   8.37815835e+00  -9.58612285e-01   6.57085269e+00
   6.34220091e-01   1.72254106e+00  -6.77930465e-01  -8.46183679e-01
  -1.16357548e+00   3.52554139e+00  -8.76401983e+00  -1.92717564e+00
  -6.87795966e+00   4.47138806e+00   6.03851916e-01   9.50989460e+00
   4.68779834e+00   3.41906322e+00   1.31339429e+01  -5.67266416e+00
  -2.58530877e+00  -6.35470508e+00  -5.23340480e+00  -2.33391071e+00
  -1.11270494e+01  -2.21179220e-01   9.47692668e+00   5.01763456e+00
   1.39901481e+01   7.96383394e+00   5.79810644e+00   1.14577890e+01
   1.02713550e+01  -6.87254041e+00  -1.16995102e+01  -9.87992841e+00
  -1.33394598e+01  -1.52933686e+01  -1.13868268e+01  -1.60171997e+01
  -6.35680734e+00  -1.54948588e+01   2.49034817e+00   3.64840049e+00
  -2.27528666e+00  -6.44265684e+00   6.31203415e+00   6.08488708e+00
   5.12631116e+00   4.94814660e+00   1.56022297e+00   5.86349685e+00
   7.92933353e+00   5.01878329e+00   5.93952581e+00   4.51645036e-01
   3.91895540e+00   1.27727527e+01   1.85489607e+01   1.20872407e+01
   1.02281540e+01   1.75347348e+01   1.94804961e+01  -5.40113023e+00
   2.40162635e-01  -3.52528854e-01   1.05018301e+01   2.39317011e+00
   3.77365620e+00   1.11763928e+01   1.40209995e+01   1.46306065e+01
   1.29994591e+01   1.14743177e+01  -1.71281797e+01   7.23911978e+00
   7.89591393e+00   7.65909505e+00   6.45541038e+00   8.70473702e+00
   1.24854320e+01  -4.21198965e-01   1.28954391e+01   2.50478653e+00
   1.79305349e+00  -1.22541996e+01   4.19760484e-01  -6.20773102e+00
   1.05994014e+00   1.23992384e+01   9.76004438e-01   6.05600248e+00
   5.52587831e-01  -4.80961444e+00  -2.63961568e+00   3.77742885e+00
   2.34466989e+00  -6.82109440e+00  -9.38219051e-01   6.98226771e+00
   2.19819315e+00  -5.14118720e+00  -2.21269059e-01   6.87760329e+00
  -5.07990529e+00   6.07700374e+00   1.51443865e+00  -4.12570503e+00
   1.17420745e+01   2.68636054e+00   8.43724576e-01   9.04821693e+00
   5.69062465e+00   5.09391771e+00   2.38727241e-02   1.30291002e-01
  -2.21696494e+00   1.62061822e+00   8.56541019e-01  -1.04122429e+01
   5.31662513e+00   7.84844534e+00  -2.53040835e+00  -1.29182785e+01
  -1.03348326e+01   6.82147467e-01  -5.43888083e+00  -1.63331640e+01
  -1.00525923e+01   1.28068930e+01   1.33231200e+01   1.13147850e+01
   8.91862856e+00   1.34008824e+00  -7.70197618e+00   1.28705348e+01
  -4.80093992e+00  -1.06846372e+01   1.64133374e+00  -4.94311753e+00
   1.06232067e+01   1.71831499e+00   6.82508338e+00   1.34554586e+01
   9.39502862e+00   1.33799335e+01   8.08492152e+00   1.49620726e+01
   7.17581227e+00   9.34415664e+00   9.33960828e+00   7.86356836e+00
   2.14118801e+00  -2.13592073e+00   1.11653492e+01  -1.04491366e+00
  -3.94541357e+00   4.56132149e+00   2.05559793e+00   1.61866475e+01
   4.37683236e+00  -9.62052240e-02  -1.32389449e+00   1.27283361e+01
   9.29013078e+00   1.41483410e+01   4.51357809e+00   2.59420973e+00
   8.68599402e+00  -4.13381693e+00  -7.02643152e+00   6.71717387e+00
  -6.84623064e+00   7.91850548e+00   4.08618847e+00   6.93902558e+00
   6.33680507e+00   6.93103762e+00  -4.92035361e+00   1.07324064e+01
   1.24797655e+01   1.30762501e+01   9.40172112e+00   1.35612664e+01
  -3.93549392e+00   1.32420591e+01   4.73654349e+00   1.95419045e+00
   4.41819042e+00   9.07426636e+00  -4.46064879e+00   7.36219691e+00
   5.31371422e+00  -1.01825385e+01  -1.35445035e+01   1.15390490e+01
  -1.04932638e+01   1.07819865e+01   8.77818200e+00   5.04039534e+00
   1.36233502e+01  -1.06959043e+01  -1.20470341e+01  -6.57222515e+00
   5.13528176e+00  -6.77706748e+00  -8.89440513e+00  -5.80165564e+00
  -1.87681654e+00   4.56814347e+00   9.99379256e+00   3.15301675e-01
   5.18439757e+00   3.71502472e+00   9.15767108e+00   8.84303133e+00
  -6.96951400e-01   8.49256956e+00   1.01068733e+01   1.04815898e+01
  -6.15395530e+00   1.42880924e+01   7.61468791e+00  -6.26532565e-02
  -1.09296969e+00   3.37597493e+00   1.87111720e+00   1.16395328e+01
   7.51366451e+00   3.78312476e+00   1.33321724e+01   5.62117699e+00
  -9.75671342e+00  -1.88249908e+00   3.53846998e+00   4.41674517e+00
   3.11406313e-01   6.77850208e+00   4.84468461e+00   9.93418932e+00
   5.37455218e+00   1.16300889e+01   7.76200432e-01  -7.16517836e+00
   7.30956796e+00  -4.03537992e+00   1.21840926e+01   1.34320669e+01
   1.12880424e+01   7.34254268e+00  -8.53760892e+00   9.96101519e-01
   1.11955136e+01   1.16191430e+01   8.51748445e+00   1.10262393e+01
   2.67367744e+01   5.69619180e+00   5.99652302e+00   1.73695447e+01
  -7.68083524e+00   1.32947737e+01  -1.19518330e+00   1.09855694e+01
  -3.93538833e+00   1.04744223e+01   1.08595847e+01   1.25569475e+01
  -1.01545515e+01  -1.39919828e+01   5.43785465e+00   1.01277835e+01
   1.18250355e+01  -5.53136455e+00   5.74723755e-02   8.65924601e-02
   1.77912205e+00   3.83230801e+00   1.16182396e+01   6.44742437e+00
   9.69097559e+00  -1.02640487e+01   8.21685873e-01  -8.25560043e+00
  -6.92170755e+00  -5.61494893e+00  -4.36365563e-01   7.32723893e+00
  -6.56976697e+00  -1.30178755e+01   1.08404587e+01   3.45203125e+00
  -4.68373413e+00   6.36076304e+00   2.92399045e+00   8.78764678e+00
   2.16693344e+00   6.80021418e+00  -7.02931956e+00   1.27833562e+01
   2.23739799e+01   6.26276605e+00   9.66522738e+00   3.38021206e+00
   1.06152100e+01   7.41959756e+00   5.79736793e+00   9.06006772e+00
   5.43146505e+00  -5.39811599e+00   9.48236309e+00   9.03057522e+00
   9.67984449e+00   1.20660153e+01   8.14266430e+00   1.01896935e+01
   1.18091326e+01   3.91897983e+00   1.05349565e+01   1.44086876e+01
  -8.71950416e+00   4.18190750e+00  -8.67530064e-01   7.93708914e+00
   8.70317666e+00   5.87257428e+00   7.08372547e+00   8.58931682e+00
   1.25187431e+01   8.66464885e+00   7.36489793e+00  -1.27530341e+01
  -8.35661199e+00   1.30629911e+01   1.08040910e+01   8.77241417e+00
   9.50381783e+00   1.24992913e+01  -8.31521021e+00   2.00673606e+00
   4.58585526e+00   3.16349264e+00  -1.10608458e+00   2.95561319e+00
   7.34034207e+00   8.11673502e+00   7.27750323e+00   1.57382437e+01
   1.06340794e+01   1.15884364e+01   9.62182749e+00  -1.13317133e+01
  -7.87394324e+00  -1.97542950e+00  -2.80337867e+00  -5.70773108e+00
   1.97880168e+01   8.16372094e+00   1.15713887e+01   8.96707599e+00
   1.54352343e+01]

import pandas as pd
import numpy as np
#使用pandas分别读取训练数据和测试数据集
digit_train = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/optdigits/optdigits.tra',header=None)
digit_test = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/optdigits/optdigits.tes',header=None)

#分割训练数据的特征向量和标记
X_train = digit_train[np.arange(64)]
y_train = digit_train[64]

X_test = digit_test[np.arange(64)]
y_test = digit_test[64]

'''
导入基于线性核的支持向量机分类
'''
from sklearn.svm import LinearSVC
from sklearn.decomposition import PCA
#使用默认配置初始化LinearSVC,对原始64像素特征的训练数据进行建模，并在测试数据上做出预测，存储在y_predict中
svc = LinearSVC()
svc.fit(X_train,y_train)
y_predict = svc.predict(X_test)

#使用PCA将64维图像数据压缩到20维
estimator = PCA(n_components=20)

#利用训练特征决定（fit）20个正交维度方向，并转化（transform)原训练特征
pca_X_train = estimator.fit_transform(X_train)
#测试特征也按照上述20个正交维度方向进行转化（transform）
pca_X_test = estimator.transform(X_test)

#使用默认初始化配置LinearSVC,对压缩后的20位特征的训练数据进行建模，并在测试数据上做出预测，存储在pca_y_predict中
pca_svc = LinearSVC()
pca_svc.fit(pca_X_train,y_train)
pca_y_predict = pca_svc.predict(pca_X_test)

'''
原始像素特征与PCA压缩重建的低维特征，在相同配置的支持向量机（分类）模型上识别性能的差异
'''
#从sklearn.metrics导入classification_report用于更加细致的分类性能分析
from sklearn.metrics import classification_report

#对原始图像高维像素特征训练的支持向量机分类器的性能做出评估
print(svc.score(X_test,y_test))
print(classification_report(y_test,y_predict,target_names=list(np.arange(10).astype(str))))

#使用PCA压缩重建的低维图像特征训练的支持向量机分类器的性能进行评估
print(pca_svc.score(pca_X_test,y_test))
print(classification_report(y_test,pca_y_predict,target_names=list(np.arange(10).astype(str))))

0.920979410128
             precision    recall  f1-score   support

          0       0.99      0.98      0.99       178
          1       0.89      0.89      0.89       182
          2       0.99      0.96      0.97       177
          3       0.99      0.84      0.91       183
          4       0.95      0.97      0.96       181
          5       0.89      0.96      0.92       182
          6       0.99      0.96      0.98       181
          7       0.99      0.90      0.94       179
          8       0.72      0.91      0.80       174
          9       0.87      0.84      0.86       180

avg / total       0.93      0.92      0.92      1797

0.927657206455
             precision    recall  f1-score   support

          0       0.96      0.96      0.96       178
          1       0.93      0.85      0.89       182
          2       0.97      0.95      0.96       177
          3       0.96      0.91      0.93       183
          4       0.94      0.97      0.95       181
          5       0.90      0.97      0.93       182
          6       0.97      0.98      0.98       181
          7       0.98      0.92      0.95       179
          8       0.89      0.86      0.87       174
          9       0.81      0.91      0.86       180

avg / total       0.93      0.93      0.93      1797