機械学習:scikit-learn笑顔認識(SVM,KNN,Logisitc regression)

scikit-learnはPythonの非常に強力な機械学習のパッケージで、今日scikit-learnの中でいくつかのよく使われる分類器SVM、KNNとlogistic regressionを紹介して、笑顔の識別に使います.
ここではGENKI 4 Kというデータベースを用いており,各画像はまず顔検出とカットを行い,HOGの特徴を抽出する.このデータベースには4000枚の図があり、4つのグループに分けてcross validationを作り、平均値を最終的な識別率とします.

import string, os, sys
import numpy as np
import matplotlib.pyplot as plt
import scipy.io
import random
from sklearn import neighbors, linear_model, svm


dir = '/GENKI4K/Feature_Data'  
print '----------- no sub dir'  

# prepare the data
files = os.listdir(dir)  
for f in files:  
    print dir + os.sep + f

file_path=dir+os.sep+files[14]

#print file_path

dic_mat = scipy.io.loadmat(file_path)

data_mat=dic_mat['Hog_Feat']

print 'feature: ',  data_mat.shape

#print data_mat.dtype

file_path2=dir+os.sep+files[15]

#print file_path2

dic_label=scipy.io.loadmat(file_path2)

label_mat=dic_label['Label']

file_path3=dir+os.sep+files[16]

print 'fiel 3 path: ', file_path3

dic_T=scipy.io.loadmat(file_path3)

T=dic_T['T']
T=T-1

print T.shape

label=label_mat.ravel()

# Acc=np.zeros((1,4))

Acc=[0,0,0,0]

for i in range (0, 4):
    print "the fold %d" % (i+1)
    train_ind=[]
    for j in range (0, 4):
        if j==i:
            test_ind=T[j]
        else:
            train_ind.extend(T[j])
#    print len(test_ind), len(train_ind)
#    print max(test_ind), max(train_ind)
    train_x=data_mat[train_ind, :]
    test_x=data_mat[test_ind, :]
    train_y=label[train_ind]
    test_y=label[test_ind]
#   SVM   
    clf=svm.LinearSVC()
#   KNN 
#    clf = neighbors.KNeighborsClassifier(n_neighbors=15)
#    Logistic regression
#    clf = linear_model.LogisticRegression()

    clf.fit(train_x, train_y)
    predict_y=clf.predict(test_x)
    Acc[i]=np.mean(predict_y == test_y)
    print "Accuracy: %.2f" % (Acc[i])

print "The mean average classification accuracy: %.2f" % (np.mean(Acc))

# SVM      
(4, 1000)
the fold 1
Accuracy: 0.89
the fold 2
Accuracy: 0.88
the fold 3
Accuracy: 0.89
the fold 4
Accuracy: 0.90
The mean average classification accuracy: 0.89

# KNN      
(4, 1000)
the fold 1
Accuracy: 0.83
the fold 2
Accuracy: 0.84
the fold 3
Accuracy: 0.84
the fold 4
Accuracy: 0.85
The mean average classification accuracy: 0.84

# logistic regression      
(4, 1000)
the fold 1
Accuracy: 0.91
the fold 2
Accuracy: 0.91
the fold 3
Accuracy: 0.90
the fold 4
Accuracy: 0.92
The mean average classification accuracy: 0.91