def cross_validation(file):
n = int(0.2 * datanp.shape[0]) #抽樣以五折驗證，所以乘0.2
mean_accuracy_svm = 0
fs_list = []
fs_dict = {} #計算每一折的list 總list feature list 出現結果
for i in range(5):
i = i + 1
print str(i) + "-fold"
if (i == 1): #第一折
idx = range(0, n)
elif (i == 2): #第二折
idx = range(n, n * 2)
elif (i == 3): #第三折
idx = range(n * 2, n * 3)
elif (i == 4): #第四折
idx = range(n * 3, n * 4)
else: #第五折
idx = range(n * 4, file.shape[0])
testdata = file[idx, :]
idx_IN_columns = [i for i in xrange(np.shape(file)[0]) if i not in idx]
traindata = file[idx_IN_columns, :]
#accuracy_svm,feature_select_index_list = SVM.SVM_main(traindata,testdata)
feature_select_index_list = SVM.SVM_main(traindata, testdata)
#mean_accuracy_svm = accuracy_svm + mean_accuracy_svm
fs_list.extend(feature_select_index_list)
#計算總結果list出現頻率
for i in fs_list:
if fs_list.count(i) > 1:
fs_dict[i] = fs_list.count(i)
print 'feature frequency:' + str(fs_dict) #fs_dict 顯示feature 各自出現的頻率
f = sum(list(fs_dict.values())) / len(list(
fs_dict.values())) #計算出現頻率的平均 f 門檻值
print 'feature frequency avg:' + str(f)
fs_list = find_key(fs_dict, f)
print 'feature result:' + str(fs_list)
