y_train,
epochs=100,
C=C,
gamma=0.01,
schedule_func=schedule_B)
print('C:', C, 'weights:', w)
err_train = SVM.SVM_primal_test(X_train, y_train, w)
err_test = SVM.SVM_primal_test(X_test, y_test, w)
print('training error:', err_train, 'test error', err_test)
### Q3 SVM dual and kernal
# SVM dual
print('\nSVM dual')
for C in C_vals:
w, b = SVM.SVM_dual(X_train, y_train, C)
print('C:', C, 'weights:', w, 'bias:', b)
err_train = SVM.SVM_dual_test(X_train, y_train, w, b)
err_test = SVM.SVM_dual_test(X_test, y_test, w, b)
print('training error:', err_train, 'test error', err_test)
# SVM kernel
print('\nSVM with kernel')
gammas = [0.1, 0.5, 1, 5, 100]
for C in C_vals:
# find overlap SV for C=500/875
if C == 500.0 / 873:
prev_sv = np.zeros(X_train.shape[0])
for gamma in gammas:
w, b, alphas = SVM.SVM_kernel(X_train,
y_train,
def SVM_valabel(root_dir):
splits_num = 5
#先获取全部数据
x = []
y1 = []
y2 = []
for i in range(splits_num):
#从文件获取数据
subjects = np.loadtxt(root_dir + 'subject/subject_video_' + str(i) +
'.txt',
dtype=int)
features = np.loadtxt(root_dir + 'feature/EEG_feature_' + str(i) +
'.txt')
va_labels = np.loadtxt(root_dir + 'valabel/valence_arousal_label_' +
str(i) + '.txt',
dtype=int)
if 'HCI' in root_dir:
emotions = np.loadtxt(root_dir + 'emotion/EEG_emotion_category_' +
str(i) + '.txt',
dtype=int)
# #归一化
# subjects = subjects / np.array([np.max(subjects[:,0]),np.max(subjects[:,1])])
# features /= np.max(features)
# if 'HCI' in root_dir:
# emotions /= np.max(emotions)
# # va_labels = va_labels / np.array([np.max(va_labels[:,0]),np.max(va_labels[:,1])])
x_temp = []
y1_temp = []
y2_temp = []
length = len(subjects)
for j in range(length):
x_temp.append([])
x_temp[j].extend(subjects[j])
if 'HCI' in root_dir:
x_temp[j].append(emotions[j])
x_temp[j].extend(features[j])
y1_temp.append(va_labels[j][0])
y2_temp.append(va_labels[j][1])
x.append(x_temp)
y1.append(y1_temp)
y2.append(y2_temp)
# y = np.array(y,dtype=int)
# y_valenc,y_arousal = np.split(y, 2, axis = 1)
cvscores = [[], []]
#交叉验证
for i in range(splits_num):
x_train = []
y1_train = []
y2_train = []
x_test = []
y1_test = []
y2_test = []
for j in range(splits_num):
if j == i:
x_test.extend(x[j])
y1_test.extend(y1[j])
y2_test.extend(y2[j])
else:
x_train.extend(x[j])
y1_train.extend(y1[j])
y2_train.extend(y2[j])
x_train = np.array(x_train)
y1_train = np.array(y1_train)
y2_train = np.array(y2_train)
x_test = np.array(x_test)
y1_test = np.array(y1_test)
y2_test = np.array(y2_test)
# svm = SVC(C=10)
svm = SVM.SVM_dual(100, ['rbf', 1], C=6, toler=1e-3)
svm.fit(x_train, y1_train)
score1 = svm.score(x_test, y1_test)
print(" on v_label, %s: %.2f%%" % ('acc', score1 * 100))
cvscores[0].append(score1)
# svm = SVC(C=10)
svm = SVM.SVM_dual(100, ['rbf', 1], C=6, toler=1e-3)
svm.fit(x_train, y2_train)
score2 = svm.score(x_test, y2_test)
print(" on a_label, %s: %.2f%%\n" % ('acc', score2 * 100))
cvscores[1].append(score2)
cvscores = np.array(cvscores)
average_score = np.sum(cvscores, axis=1) / splits_num
write_score(cvscores, average_score, root_dir + 'SVM/valabel_cvscores.txt')
return average_score
