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,
epochs=100,
C=C,
gamma=gamma)
print('C:', C, 'gamma:', gamma, 'weights:', w, 'bias:', b)
# find number of support vectors
print('# support vectors:', np.sum(alphas != 0.0))
# find overlap SV
if C == 500.0 / 873:
curr_sv = np.argwhere(alphas > 0)
overlap_sv = np.intersect1d(prev_sv, curr_sv)
print('number of overlap SV:', overlap_sv.size)
prev_sv = curr_sv
# calculate training and test errors
err_train = SVM.SVM_kernel_test(X_train,
y_train,
