def run_plot(if_fast: bool, if_me: bool, if_simu: bool, latency_ms,
latency_serv, performance):
eval_type = 'sdr'
# 设定x axis
source_list = []
# 设定 ms_MeICA
time_list_ms_MeICA = []
snr_list_ms_MeICA = []
# 设定 Me_ICA
time_list_MeICA = []
snr_list_MeICA = []
# 设定 Fast_ICA
time_list_FastICA = []
snr_list_FastICA = []
for s in PGB(range(2, source_number + 1, 1)):
print(f'starting for source_number:{s}')
print('running Ms_MeICA...')
source_list.append(s)
Source_controller = hop_Source_controller(folder_address, duration, s,
mixing_type, max_min,
mu_sigma)
B_Manager = hop_B_Manager(Source_controller)
# use same init B
B_init = B_Manager.get_B()
# Part MS_MeICA
pyfbss_tb.timer_start()
MeICA_controller = hop_MeICA_controller(max_iter, tol, break_coef,
ext_multi_ica,
Source_controller, B_Manager)
hat_S_ms = MeICA_controller.get_hat_S()
time_ms_MeICA = pyfbss_tb.timer_value()
# Time simulator
if if_simu:
# reset time for ms_MeICA
# 算法：
# client发送接受Source_Controller和B_Manager的信息，latency*4
# client将信息发送MeICA_Controller最后接受S，latency*2
# MeICA_Controller接受B，最后更新B，latency*2*(grad+1)
# MeICA_Controller发送接收B_temp到iter,latency*2*(grad+1)
time_ms_MeICA = time_ms_MeICA + \
(4+2+4*MeICA_controller.get_grad()+4)*latency_ms
time_list_ms_MeICA.append(time_ms_MeICA)
snr_list_ms_MeICA.append(
pyfbss_tb.bss_evaluation(Source_controller.get_S(), hat_S_ms,
eval_type))
# Part FastICA
if if_fast:
print('running FastICA...')
pyfbss_tb.timer_start()
X = Source_controller.get_X()
X, V = pyfbss.whiten(X)
B1 = B_init
B2 = pyfbss.newton_iteration(B1, X, max_iter, tol)[0]
hat_S_Fast = np.dot(B2, X)
time_Fast = pyfbss_tb.timer_value()
# Time Simulator
if if_simu:
# 算法：
# 中间的运算时间根据性能比线性缩短，之后latency*2
time_Fast = time_Fast / performance + latency_serv * 2
pass
time_list_FastICA.append(time_Fast)
snr_list_FastICA.append(
pyfbss_tb.bss_evaluation(Source_controller.get_S(), hat_S_Fast,
eval_type))
# Part MeICA
if if_me:
print('running MeICA...')
pyfbss_tb.timer_start()
X = Source_controller.get_X()
pyfbss.Stack = []
B1 = B_init
B2 = pyfbss.multi_level_extraction_newton_iteration(
X, B1, max_iter, tol, break_coef, ext_multi_ica)
hat_S_MeICA = np.dot(B2, X)
time_MeICA = pyfbss_tb.timer_value()
# Time Simulator
if if_simu:
# 算法和FastICA一样
time_MeICA = time_MeICA / 10 + latency_serv * 2
time_list_MeICA.append(time_MeICA)
snr_list_MeICA.append(
pyfbss_tb.bss_evaluation(Source_controller.get_S(),
hat_S_MeICA, eval_type))
# plot
plt.figure()
plt.subplot(1, 2, 1)
plt.plot(np.array(source_list),
np.array(time_list_ms_MeICA),
'o-',
label='MircoService_MeICA')
if if_fast:
plt.plot(np.array(source_list),
np.array(time_list_FastICA),
'o-',
label='FastICA')
if if_me:
plt.plot(np.array(source_list),
np.array(time_list_MeICA),
'o-',
label='MeICA')
plt.title(f'Time Use, base on Simulator:{if_simu}')
plt.xlabel('Source Number')
plt.ylabel('Time[ms]')
# 添加图例
plt.legend(loc='best')
plt.subplot(1, 2, 2)
plt.plot(np.array(source_list),
np.array(snr_list_ms_MeICA),
'o-',
label='MircoService_MeICA')
if if_fast:
plt.plot(np.array(source_list),
np.array(snr_list_FastICA),
'o-',
label='FastICA')
if if_me:
plt.plot(np.array(source_list),
np.array(snr_list_MeICA),
'o-',
label='MeICA')
plt.title(f'SNR')
plt.xlabel('Source Number')
plt.ylabel('SNR[dB]')
# 添加图例
plt.legend(loc='best')
plt.show()
tmp_fastica = [source_number]
tmp_cdica = [source_number]
tmp_aeica = [source_number]
tmp_meica = [source_number]
for test_i in np.arange(1, 2, 1):
S, A, X = pyfbss_tb.generate_matrix_S_A_X(
folder_address, duration, source_number, mixing_type="normal", max_min=(1, 0.01), mu_sigma=(0, 1))
print('type eval_dB time(ms) for ' +
str(source_number)+' sources, ' + str(test_i) + '-th test.')
print(
'--------------------------------------------------------------------------------')
eval_type = 'sdr'
# time and accuracy of FastICA
pyfbss_tb.timer_start()
hat_S = pyfbss.fastica(X, max_iter=100, tol=1e-04)
time = pyfbss_tb.timer_value()
Eval_dB = pyfbss_tb.bss_evaluation(S, hat_S, eval_type)
tmp_fastica.extend([Eval_dB, time])
print('FastICA: ', Eval_dB, '; ', time)
# time and accuracy of CdICA
pyfbss_tb.timer_start()
hat_S = pyfbss.cdica(
X, max_iter=100, tol=1e-04, ext_initial_matrix=0)
time = pyfbss_tb.timer_value()
Eval_dB = pyfbss_tb.bss_evaluation(S, hat_S, eval_type)
tmp_cdica.extend([Eval_dB, time])
print('CdICA: ', Eval_dB, '; ', time)