def update_RESAMPLED_DVH_BIN_COUNT(attr, old, new):
try:
bin_count = int(new)
except ValueError:
bin_count = 10
options.RESAMPLED_DVH_BIN_COUNT = bin_count
save_options(options)
def update_ALPHA_val(attr, old, new):
try:
alpha = float(new)
except ValueError:
alpha = 1.
setattr(options, input_ALPHA_var.value, alpha)
save_options(options)
def update_ENDPOINT_COUNT(attr, old, new):
try:
ep_count = int(new)
except ValueError:
ep_count = 10
options.ENDPOINT_COUNT = ep_count
save_options(options)
def update_LINE_WIDTH_val(attr, old, new):
try:
line_width = float(new)
except ValueError:
line_width = 1.
setattr(options, input_LINE_WIDTH_var.value, line_width)
save_options(options)
def update_OPTIONAL_TABS(attr, old, new):
for i in range(len(input_OPTIONAL_TABS.labels)):
key = input_OPTIONAL_TABS.labels[i]
if i in new:
options.OPTIONAL_TABS[key] = True
else:
options.OPTIONAL_TABS[key] = False
save_options(options)
def update_SIZE_val(attr, old, new):
if 'FONT' in input_SIZE_var.value:
try:
size = str(int(new)) + 'pt'
except ValueError:
size = '10pt'
else:
try:
size = float(new)
except ValueError:
size = 1.
setattr(options, input_SIZE_val.value, size)
save_options(options)
def update_LINE_DASH_val(attr, old, new):
setattr(options, input_LINE_DASH_var.value, new)
save_options(options)
def update_input_COLORS_val(attr, old, new):
setattr(options, input_COLORS_var.value, new)
save_options(options)
def update_LITE_VIEW(attr, old, new):
options.LITE_VIEW = bool(1 - new)
save_options(options)
def update_DISABLE_BACKUP_TAB(attr, old, new):
options.DISABLE_BACKUP_TAB = bool(1 - new)
save_options(options)
def update_AUTH_USER_REQ(attr, old, new):
options.AUTH_USER_REQ = bool(1 - new)
save_options(options)
def main():
opt = {}
# Basic setting
opt['batch_fraction'] = 20 # use for LIBSVM data
opt['dtype'] = tf.float32
opt['repetition'] = 1 # the number of executions
# Logistic regression parameter
opt['LR_lambda'] = 1. # regularization coefficient
# SANC parameter setting
opt['SANC_eta1'] = 0.2
opt['SANC_eta2'] = 0.8
opt['SANC_gamma'] = 2.
opt['SANC_epsilon'] = 1e-4
opt['SANC_lanczos_max_iters'] = 5
# CR parameter setting
opt['CR_sigma'] = 5.
opt['CR_lanczos_max_iters'] = 5
# SCR parameter setting
opt['SCR_eta1'] = 0.2
opt['SCR_eta2'] = 0.8
opt['SCR_gamma'] = 2.
opt['SCR_lanczos_max_iters'] = 5
# NCD parameter setting
opt['NCD_lanczos_max_iters'] = 5
opt['NCD_L1_nc'] = 10.
opt['NCD_L2_nc'] = 10.
opt['NCD_epsilon'] = 1e-4
# # SGD parameter setting
opt['SGD_learning_rate'] = 0.01
# CRM parameter setting
opt['CRM_beta'] = 12
data = 'w1a' # dataset
opt['oraclecall_limit'] = 30000
_main = _main_lr
# Test for figure 1 in the paper
# opt['SANC_sigma_init'] = 0.001
# opt['SCR_sigma_init'] = 0.001
# fig_data['SANC'] = _main(method = 'SANC', data = data, opt = opt)
# fig_data['SCR'] = _main(method = 'SCR', data = data, opt = opt)
# fig_data['CR'] = _main(method = 'CR', data = data, opt = opt)
# fig_data['CRM'] = _main(method = 'CRM', data = data, opt = opt)
# filename_suffix = datetime.datetime.now().strftime("%y%m%d_%H%M%S")
# save_figures(fig_data, dataname = data, fn_prefix=filename_suffix, log_scale = False, show = True)
# save_options(opt, dataname = data, fn_prefix=filename_suffix)
# save_result_data(fig_data, dataname= data, fn_prefix = filename_suffix)
# # Test for figure 1(a) top in the paper
opt['SANC_sigma_init'] = 1.
opt['SCR_sigma_init'] = 1.
opt['oraclecall_limit'] = 50000
opt['repetition'] = 1 # the number of executions
# fig_data2 = {}
# fig_data2['SANC'] = _main(method = 'SANC', data = data, opt = opt)
# fig_data2['SCR'] = _main(method = 'SCR', data = data, opt = opt)
# fig_data2['CR'] = _main(method = 'CR', data = data, opt = opt)
# fig_data2['SGD'] = _main(method = 'SGD', data = data, opt = opt)
# fig_data2['NCD'] = _main(method = 'NCD', data = data, opt = opt)
# fig_data2['CRM'] = _main(method = 'CRM', data = data, opt = opt)
# filename_suffix = datetime.datetime.now().strftime("%y%m%d_%H%M%S")
# save_figures(fig_data2, dataname = data, fn_prefix=filename_suffix, log_scale = False, show = True)
# save_options(opt, dataname = data, fn_prefix=filename_suffix)
# save_result_data(fig_data2, dataname= data, fn_prefix = filename_suffix)
# figure 1(a) bottom
# data = 'higgs'
# opt['SANC_sigma_init'] = 1.
# opt['SCR_sigma_init'] = 1.
# opt['oraclecall_limit'] = 60000000
# opt['repetition'] = 1 # the number of executions
# opt['SANC_L1_nc'] = 10.
# opt['SANC_L2_nc'] = 100.
# fig_data = {}
# fig_data['SANC'] = _main(method = 'SANC', data = data, opt = opt)
# fig_data['SCR'] = _main(method = 'SCR', data = data, opt = opt)
# fig_data['CR'] = _main(method = 'CR', data = data, opt = opt)
# fig_data['SGD'] = _main(method = 'SGD', data = data, opt = opt)
# fig_data['NCD'] = _main(method = 'NCD', data = data, opt = opt)
# fig_data['CRM'] = _main(method = 'CRM', data = data, opt = opt)
# filename_suffix = datetime.datetime.now().strftime("%y%m%d_%H%M%S")
# save_figures(fig_data, dataname = data, fn_prefix=filename_suffix, log_scale = False, show = True)
# save_options(opt, dataname = data, fn_prefix=filename_suffix)
# save_result_data(fig_data, dataname= data, fn_prefix = filename_suffix)
# exit()
opt["MLP_alpha"] = 0.05
# figure 1(b) top
_main = _main_mlp
# fig_data = {}
# opt['SANC_L1_nc'] = 100.
# opt['SANC_L2_nc'] = 100.
# opt['SANC_eta1'] = 0.1
# opt['SANC_eta2'] = 0.3
# opt['SCR_eta1'] = 0.1
# opt['SCR_eta2'] = 0.3
# opt['NCD_L1_nc'] = 100.
# opt['NCD_L2_nc'] = 100.
# opt['batch_size'] = 128
# opt['oraclecall_limit'] = 40000
# data = 'seismic'
# opt['num_classes'] = 3
# opt['CRM_beta'] = 2
# opt['repetition'] = 10 # the number of executions
# fig_data['CR'] = _main(method = 'CR', data = data, opt = opt)
# fig_data['CRM'] = _main(method = 'CRM', data = data, opt = opt)
# fig_data['NCD'] = _main(method = 'NCD', data = data, opt = opt)
# fig_data['SCR'] = _main(method = 'SCR', data = data, opt = opt)
# fig_data['SANC'] = _main(method = 'SANC', data = data, opt = opt)
# fig_data['SGD'] = _main(method = 'SGD', data = data, opt = opt)
# filename_suffix = datetime.datetime.now().strftime("%y%m%d_%H%M%S")
# save_figures(fig_data, dataname = data, fn_prefix=filename_suffix, log_scale = False, show = True)
# save_options(opt, dataname = data, fn_prefix=filename_suffix)
# save_result_data(fig_data, dataname= data, fn_prefix = filename_suffix)
# figure 1(b) bottom
data = 'segment'
opt['num_classes'] = 7
opt["MLP_alpha"] = 0.05
opt['SANC_L1_nc'] = 100.
opt['SANC_L2_nc'] = 100.
opt['SANC_eta1'] = 0.1
opt['SANC_eta2'] = 0.3
opt['SCR_eta1'] = 0.1
opt['SCR_eta2'] = 0.3
opt['NCD_L1_nc'] = 100.
opt['NCD_L2_nc'] = 100.
opt['CRM_beta'] = 2
opt['repetition'] = 10 # the number of executions
opt['batch_size'] = 128
opt['oraclecall_limit'] = 50000
fig_data = {}
fig_data['CR'] = _main(method='CR', data=data, opt=opt)
fig_data['CRM'] = _main(method='CRM', data=data, opt=opt)
fig_data['NCD'] = _main(method='NCD', data=data, opt=opt)
fig_data['SCR'] = _main(method='SCR', data=data, opt=opt)
fig_data['SANC'] = _main(method='SANC', data=data, opt=opt)
fig_data['SGD'] = _main(method='SGD', data=data, opt=opt)
filename_suffix = datetime.datetime.now().strftime("%y%m%d_%H%M%S")
save_figures(fig_data,
dataname=data,
fn_prefix=filename_suffix,
log_scale=False,
show=True)
save_options(opt, dataname=data, fn_prefix=filename_suffix)
save_result_data(fig_data, dataname=data, fn_prefix=filename_suffix)