def test_shufflenet():
shufflenet_creator = shuffle.Model("shufflenet")
mg = ex.model_generator(name='shufflenet_profile')
mg.set_creator(shufflenet_creator.model_creator)
mg.set_stats_updater(shufflenet_creator.stat_updater)
# Mapping out_channel with nr_group
out_ch_map = {1: 144, 2: 200, 3: 240, 4: 272, 8: 384}
param_list = []
stat_list = []
for filter_group in [1, 2, 3, 4, 8]:
for complexity_scale_factor in [0.25, 0.5, 1.0]:
param = {
'filter_group': filter_group,
'complexity_scale_factor': complexity_scale_factor,
'out_channel': out_ch_map[filter_group]
}
param_list.append(param)
stat_list.append(mg.set_and_stats(param))
param_fields = ['filter_group', 'complexity_scale_factor', 'out_channel']
stat_fields = [
'param', 'flops', 'single_thread_mean', 'single_thread_var',
'multi_thread_mean', 'multi_thread_var', 'file_size'
]
report.write_data_to_csv(param_list, param_fields,
'model_parameters_shufflenet')
report.write_data_to_csv(stat_list, stat_fields,
'model_behaviour_shufflenet')
# Prepare for execution graph plotting
x_label = []
y_single_mean = []
y_multi_mean = []
y_single_error = []
y_multi_error = []
def create_label(d):
ret = ""
for key, value in d.items():
if ret == "":
ret = str(value)
else:
ret = ret + "_" + str(value)
return ret
for i in range(len(stat_list)):
x_label.append(create_label(param_list[i]))
y_single_mean.append(stat_list[i]['single_thread_mean'])
y_multi_mean.append(stat_list[i]['multi_thread_mean'])
y_single_error.append(stat_list[i]['single_thread_var'])
y_multi_error.append(stat_list[i]['multi_thread_var'])
name = ['single', 'multi']
y_data_mean = [y_single_mean, y_multi_mean]
y_data_var = [y_single_error, y_multi_error]
report.graph_data_bar(x_label, y_data_mean, y_data_var, name, multi=True)
def test_mobilenet():
mobilenet_creator = mobile.Model("mobilenet_v1")
mg = ex.model_generator(name='mobilenet_profile')
mg.set_creator(mobilenet_creator.model_creator)
mg.set_stats_updater(mobilenet_creator.stat_updater)
param_list = []
stat_list = []
for res in [1, 0.858, 0.715, 0.572]:
for width in [1, 0.75, 0.5]:
for depth in [1, 2]:
param = {
'resolution_multiplier': res,
'width_multiplier': width,
'depth_multiplier': depth
}
param_list.append(param)
stat_list.append(mg.set_and_stats(param))
param_fields = [
'resolution_multiplier', 'width_multiplier', 'depth_multiplier'
]
stat_fields = [
'param', 'flops', 'single_thread_mean', 'single_thread_var',
'multi_thread_mean', 'multi_thread_var', 'file_size'
]
report.write_data_to_csv(param_list, param_fields,
'model_parameters_mobilenet')
report.write_data_to_csv(stat_list, stat_fields,
'model_behaviour_mobilenet')
# Prepare for execution graph plotting
x_label = []
y_single_mean = []
y_multi_mean = []
y_single_error = []
y_multi_error = []
def create_label(d):
ret = ""
for key, value in d.items():
if ret == "":
ret = str(value)
else:
ret = ret + "_" + str(value)
return ret
for i in range(len(stat_list)):
x_label.append(create_label(param_list[i]))
y_single_mean.append(stat_list[i]['single_thread_mean'])
y_multi_mean.append(stat_list[i]['multi_thread_mean'])
y_single_error.append(stat_list[i]['single_thread_var'])
y_multi_error.append(stat_list[i]['multi_thread_var'])
name = ['single', 'multi']
y_data_mean = [y_single_mean, y_multi_mean]
y_data_var = [y_single_error, y_multi_error]
report.graph_data_bar(x_label, y_data_mean, y_data_var, name, multi=True)
def test_squeezenet():
squeezenet_creator = squeeze.Model("squeezenet")
mg = ex.model_generator(name = 'squeezenet_profile')
mg.set_creator(squeezenet_creator.model_creator)
mg.set_stats_updater(squeezenet_creator.stat_updater)
param_list = []
stat_list = []
for base_expand_kernels in [128]:
for expansion_increment in [128]:
for pct in [0.5]:
for freq in [2]:
for SR in [0.125]:
param = {'base_expand': base_expand_kernels,
'expansion_increment': expansion_increment,
'expansion_filter_ratio': pct,
'filter_expansion_freq': freq,
'squeeze_ratio': SR}
param_list.append(param)
stat_list.append(mg.set_and_stats(param))
param_fields = ['base_expand', 'expansion_increment',
'expansion_filter_ratio', 'filter_expansion_freq'
'squeeze_ratio']
stat_fields = ['param','flops','single_thread_mean','single_thread_var',
'multi_thread_mean','multi_thread_var', 'file_size']
report.write_data_to_csv(param_list, param_fields, 'model_parameters')
report.write_data_to_csv(stat_list, stat_fields, 'model_behaviour')
# Prepare for execution graph plotting
x_label = []
y_single_mean = []
y_multi_mean = []
y_single_error = []
y_multi_error = []
def create_label(d):
ret = ""
for key, value in d.items():
if ret=="":
ret = str(value)
else:
ret = ret+"_"+str(value)
return ret
for i in range(len(stat_list)):
x_label.append(create_label(param_list[i]))
y_single_mean.append(stat_list[i]['single_thread_mean'])
y_multi_mean.append(stat_list[i]['multi_thread_mean'])
y_single_error.append(stat_list[i]['single_thread_var'])
y_multi_error.append(stat_list[i]['multi_thread_var'])
name = ['single', 'multi']
y_data_mean = [y_single_mean, y_multi_mean]
y_data_var = [y_single_error, y_multi_error]
report.graph_data_bar(x_label, y_data_mean, y_data_var, name, multi=True)