def Tune():
'''
for small net
re_max = -4
re_min = -14
lr_max = -6
lr_min = -10
'''
re_max = -4
re_min = -14
lr_max = -6
lr_min = -10
distribution = np.array(['uniform', 'normal'])
iteration = 1
max_batch = 1000
verbose = True
#list_lr = np.exp(np.linspace(start=-9, stop=-13, num=iteration))
list_lr = np.exp(np.random.uniform(low=lr_min, high=lr_max, size=iteration))
list_re = np.exp(np.random.uniform(low=re_min, high=re_max, size=iteration))
mask_distribution = np.random.randint(low=0, high=len(distribution), size=iteration)
list_dstribution = distribution[mask_distribution]
_Notice()
list_report = []
list_logs = []
mcg = ConfigUtils.ModelCfgGenerator()
ccg = ConfigUtils.TrainCfgGenerator()
for i in range(iteration):
print('number of hyperparam setting tried: {}/{}'.format(i+1, iteration))
cfg_model = mcg.GetContent(distribution=list_dstribution[i], reg=list_re[i])
cfg_train = ccg.GetContent(lr=list_lr[i], max_batch=max_batch, verbose=verbose)
mnist = input_data.read_data_sets(cfg_train.data_dir , one_hot=False, seed=231)
report, logs = Run(cfg_model,cfg_train, mnist)
log_df = Logs2Df(logs)
list_report.append(report)
list_logs.append(log_df)
df = Reports2Df(list_report)
return df, list_logs
