def main():
# env = SinEnvironment(bo_param2model_param_dic=bo_param2model_param_dic, result_filename=result_filename,
# output_dir=output_dir,
# reload=reload)
env = OneDimGaussianEnvironment(bo_param2model_param_dic=bo_param2model_param_dic, result_filename=result_filename,
output_dir=output_dir,
reload=reload)
agent = GMRF_BO(bo_param_list, env, GAMMA=GAMMA, GAMMA0=GAMMA0, GAMMA_Y=GAMMA_Y, ALPHA=ALPHA,
is_edge_normalized=IS_EDGE_NORMALIZED, gt_available=True, n_early_stopping=N_EARLY_STOPPING,
burnin=BURNIN,
normalize_output=NORMALIZE_OUTPUT, update_hyperparam_func=UPDATE_HYPERPARAM_FUNC,
initial_k=INITIAL_K, initial_theta=INITIAL_THETA, acquisition_func=ACQUISITION_FUNC,
acquisition_param_dic=ACQUISITION_PARAM_DIC)
#
# agent = GP_BO(bo_param_list, env, gt_available=True, my_kernel=kernel, burnin=BURNIN,
# normalize_output=NORMALIZE_OUTPUT, acquisition_func=ACQUISITION_FUNC,
# acquisition_param_dic=ACQUISITION_PARAM_DIC)
# for i in tqdm(range(n_iter)):
for i in range(n_iter):
try:
flg = agent.learn()
agent.plot(output_dir=output_dir)
agent.save_mu_sigma_csv()
if flg == False:
print("Early Stopping!!!")
print(agent.bestX)
print(agent.bestT)
break
except KeyboardInterrupt:
print("Learnig process was forced to stop!")
break
plot_1dim(agent.point_info_manager.T_seq, 'reward.png')
initial_k=INITIAL_K, initial_theta=INITIAL_THETA, acquisition_func=ACQUISITION_FUNC,
acquisition_param_dic=ACQUISITION_PARAM_DIC, n_ctr=n_total_exp)
# for i in tqdm(range(n_iter)):
agent.save_mu_sigma_csv(outfn=mu_sigma_fn, point_info_fn=point_fn)
agent.plot_click_distribution(output_dir)
# agent.learn_from_clicks()
for i in range(n_iter):
try:
flg = agent.learn_from_clicks()
# agent.sample_randomly()
agent.plot_click_distribution(output_dir)
if flg == False:
print("Early Stopping!!!")
print(agent.bestX)
print(agent.bestT)
break
except KeyboardInterrupt:
print("Learnig process was forced to stop!")
# print(agent.X)
# print(agent.Treal)
break
plot_1dim([agent.total_clicked_ratio_list, agent.randomly_total_clicked_ratio_list], 'total_clicked_ratio_list.png')
print(agent.total_clicked_ratio_list)
print(agent.randomly_total_clicked_ratio_list)
def test(ACQUISITION_FUNC):
# ### temporary ###
if os.path.exists(OUTPUT_DIR):
shutil.rmtree(OUTPUT_DIR)
##################
print('GAMMA: ', GAMMA)
print('GAMMA_Y: ', GAMMA_Y)
print('GAMMA0:', GAMMA0)
mkdir_if_not_exist(OUTPUT_DIR)
param_names = sorted(
[x.replace('.csv', '') for x in os.listdir(PARAMETER_DIR)])
bo_param2model_param_dic = {}
bo_param_list = []
for param_name in param_names: # param_name is a param file's name
param_df = pd.read_csv(
os.path.join(PARAMETER_DIR, param_name + '.csv'),
dtype=str) #makes index column type str instead of float
# always read the column of the same name as the file name -- param_name
bo_param_list.append(param_df[param_name].values)
# param_df has a column of its csv file name, e.g. "x"
# and this column is set as the index column
param_df.set_index(param_name, inplace=True)
# dict: param_file name -> column dict (the column with the name "bo_"+param_file name)
# column dict: index column element -> cell value #index column is type str
bo_param2model_param_dic[param_name] = param_df.to_dict()['bo_' +
param_name]
# bo_param_list is a list of every "bo_" column in all the param files of param_names
# print("bo_param_list", bo_param_list)
# env = SinEnvironment(bo_param2model_param_dic=bo_param2model_param_dic,
# result_filename=RESULT_FILENAME,
# output_dir=OUTPUT_DIR,
# reload=RELOAD)
env = OneDimGaussianEnvironment(
bo_param2model_param_dic=bo_param2model_param_dic,
result_filename=RESULT_FILENAME,
output_dir=OUTPUT_DIR,
reload=RELOAD)
# agent = GMRF_BO(bo_param_list, env, GAMMA=GAMMA, GAMMA0=GAMMA0, GAMMA_Y=GAMMA_Y, ALPHA=ALPHA,
# is_edge_normalized=IS_EDGE_NORMALIZED,
# gt_available=True,
# n_early_stopping=N_EARLY_STOPPING,
# burnin=BURNIN,
# normalize_output=NORMALIZE_OUTPUT,
# update_hyperparam_func=UPDATE_HYPERPARAM_FUNC,
# initial_k=INITIAL_K,
# initial_theta=INITIAL_THETA,
# acquisition_func=ACQUISITION_FUNC,
# acquisition_param_dic=ACQUISITION_PARAM_DIC)
agent = GP_BO(bo_param_list,
env,
gt_available=True,
my_kernel=kernel,
burnin=BURNIN,
normalize_output=NORMALIZE_OUTPUT,
acquisition_func=ACQUISITION_FUNC,
acquisition_param_dic=ACQUISITION_PARAM_DIC)
nIter = 200
for i in range(nIter):
flg = agent.learn(drop=True if i < nIter - 1 else False)
agent.plot(output_dir=OUTPUT_DIR)
agent.save_mu_sigma_csv()
if flg == False:
print("Early Stopping!!!")
print("bestX =", agent.bestX)
print("bestT =", agent.bestT)
break
plot_1dim(agent.point_info_manager.T_seq, 'reward.png')
subprocess.call([
"./convert_pngs2gif.sh demo_%s_iter_%d_eps_%f.gif" %
(ACQUISITION_FUNC, nIter, ACQUISITION_PARAM_DIC["eps"])
],
shell=True)
os.system("mv ./output/*.gif ./")