def create_D_function(camp, max_bid):
large_storage_media = "/media/onetbssd/rlb/"
src = "ipinyou"
obj_type = "clk"
clk_vp = 1
# N = 10000
N = 500
c0 = 1 / 8
gamma = 1
overwrite = False
if src == "ipinyou":
camps = config.ipinyou_camps_to_test
data_path = config.ipinyouPath
max_market_price = config.ipinyou_max_market_price
camp_info = config.get_camp_info(camp, src)
aution_in_file = data_path + camp + "/test.theta.txt"
opt_obj = Opt_Obj(obj_type, int(clk_vp * camp_info["cost_train"] / camp_info["clk_train"]))
B = int(camp_info["cost_train"] / camp_info["imp_train"] * c0 * N)
large_storage_folder = large_storage_media + src + "/" + camp + "/bid-model/"
if not os.path.exists(large_storage_folder):
os.makedirs(large_storage_folder)
env = BidEnv(camp_info, aution_in_file)
agent = bidding_agent_rtb_rl_dp_tabular()
agent.init(env, camp_info, opt_obj, gamma)
setting = "{}, camp={}, algo={}, N={}, c0={}" \
.format(src, camp, "rlb_rl_fa", N, c0)
bid_log_path = config.projectPath + "bid-log/{}.txt".format(setting)
# Approximating D(t,b) function
m_pdf = calc_m_pdf(camp_info["price_counter_train"])
D_function_path = large_storage_folder + "rlb_dnb_gamma={}_N={}_{}.txt".format(gamma, N, obj_type)
print("D_function_path = " + D_function_path)
if (not os.path.isfile(D_function_path)) or overwrite:
# print("START: Approximating V function by dynamic programming... ")
agent.calc_Dnb(N, B, max_market_price, m_pdf, D_function_path)
# print("END: Approximating V function by dynamic programming.")
else:
agent.load_Dnb(N,B,D_function_path)
X, Y = agent.run_and_output_datasets(bid_log_path, N, c0, max_bid, delimiter=" ", save_log=False)
# Then train a NN using the Dnd function
NN_model_path = large_storage_folder + "fa_dnb_gamma={}_N={}_{}.pickle".format(gamma, N, obj_type)
NN_model_txt_path = large_storage_folder + "fa_dnb_gamma={}_N={}_{}.txt".format(gamma, N, obj_type)
return agent, src, N, D_function_path, large_storage_folder, NN_model_path, NN_model_txt_path, opt_obj, camp_info, X, Y
dim = 2
net_type = "nn"
src = "ipinyou"
camp = "1458"
if len(_argv) == 2:
camp = _argv[1]
obj_type = "clk"
clk_vp = 1
N = 5000
tag = src + "_" + camp + "_" + model + "_" + net_type + "_N={}_{}".format(
N, obj_type) + "_" + getTime()
if src == "ipinyou":
data_path = config.ipinyouPath
camp_info = config.get_camp_info(camp, src)
elif src == 'vlion':
data_path = config.vlionPath
camp_info = config.get_camp_info(camp, src)
elif src == "yoyi":
data_path = config.yoyiPath
camp_info = config.get_camp_info(camp, src)
opt_obj = Opt_Obj(
obj_type,
int(clk_vp * camp_info["cost_train"] / camp_info["clk_train"]))
avg_theta = camp_info["clk_train"] / camp_info["imp_train"]
if obj_type == "profit":
avg_theta *= opt_obj.clk_v
b_bound = 800
if model == "dnb":
dim = 2
net_type = "nn"
src = "ipinyou"
camp = "1458"
if len(_argv) == 2:
camp = _argv[1]
obj_type = "clk"
clk_vp = 1
N = 5000
tag = src + "_" + camp + "_" + model + "_" + net_type + "_N={}_{}".format(N, obj_type) + "_" + getTime()
if src == "ipinyou":
data_path = config.ipinyouPath
camp_info = config.get_camp_info(camp, src)
elif src == 'vlion':
data_path = config.vlionPath
camp_info = config.get_camp_info(camp, src)
elif src == "yoyi":
data_path = config.yoyiPath
camp_info = config.get_camp_info(camp, src)
opt_obj = Opt_Obj(obj_type, int(clk_vp * camp_info["cost_train"] / camp_info["clk_train"]))
avg_theta = camp_info["clk_train"] / camp_info["imp_train"]
if obj_type == "profit":
avg_theta *= opt_obj.clk_v
b_bound = 800
n_bound = 50
max_train_round = 500
# Generate testing and training datasets for target campaing.
camp_to_test = config.ipinyou_camps_to_test[0]
camp_to_test_file = data_path + camp_to_test + "/test.theta.txt"
training_filename, testing_filename, len_training, len_testing = generate_training_and_evaluation_files_for_target_camp(
camp_to_test_file, 0.8)
T_shoots_list = [.1, .2, .3, .4, .5, .6, .7, .8, .9, 1]
#T_shoots_list = [.1]
T_shoots_list = np.array(T_shoots_list)
T_shoots_list = T_shoots_list * len_training
T_shoots_list = T_shoots_list.astype(int)
for T_shoots_learning_size in T_shoots_list:
for camp in camps:
camp_info = config.get_camp_info(camp, src)
aution_in_file = data_path + camp + "/test.theta.txt"
opt_obj = Opt_Obj(
obj_type,
int(clk_vp * camp_info["cost_train"] / camp_info["clk_train"]))
B = int(camp_info["cost_train"] / camp_info["imp_train"] * c0 * N)
large_storage_media = "/media/onetbssd/rlb/"
# Create log file folder location.
if not os.path.exists(config.projectPath + "bid-log"):
os.makedirs(config.projectPath + "bid-log")
# Create training merged set:
train_camps = []
