import sys
import pandas as pd
import matplotlib.pyplot as plt
sys.path.append('..')
from threshold import get_threshold
# %% load datasets
test = pd.read_csv(
'../../data/predicted/linear_regression/Safaricom-Ltd(SCOM).csv')
data = pd.read_csv('../../data/processed/Safaricom-Ltd(SCOM).csv')
threshold = get_threshold(data)
test.index = data[threshold:].index
# %% plot and save price prediction
plt.plot(data['Price'], label='Actual Price')
plt.plot(test['Prediction'], label='Predicted Price')
plt.xlabel('Time (Days)')
plt.ylabel('Share Price (KSh.)')
plt.title('Safaricom Price Prediction (Linear Regression)')
plt.legend()
plt.savefig('../../reports/figures/Safaricom_linear_regression.png')
plt.show()
def start_bidding(fo, reqs, cpms, pred_ctr):
day = 12 # 20130612这一天
BGT = 10000000 # 一天的总预算30297100
print('今天的总预算:', BGT)
total_cost = 0 # 模拟过程中的实际总花费
total_cost_true = 0 # 数据集中的实际总花费
n = 12 # 将一天划分为12个时段
nums = []
for j in range(0, 12):
nums.append(0)
p = get_p(BGT) # 分发概率 budgetpacing.py
length = 0
for i in range(0, 24, int(24 / n)): # 划分时段
# 2、无预算分配策略
# budget = BGT - total_cost
# pt_next = p_next(int(i / 2), BGT) * (BGT - total_cost)
# 2、平均分配
# budget = BGT / n
# pt_next = p_next(int(i / 2), BGT) * (BGT - total_cost)
# 3、基于点击率分配
# sum_tmp = 0
# ii = int(i / 2)
# for k in range(ii, 12):
# sum_tmp += pred_ctrs[k]
# budget = (BGT - total_cost) * pred_ctrs[int(i / 2)] / sum_tmp
# pt_next = p_next(int(i / 2), BGT) * (BGT - total_cost)
# 4、基于流量分配
# sum_tmp = 0
# ii = int(i / 2)
# for k in range(ii, 12):
# sum_tmp += reqs[k]
# budget = (BGT - total_cost) * reqs[int(i / 2)] / sum_tmp
# pt_next = p_next(int(i / 2), BGT) * (BGT - total_cost)
# 5、按时段长度占剩余时段总长度的比例分配
# budget = (BGT - total_cost) * (1 / (12 - (i / 2)))
# pt_next = p_next(int(i / 2), BGT) * (BGT - total_cost)
# 6、我们的
if i == 0:
budget = reqs[int(i / 2)] * 1 * cpms[int(i / 2)] * p[int(i / 2)]
# budget = BGT
pt_next = p_next(int(i / 2), BGT) * (BGT - total_cost)
if i != 22 and i != 0:
pt_next = p_next(int(i / 2), BGT) * (BGT - total_cost)
# budget = (BGT - tocost) * pt_next / wr
budget = reqs[int(i / 2)] * 1 * cpms[int(
i / 2)] * pt_next # 因为用了赢标日志模拟广告请求，所以将预测的赢标率都设为了1
if i == 22:
pt_next = p_next(int(i / 2), BGT) * (BGT - total_cost)
budget = min(reqs[int(i / 2)] * 1 * cpms[int(i / 2)] * pt_next,
BGT - total_cost)
num = reqs[int(i / 2)] * pt_next
nums[int(i / 2)] += int(num)
threshold = float(get_threshold(int(i / 2), int(num)))
# threshold = 0
ccfm, cost_t, pctr = get_timeSlot_data(day, i, length)
length += len(ccfm)
total_cost_true += cost_t
if total_cost <= BGT:
print('\n{0}-{1}时段log中的实际花费:'.format(i, i + 2), cost_t)
print('本时段预算:', budget)
result, cost, winrate, cpm = simulateBidding(
ccfm, budget, pctr, threshold, cpms[int(i / 2)], int(i / 2))
fo.write(result + '\n')
total_cost += cost
print('预算剩余:', BGT - total_cost)
header = 'budget\t\tspend\t\tbid\t\tipm\t\tclk\t\tclk_true\t\tcnv\t\tecpc\t\twr'
print(header + '\n')
print(result)
else:
print('今天的预算已用完！！！')
break
fo.close()
print('\nlog中统计得到的一天的实际总花费:', total_cost_true)