def make_feats(end_date):
t0 = time.time()
print('数据key为：{}'.format(end_date))
result_path = cache_path + 'train_set_{0}.hdf'.format(end_date)
if os.path.exists(result_path) & 0:
result = pd.to_hdf(result_path, 'w', complib='blosc', complevel=5)
else:
print('添加label')
label = get_label(end_date)
print('开始构造特征...')
result = []
result.append(get_lastdays_of_st(label, end_date, 30)) # 前一周每天的值
# result.append(get_sum_of_store_item(label, end_date, 1)) # 前1天的和
# result.append(get_sum_of_store_item(label, end_date, 3)) # 前3天的和
result.append(get_sum_of_store_item(label, end_date, 7)) # 前7天的和
result.append(get_sum_of_store_item(label, end_date, 14)) # 前14天的和
result.append(get_sum_of_store_item(label, end_date, 21)) # 前21天的和
result.append(get_sum_of_store_item(label, end_date, 28)) # 前28天的和
result.append(get_sum_of_store_item(label, end_date, 42)) # 前42天的和
result.append(get_sum_of_store_item(label, end_date, 70)) # 前70天的和
# result.append(get_sum_of_store_item(label, end_date, 98)) # 前98天的和
result.append(get_sum_of_store_item(label, end_date, 140)) # 前140天的和
result.append(get_lastdays_of_prom(label, end_date, 7)) # 前7天是否促销
result.append(get_sum_of_prom(label, end_date, 14)) # 前14天促销次数
result.append(get_sum_of_prom(label, end_date, 28)) # 前28天促销次数
result.append(get_sum_of_prom(label, end_date, 140)) # 前140天促销次数
# result.append(get_sum_of_week(label, end_date, 140)) #获取前一个月的week和
#上次购买时间
result.append(get_promo_of_store_item(label,
end_date)) # 上次促销开始的时间和结束时间
result.append(label)
print('开始合并特征...')
result = concat(result).reindex()
result = second_feat(result)
# print('存储数据...')
# result.to_hdf(result_path, 'w', complib='blosc', complevel=5)
print('特征矩阵大小：{}'.format(result.shape))
print('生成特征一共用时{}秒'.format(time.time() - t0))
return result
def make_feats(end_date):
t0 = time.time()
print('数据key为：{}'.format(end_date))
result_path = cache_path + 'train_set_{0}.hdf'.format(end_date)
if os.path.exists(result_path) & 0:
result = pd.to_hdf(result_path, 'w', complib='blosc', complevel=5)
else:
print('添加label')
label = get_label(end_date)
base_feat = get_base_feat(label,end_date) # 基础特征
date_feat = get_date_feat(label,end_date) # 时间特征
print('开始合并特征...')
result = concat([label,base_feat,date_feat])
result = second_feat(result)
# print('存储数据...')
# result.to_hdf(result_path, 'w', complib='blosc', complevel=5)
print('特征矩阵大小：{}'.format(result.shape))
print('生成特征一共用时{}秒'.format(time.time() - t0))
return result
def make_feats(end_date):
t0 = time.time()
print('数据key为：{}'.format(end_date))
result_path = cache_path + 'train_set_{0}.hdf'.format(end_date)
if os.path.exists(result_path) & 0:
result = pd.to_hdf(result_path, 'w', complib='blosc', complevel=5)
else:
print('添加label')
label = get_label(end_date)
base_feat = get_base_feat(label, end_date) # 基础特征
print('开始构造特征...')
result = [base_feat]
result.append(get_lastdays_of_store_item(base_feat, end_date,
30)) # 前一周每天的值
result.append(get_sum_of_store_item(base_feat, end_date, 7)) # 前7天的和
result.append(get_sum_of_store_item(base_feat, end_date, 14)) # 前14天的和
result.append(get_sum_of_store_item(base_feat, end_date, 21)) # 前21天的和
result.append(get_sum_of_store_item(base_feat, end_date, 28)) # 前28天的和
result.append(get_sum_of_store_item(base_feat, end_date, 42)) # 前42天的和
result.append(get_sum_of_store_item(base_feat, end_date, 70)) # 前70天的和
result.append(get_sum_of_store_item(base_feat, end_date,
140)) # 前140天的和
result.append(get_lastdays_of_prom(base_feat, end_date, 7)) # 前7天是否促销
result.append(get_sum_of_prom(base_feat, end_date, 14)) # 前14天促销次数
result.append(get_sum_of_prom(base_feat, end_date, 28)) # 前28天促销次数
result.append(get_sum_of_prom(base_feat, end_date, 140)) # 前140天促销次数
result.append(get_sum_of_week(base_feat, end_date, 28)) # 获取前一个月的week和
result.append(get_sum_of_week(base_feat, end_date,
140)) # 获取前一个月的week和
result.append(get_sum_of_week(base_feat, end_date,
490)) # 获取前一个月的week和
result.append(get_lastyear_of_store_item(base_feat,
end_date)) # 获取去年同期的数据销量数据
result.append(get_lastday_of_store_item(base_feat, end_date)) #上次购买时间
result.append(get_promo_of_store_item(base_feat, end_date)) # 是否促销
result.append(get_lastdays_of_item(base_feat, end_date,
30)) # item 前7天的值
result.append(get_sum_of_item(base_feat, end_date, 7)) # item 前7天的和
result.append(get_sum_of_item(base_feat, end_date, 28)) # item 前7天的和
result.append(get_sum_of_item(base_feat, end_date, 70)) # item 前7天的和
result.append(get_sum_of_item(base_feat, end_date, 140)) # item 前7天的和
# result.append(get_sum_of_week_item(base_feat, end_date, 140)) # 获取前一个月的week和
result.append(get_lastyear_of_item(base_feat,
end_date)) # 获取去年同期的数据销量数据
# result.append(get_lastdays_of_store(base_feat, end_date, 7)) # store 前7天的值
# result.append(get_sum_of_week_store(base_feat, end_date, 140)) # 获取前一个月的week和
# result.append(get_lastyear_of_store(base_feat, end_date)) # 获取去年同期的数据销量数据
result.append(get_target_store(base_feat, end_date)) # 商场的销量
# 商场 品类特征
# result.append(get_sum_of_store_class(base_feat, end_date, 7)) # store×class 前7天的
# 商店 大类特征
# result.append(get_sum_of_store_family(base_feat, end_date, 7)) # store×class 前7天的
########################日期趋势信息###########################
result.append(label)
print('开始合并特征...')
result = concat(result)
result = second_feat(result)
# print('存储数据...')
# result.to_hdf(result_path, 'w', complib='blosc', complevel=5)
print('特征矩阵大小：{}'.format(result.shape))
print('生成特征一共用时{}秒'.format(time.time() - t0))
return result
def write_hdf(file_name, key, **kwargs):
pd.to_hdf(file_name, key, **kwargs)
def write_hdf(file_name, key, **kwargs):
pd.to_hdf(file_name, key, **kwargs)
def make_feats(end_date,n_day):
t0 = time.time()
key = end_date,n_day
print('数据key为：{}'.format(key))
result_path = cache_path + 'train_set_{0}.hdf'.format(end_date)
if os.path.exists(result_path) & 0:
result = pd.to_hdf(result_path, 'w', complib='blosc', complevel=5)
else:
print('添加label')
label = get_label(end_date,n_day)
print('开始构造特征...')
result = [label]
result.append(get_store_visitor_feat(label, key, 1000)) # 商店特征
result.append(get_store_visitor_feat(label, key, 60)) # 商店特征
result.append(get_store_visitor_feat(label, key, 30)) # 商店特征
result.append(get_store_exp_visitor_feat(label, key, 1000)) # 商店特征
result.append(get_store_week_feat(label, key, 1000)) # 商店×星期特征
result.append(get_store_week_feat(label, key, 60)) # 商店×星期特征
result.append(get_store_week_feat(label, key, 30)) # 商店×星期特征
result.append(get_store_week_exp_feat(label, key, 1000)) # 商店×星期指数特征
# result.append(get_area_visitor_feat(label, key, 1000)) # 地域特征
# result.append(get_area_visitor_feat(label, key, 60)) # 地域特征
# result.append(get_area_visitor_feat(label, key, 30)) # 地域特征
# result.append(get_area_exp_visitor_feat(label, key, 1000)) # 地域特征
# result.append(get_area_week_feat(label, key, 1000)) # 地域×星期特征
# result.append(get_area_week_feat(label, key, 60)) # 地域×星期特征
# result.append(get_area_week_feat(label, key, 30)) # 地域×星期特征
# result.append(get_area_week_exp_feat(label, key, 1000)) # 地域×星期指数特征
result.append(get_genre_visitor_feat(label, key, 1000)) # 品类特征
result.append(get_genre_visitor_feat(label, key, 60)) # 品类特征
result.append(get_genre_visitor_feat(label, key, 30)) # 品类特征
result.append(get_genre_exp_visitor_feat(label, key, 1000)) # 品类特征
result.append(get_genre_week_feat(label, key, 1000)) # 品类×星期特征
result.append(get_genre_week_feat(label, key, 60)) # 品类×星期特征
result.append(get_genre_week_feat(label, key, 30)) # 品类×星期特征
result.append(get_genre_week_exp_feat(label, key, 1000)) # 品类×星期指数特征
# result.append(get_area_genre_visitor_feat(label, key, 1000)) # 地域×品类特征
# result.append(get_area_genre_visitor_feat(label, key, 60)) # 地域×品类特征
# result.append(get_area_genre_visitor_feat(label, key, 30)) # 地域×品类特征
# result.append(get_area_genre_exp_visitor_feat(label, key, 1000))# 品类特征
# result.append(get_area_genre_week_feat(label, key, 1000)) # 地域×品类×星期特征
# result.append(get_area_genre_week_feat(label, key, 60)) # 地域×品类×星期特征
# result.append(get_area_genre_week_feat(label, key, 30)) # 地域×品类×星期特征
# result.append(get_area_genre_week_exp_feat(label, key, 1000)) # 地域×品类×星期指数特征
result.append(get_reserve_feat(label,key)) # air_reserve当天特征
# result.append(get_reserve_history(label,key)) # air_reserve统计特征
result.append(get_first_last_time(label,key,1000)) # 第一次和最后一次有销量的时间
result.append(label)
print('开始合并特征...')
result = concat(result)
result = second_feat(result)
# print('存储数据...')
# result.to_hdf(result_path, 'w', complib='blosc', complevel=5)
print('特征矩阵大小：{}'.format(result.shape))
print('生成特征一共用时{}秒'.format(time.time() - t0))
return result
def make_feats(end_date):
t0 = time.time()
print('time key:{}'.format(end_date))
result_path = cache_path + 'train_set_{0}.hdf'.format(end_date)
if os.path.exists(result_path) & load:
result = pd.to_hdf(result_path, 'w', complib='blosc', complevel=5)
else:
print('add label')
label = get_label(end_date)
print('make feature...')
result = []
result.append(get_sum_of_store_item(label, end_date, 140, None, holiday_no, 'holiday_no'))
result.append(
get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], holiday_no, 'holiday_no'))
result.append(get_sum_of_store_item(label, end_date, 140, None, holiday_yes, 'holiday_yes'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], holiday_yes,
'holiday_yes'))
result.append(get_sum_of_store_item(label, end_date, 140, None, week0, 'week0'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], week0, 'week0'))
result.append(get_sum_of_store_item(label, end_date, 140, None, week1, 'week1'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], week1, 'week1'))
result.append(get_sum_of_store_item(label, end_date, 140, None, week2, 'week2'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], week2, 'week2'))
result.append(get_sum_of_store_item(label, end_date, 140, None, week3, 'week3'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], week3, 'week3'))
result.append(get_sum_of_store_item(label, end_date, 140, None, week4, 'week4'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], week4, 'week4'))
result.append(get_sum_of_store_item(label, end_date, 140, None, week5, 'week5'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], week5, 'week5'))
result.append(get_sum_of_store_item(label, end_date, 140, None, weekend_no, 'weekend_no'))
result.append(
get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], weekend_no, 'weekend_no'))
result.append(get_sum_of_store_item(label, end_date, 140, None, weekend_yes, 'weekend_yes'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], weekend_yes,
'weekend_yes'))
result.append(get_sum_of_store_item(label, end_date, 140, None, onpromotion_yes, 'onpromotion_yes'))
result.append(
get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], onpromotion_yes,
'onpromotion_yes'))
result.append(get_sum_of_store_item(label, end_date, 140, None, onpromotion_no, 'onpromotion_no'))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)], onpromotion_no,
'onpromotion_no'))
result.append(get_sum_of_store_item(label, end_date, 140, [i for i in range(140)]))
result.append(get_sum_of_store_item(label, end_date, 140, [i + 140 for i in range(140)]))
result.append(get_sum_of_store_item(label, end_date, 140, [i + 280 for i in range(140)]))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 2 for i in range(140)]))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 3 for i in range(140)]))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 4 for i in range(140)]))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 5 for i in range(140)]))
result.append(get_sum_of_store_item(label, end_date, 140, [i ** 6 / 1000 for i in range(140)]))
result.append(get_sum_of_item_state(label, end_date, 7)) # 前7天同state同商品销量
result.append(get_sum_of_item_state(label, end_date, 14)) # 前14天同state同商品销量
result.append(get_sum_of_item_state(label, end_date, 28)) # 前28天同state同商品销量
result.append(get_sum_of_item_state(label, end_date, 42)) # 前42天同state同商品销量
result.append(get_sum_of_item_state(label, end_date, 70)) # 前70天同state同商品销量
result.append(get_sum_of_item_state(label, end_date, 140)) # 前140天同state同商品销量
result.append(get_sum_of_item_type(label, end_date, 7)) # 前7天同type同商品销量
result.append(get_sum_of_item_type(label, end_date, 14)) # 前14天同type同商品销量
result.append(get_sum_of_item_type(label, end_date, 28)) # 前28天同type同商品销量
result.append(get_sum_of_item_type(label, end_date, 42)) # 前42天同type同商品销量
result.append(get_sum_of_item_type(label, end_date, 70)) # 前70天同type同商品销量
result.append(get_sum_of_item_type(label, end_date, 140)) # 前140天同type同商品销量
result.append(get_sum_of_item_cluster(label, end_date, 7)) # 前7天同cluster同商品销量
result.append(get_sum_of_item_cluster(label, end_date, 14)) # 前14天同cluster同商品销量
result.append(get_sum_of_item_cluster(label, end_date, 28)) # 前28天同cluster同商品销量
result.append(get_sum_of_item_cluster(label, end_date, 42)) # 前42天同cluster同商品销量
result.append(get_sum_of_item_cluster(label, end_date, 70)) # 前70天同cluster同商品销量
result.append(get_sum_of_item_cluster(label, end_date, 140)) # 前140天同cluster同商品销量
result.append(get_sum_of_store_class(label, end_date, 7)) # 前7天同店铺同class销量
result.append(get_sum_of_store_class(label, end_date, 14)) # 前14天同店铺同class销量
result.append(get_sum_of_store_class(label, end_date, 28)) # 前28天同店铺同class销量
result.append(get_sum_of_store_class(label, end_date, 42)) # 前42天同店铺同class销量
result.append(get_sum_of_store_class(label, end_date, 70)) # 前70天同店铺同class销量
result.append(get_sum_of_store_class(label, end_date, 140)) # 前140天同店铺同class销量
result.append(get_sum_of_trans(label, end_date, 7)) # 前7天trans
result.append(get_sum_of_trans(label, end_date, 28)) # 前28天trans
result.append(get_sum_of_trans(label, end_date, 42)) # 前42天trans
result.append(get_sum_of_trans(label, end_date, 70)) # 前70天trans
result.append(get_sum_of_trans(label, end_date, 140)) # 前140天trans
result.append(get_sum_of_oil(label, end_date, 7)) # 前7天油价
result.append(get_sum_of_oil(label, end_date, 28)) # 前28天油价
result.append(get_sum_of_oil(label, end_date, 42)) # 前42天油价
result.append(get_sum_of_oil(label, end_date, 70)) # 前70天油价
result.append(get_sum_of_oil(label, end_date, 140)) # 前140天油价
result.append(get_item_info(label)) # 添加商品信息
result.append(get_store_info(label)) # 添加店铺信息
result.append(get_sum_of_store_family(label, end_date, 7)) # 前7天同店铺同family销量
result.append(get_sum_of_store_family(label, end_date, 14)) # 前14天同店铺同family销量
result.append(get_sum_of_store_family(label, end_date, 28)) # 前28天同店铺同family销量
result.append(get_sum_of_store_family(label, end_date, 42)) # 前42天同店铺同family销量
result.append(get_sum_of_store_family(label, end_date, 70)) # 前70天同店铺同family销量
result.append(get_sum_of_store_family(label, end_date, 140)) # 前140天同店铺同family销量
result.append(get_sum_of_item_city(label, end_date, 7)) # 前7天同城市同商品销量
result.append(get_sum_of_item_city(label, end_date, 14)) # 前14天同城市同商品销量
result.append(get_sum_of_item_city(label, end_date, 21)) # 前21天同城市同商品销量
result.append(get_sum_of_item_city(label, end_date, 28)) # 前28天同城市同商品销量
result.append(get_sum_of_item_city(label, end_date, 42)) # 前42天同城市同商品销量
result.append(get_sum_of_item_city(label, end_date, 70)) # 前70天同城市同商品销量
result.append(get_sum_of_item_city(label, end_date, 140)) # 前140天同城市同商品销量
result.append(get_sum_of_item(label, end_date, 7)) # 前7天商品销量
result.append(get_sum_of_item(label, end_date, 14)) # 前14天商品销量
result.append(get_sum_of_item(label, end_date, 21)) # 前21天商品销量
result.append(get_sum_of_item(label, end_date, 28)) # 前28天商品销量
result.append(get_sum_of_item(label, end_date, 42)) # 前42天商品销量
result.append(get_sum_of_item(label, end_date, 70)) # 前70天商品销量
result.append(get_sum_of_item(label, end_date, 140)) # 前140天商品销量
result.append(get_lastdays_of_st(label, end_date, 30)) # 前一月每天的值
# result.append(get_sum_of_store_item(label, end_date, 1)) # 前1天的和
result.append(get_sum_of_store_item(label, end_date, 3)) # 前3天的和
result.append(get_sum_of_store_item(label, end_date, 7)) # 前7天的和
result.append(get_sum_of_store_item(label, end_date, 14)) # 前14天的和
result.append(get_sum_of_store_item(label, end_date, 21)) # 前21天的和
result.append(get_sum_of_store_item(label, end_date, 28)) # 前28天的和
result.append(get_sum_of_store_item(label, end_date, 42)) # 前42天的和
result.append(get_sum_of_store_item(label, end_date, 70)) # 前70天的和
result.append(get_sum_of_store_item(label, end_date, 98)) # 前98天的和
result.append(get_sum_of_store_item(label, end_date, 140)) # 前140天的和
result.append(get_lastdays_of_prom(label, end_date, 16)) # 该时间段内是否促销
result.append(get_sum_of_prom(label, end_date, 14)) # 前14天促销次数
result.append(get_sum_of_prom(label, end_date, 28)) # 前28天促销次数
result.append(get_sum_of_prom(label, end_date, 35)) # 前140天促销次数
result.append(get_sum_of_prom(label, end_date, 70)) # 前140天促销次数
result.append(get_sum_of_prom(label, end_date, 280)) # 前140天促销次数
result.append(get_sum_of_prom(label, end_date, 140)) # 前140天促销次数
# result.append(get_sum_of_week(label, end_date, 140)) #获取前一个月的week和
# 上次购买时间
result.append(get_promo_of_store_item(label, end_date)) # 上次促销开始的时间和结束时间
result.append(label)
print('concat feature...')
result = concat(result).reindex()
result = second_feat(result)
# print('存储数据...')
# result.to_hdf(result_path, 'w', complib='blosc', complevel=5)
print('shape：{}'.format(result.shape))
print('used {} second'.format(time.time() - t0))
return result