时间序列分析For循环Python

Question

我正在尝试使预测 (1) 每个州的总需求和 (2) 每个州每个客户的需求的过程自动化。应用的统计方法是移动平均。预测时间为 1 个月 ahead.The 数据从具有 5 列的 excel sheet 导入：客户、州、产品、数量、订单日期。 excel 文件可以通过 link https://drive.google.com/file/d/1JlIqWl8bfyJ3Io01Zx088GIAC6rRuCa8/view?usp=sharing

找到

一个客户可以与不同的州相关联，例如，Aaron Bergman 可以从华盛顿、得克萨斯州和俄克拉荷马州的商店购买椅子、艺术品 Phone。其他客户有相同的购买行为。对于 (1) 我尝试使用 For 循环，但它没有用。错误是Order_Date not in index

df = pd.read_excel("Sales_data.xlsx")
State_Name = df.State.unique()
Customer_Name = df.Customer.unique()

for x in State_Name:
   df = df[['Order_Date', 'Quantity']]
   df['Order_Date'].min(), df['Order_Date'].max()
   df.isnull().sum()

   df.Timestamp = pd.to_datetime(df.Order_Date, format= '%D-%M-%Y %H:%m')
   df.index = df.Timestamp
   df = df.resample('MS').sum()

   rolling_mean = df.Quantity.rolling(window=10).mean()

Answer 1

考虑将 for 循环行转换为定义的方法并使用 groupby 到 return 时间序列调用它。此外，请注意 pandas:

中的最佳实践

避免referencing columns as attributes with period qualifiers。相反，使用括号 [].
避免使用 [] 列表 column subsetting。相反，使用 reindex.

def rollmean_func(df):
   # BETTER COLUMN SUBSET
   df = df.reindex(['Order_Date', 'Quantity'], axis='columns')  

   # BETTER COLUMN ASSIGNMENT
   df['Timestamp'] = pd.to_datetime(df['Order_Date'], format= '%D-%M-%Y %H:%m')  
   df.index = df['Timestamp']

   df = df.resample('MS').sum()
   rolling_mean = df['Quantity'].rolling(window=10).mean()
  
   return rolling_mean

州级

state_rollmeans = df.groupby(['State']).apply(rollmean_func)
state_rollmeans
# State      Timestamp 
# Alabama    2014-04-01     NaN
#            2014-05-01     NaN
#            2014-06-01     NaN
#            2014-07-01     NaN
#            2014-08-01     NaN
# ...
# Wisconsin  2017-09-01    10.6
#            2017-10-01     7.5
#            2017-11-01     9.7
#            2017-12-01    12.3
# Wyoming    2016-11-01     NaN
# Name: Quantity, Length: 2070, dtype: float64

客户级别

customer_rollmeans = df.groupby(['Customer_Name']).apply(rollmean_func)
customer_rollmeans
# Customer_Name       Timestamp 
# Aaron Bergman       2014-02-01    NaN
#                     2014-03-01    NaN
#                     2014-04-01    NaN
#                     2014-05-01    NaN
#                     2014-06-01    NaN
# ...
# Zuschuss Donatelli  2017-02-01    1.2
#                     2017-03-01    0.7
#                     2017-04-01    0.7
#                     2017-05-01    0.0
#                     2017-06-01    0.3
# Name: Quantity, Length: 26818, dtype: float64

时间序列分析For循环Python

Time series analysis For loop Python

python

pandas

time-series

prediction

forecasting