我的任务是接纳一组客户,并将每个客户分成两个相同的组。请求的结果集将使每个商店位置的两个组在客户数量方面彼此相差1%,在订单数量方面彼此相差1%,在订购数量方面彼此相差不到1%。
以下是我提出的代码,它运作得相当好,大部分时间都能得到理想的结果,但有时候(我认为由于组中的异常值),%将进一步低于1%。< / p>
If OBJECT_ID('tempdb.dbo.#Orders') IS NOT NULL DROP TABLE #Orders
Select
StoreID
,CustomerID
,Sum(OrderID) as Orders
,Sum(OrderAmount) as AmountSold
Into #Orders
From CustomerOrders
Group by StoreID,CustomerID
IF OBJECT_ID('tempdb.dbo.#OrderRanking') IS NOT NULL DROP TABLE #OrderRanking
Select
O.*
,ROW_NUMBER() Over(Partition by StoreID Order by AmountSold, Orders) as Ranking
Into #OrderRanking
From #Orders as O
Select
R.StoreID
,Count(CustomerID) as CustomerCount
,Sum(R.Orders) as Orders
,Sum(R.AmountSold) as Amountsold
,Case When Ranking%2 = 0 Then 'A' Else 'B' End as 'Grouping'
From #OrderRanking as R
Group by
R.StoreID
,Case When Ranking%2 = 0 Then 'A' Else 'B' End
是否有更好的方法来拆分组以确保1%的差异?或者也许是一种循环几种不同分裂的方法,直到找到1%?如果循环需要一个故障安全来防止无限循环,如果不可能分裂,就像在x循环后只需要最接近的分割。
我正在使用SQL Server 2012和SSMS 2016.感谢您提供的任何帮助。
编辑: 我试图将代码转换为非公司特定的东西,我搞砸了代码。我意识到并调整了代码以显示真正受到追捧的内容。
Edit2:我自己取得了一些进展,并希望更新问题。
所以我正在研究这个问题,每次运行代码时我都可以按随机顺序对其进行排序,并让它显示每个组的方差。现在我要添加的是一种循环X次的方法,并保持整体方差最小的方法。本周末我可能会尝试更多的东西。但是现在下面是我提到的新代码。
If OBJECT_ID('tempdb.dbo.#Orders') IS NOT NULL DROP TABLE #Orders
Select
StoreID
,CustomerID
,Sum(OrderID) as Orders
,Sum(OrderAmount) as AmountSold
,Rand() as Random
Into #Orders
From CustomerOrders
Group by StoreID,CustomerID
IF OBJECT_ID('tempdb.dbo.#OrderRanking') IS NOT NULL DROP TABLE #OrderRanking
Select
O.*
,ROW_NUMBER() Over(Partition by StoreID Order by Random) as Ranking
Into #OrderRanking
From #Orders as O
If OBJECT_ID('tempdb.dbo.#Split') IS NOT NULL DROP TABLE #Split
Select
R.StoreID
,Count(CustomerID) as CustomerCount
,Sum(R.Orders) as Orders
,Sum(R.AmountSold) as Amountsold
,Case When Ranking%2 = 0 Then 'A' Else 'B' End as 'Grouping'
Into #Split
From #OrderRanking as R
Group by
R.StoreID
,Case When Ranking%2 = 0 Then 'A' Else 'B' End
Select
S.StoreID
,((Cast(Max(Case When S.[Grouping] = 'A' Then S.CustomerCount Else 0 End) as decimal(18,2))-Cast(Max(Case When S.[Grouping] = 'B' Then S.CustomerCount Else 0 End) as decimal(18,2)))
/ Cast(Max(Case When S.[Grouping] = 'B' Then S.CustomerCount Else 0 End) as decimal(18,2)))*100 as CustomerCountVar
,((Cast(Max(Case When S.[Grouping] = 'A' Then S.Orders Else 0 End) as decimal(18,2))-Cast(Max(Case When S.[Grouping] = 'B' Then S.Orders Else 0 End) as decimal(18,2)))
/ Cast(Max(Case When S.[Grouping] = 'B' Then S.Orders Else 0 End) as decimal(18,2)))*100 as OrderVar
,((Cast(Max(Case When S.[Grouping] = 'A' Then S.Amountsold Else 0 End) as decimal(18,2))-Cast(Max(Case When S.[Grouping] = 'B' Then S.Amountsold Else 0 End) as decimal(18,2)))
/ Cast(Max(Case When S.[Grouping] = 'B' Then S.Amountsold Else 0 End) as decimal(18,2)))*100 as AmountsoldVar
From #Split as S
Group by S.StoreID
答案 0 :(得分:1)
所以我们所有人的预期总是在1%之内完全不可能,但就像我说的那样,在尝试X次尝试之后我们尽可能地接近。我已经想出如何实现这一目标。下面是我目前使用10次尝试设置的代码,但可以更改为适合业务的任何数字。
If OBJECT_ID('tempdb.dbo.#TestB') IS NOT NULL DROP TABLE #TestB
Create Table #TestB
(
StoreID int
,CustomerID VarChar(11)
,Orders int
,AmountSold Float
,Random Float
,Ranking bigint
,CombinedVar Decimal(18,2)
)
If OBJECT_ID('tempdb.dbo.#BestPrep') IS NOT NULL DROP TABLE #BestPrep
Create Table #BestPrep
(
StoreID int
,CustomerID VarChar(11)
,Orders int
,AmountSold Float
,Random Float
,Ranking bigint
,CombinedVar Decimal(18,2)
)
Declare @Giveup int
Set @GiveUp = 10
WHILE @GiveUp > 0
BEGIN
If OBJECT_ID('tempdb.dbo.#Orders') IS NOT NULL DROP TABLE #Orders
Select
StoreID
,CustomerID
,Sum(OrderID) as Orders
,Sum(OrderAmount) as AmountSold
,Rand() as Random
Into #Orders
From CustomerOrders
Group by StoreID,CustomerID
IF OBJECT_ID('tempdb.dbo.#OrderRanking') IS NOT NULL DROP TABLE #OrderRanking
Select
O.*
,ROW_NUMBER() Over(Partition by StoreID Order by Random) as Ranking
Into #OrderRanking
From #Orders as O
If OBJECT_ID('tempdb.dbo.#Split') IS NOT NULL DROP TABLE #Split
Select
R.StoreID
,Count(CustomerID) as CustomerCount
,Sum(R.Orders) as Orders
,Sum(R.AmountSold) as Amountsold
,Case When Ranking%2 = 0 Then 'A' Else 'B' End as 'Grouping'
Into #Split
From #OrderRanking as R
Group by
R.StoreID
,Case When Ranking%2 = 0 Then 'A' Else 'B' End
If OBJECT_ID('Tempdb.dbo.#Var') IS NOT NULL DROP TABLE #Var
Select
S.StoreID
,ABS(((Cast(Max(Case When S.[Grouping] = 'A' Then S.CustomerCount Else 0 End) as decimal(18,2))-Cast(Max(Case When S.[Grouping] = 'B' Then S.CustomerCount Else 0 End) as decimal(18,2)))
/ Cast(Max(Case When S.[Grouping] = 'B' Then S.CustomerCount Else 0 End) as decimal(18,2)))*100) as CustomerCountVar
,ABS(((Cast(Max(Case When S.[Grouping] = 'A' Then S.Orders Else 0 End) as decimal(18,2))-Cast(Max(Case When S.[Grouping] = 'B' Then S.Orders Else 0 End) as decimal(18,2)))
/ Cast(Max(Case When S.[Grouping] = 'B' Then S.Orders Else 0 End) as decimal(18,2)))*100) as OrderVar
,ABS(((Cast(Max(Case When S.[Grouping] = 'A' Then S.Amountsold Else 0 End) as decimal(18,2))-Cast(Max(Case When S.[Grouping] = 'B' Then S.Amountsold Else 0 End) as decimal(18,2)))
/ Cast(Max(Case When S.[Grouping] = 'B' Then S.Amountsold Else 0 End) as decimal(18,2)))*100) as AmountsoldVar
,ABS(((Cast(Max(Case When S.[Grouping] = 'A' Then S.Orders Else 0 End) as decimal(18,2))-Cast(Max(Case When S.[Grouping] = 'B' Then S.Orders Else 0 End) as decimal(18,2)))
/ Cast(Max(Case When S.[Grouping] = 'B' Then S.Orders Else 0 End) as decimal(18,2)))*100)
+
ABS(((Cast(Max(Case When S.[Grouping] = 'A' Then S.Amountsold Else 0 End) as decimal(18,2))-Cast(Max(Case When S.[Grouping] = 'B' Then S.Amountsold Else 0 End) as decimal(18,2)))
/ Cast(Max(Case When S.[Grouping] = 'B' Then S.Amountsold Else 0 End) as decimal(18,2)))*100) as CombinedVar
INTO #Var
From #Split as S
Group by S.StoreID
If Exists (Select * From #Var Where (OrderVar < 1 and AmountSoldVar <1) Or CombinedVar < 2)
If Object_ID('tempdb.dbo.#TestA') IS NOT NULL DROP TABLE #TestA
Select
A.StoreID
,A.CustomerID
,A.Orders
,A.AmountSold
,A.Random
,A.Ranking
,V.CombinedVar
Into #TestA
From #OrderRanking as A
Join #var as V
on A.StoreID = V.StoreID
Where A.StoreID in
(Select StoreID From #Var Where (OrderVar < 1 and AmountSoldVar <1) Or CombinedVar < 2)
Insert Into #TestB
Select
A.StoreID
,A.CustomerID
,A.Orders
,A.AmountSold
,A.Random
,A.Ranking
,A.CombinedVar
From #TestA as A
Left Join #TestB as B
on A.CustomerID = B.CustomerID
Where
B.CustomerID is null
Insert Into #BestPrep
Select
A.StoreID
,A.CustomerID
,A.Orders
,A.AmountSold
,A.Random
,A.Ranking
,V.CombinedVar
From #OrderRanking as A
Join #Var as V
on A.StoreID = V.StoreID
Left Join #BestPrep as B
on A.CustomerID = B.CustomerID
and V.CombinedVar > B.CombinedVar
Where
B.CustomerID is null
Set @Giveup = @Giveup-1
END
If Object_ID('tempdb.dbo.#bestPrep2') IS NOT NULL DROP TABLE #bestPrep2
Select
A.StoreID
,Min(CombinedVar) as CombinedVar
Into #BestPrep2
From #BestPrep as A
Group by
A.StoreID
Select A.*
From #BestPrep as A
Join #BestPrep2 as B
on A.StoreID = B.StoreID
and A.CombinedVar = B.CombinedVar
Union
Select * From #TestB