《Microsoft SQL Server 2008 MDX Step by Step》学习笔记七：执行聚合（上）

关注

原创最新推荐文章于 2026-04-21 11:35:04 发布 2.6k 阅读 AI 写同款· GEO 优化 ›

AI使能的SVD算子：基于深度学习的矩阵分解方法奇异值分解(Singular Value Decomposition, SVD)是线性代数中一种重要的矩阵分解方法，广泛应用于信号处理、数据降维、推荐系统等领域。传统SVD算法虽然精确，但在处理大规模矩阵时面临计算复杂度高、内存需求大等问题。本文提出了一种基于深度学习神经网络的SVD算子实现方法，完全摒弃传统数值算法，仅使用PyTorch框架构建的神经网络模型来完成大规模信道矩阵的SVD分解与还原。阅读详情

SQL Server 2008中SQL应用系列及BI笔记系列--目录索引

导读:本文介绍执行聚合（Aggregation）的进阶内容，包括：

■1、用Sum和Aggregation执行求和

■2、用Avg函数计算均值

■3、用表达式计算均值

■4、用Stedv计算标准偏差

本文所用数据库和所有源码，请到微软官网下载

1、执行求和

MDX的聚合函数：Aggregation（http://msdn.microsoft.com/zh-cn/library/ms145524.aspx）

求和还有一个函数:Sum（http://msdn.microsoft.com/zh-cn/library/ms145484.aspx）

我们依旧从一个最简单的例子看起

例7-1

SELECT
{
([Measures].[Reseller Sales Amount]),
([Measures].[Reseller Transaction Count]),
([Measures].[Reseller Order Count])
} ON COLUMNS,
TopCount(
{[Product].[Subcategory].[Subcategory].Members},
5,
([Measures].[Reseller Sales Amount])
) +
{([Product].[Subcategory].[All Products])} ON ROWS
FROM [Step-by-Step]
;

需要说明的是：上例中，[Reseller Sales Amount]代表这个表中Sales Amount字段的数量，[Reseller Transaction Count]代表底层事实表数据记录的数值，[Reseller Order Count]代表底层事实表的订单数。

查询结果如下：

邀月工作室

下面我们增加对subcategory进行TopCount 5 求和

例7-2

WITH
MEMBER [Product].[Subcategory].[Top 5] AS
Sum(
TopCount(
[Product].[Subcategory].[Subcategory].Members,
5,
([Measures].[Reseller Sales Amount])
),
([Measures].CurrentMember)
)
SELECT
{
([Measures].[Reseller Sales Amount]),
([Measures].[Reseller Transaction Count]),
([Measures].[Reseller Order Count])
} ON COLUMNS,
TopCount(
[Product].[Subcategory].[Subcategory].Members,
5,
([Measures].[Reseller Sales Amount])
) +
{
([Product].[Subcategory].[Top 5]),
([Product].[Subcategory].[All Products])
} ON ROWS
FROM [Step-by-Step]
;

结果如下：
邀月工作室

前两列是累加的，因而没有问题，最后一列Reseller Order Count不是从五个分类中累加的，而是对所有产品中的子分类的订单进行汇总。很显然，这不是我们想要的结果，此时，我们需要借助于Aggregation函数

例7-3

WITH
MEMBER [Product].[Subcategory].[Top 5] AS
Aggregate(
TopCount(
[Product].[Subcategory].[Subcategory].Members,
5,
([Measures].[Reseller Sales Amount])
),
([Measures].CurrentMember)
)
SELECT
{
([Measures].[Reseller Sales Amount]),
([Measures].[Reseller Transaction Count]),
([Measures].[Reseller Order Count])
} ON COLUMNS,
TopCount(
[Product].[Subcategory].[Subcategory].Members,
5,
([Measures].[Reseller Sales Amount])
) +
{
([Product].[Subcategory].[Top 5]),
([Product].[Subcategory].[All Products])
} ON ROWS
FROM [Step-by-Step]
;

邀月工作室

Tips:在求和时，我们通常应该使用Aggregion，而非Sum。当然，这并非绝对。

2、用AVG函数计算均值

MDX中的求均值函数为Avg（http://msdn.microsoft.com/zh-cn/library/ms146067.aspx）

我们还是从一个最简单的例子入手：

例7-4

SELECT
{([Measures].[Reseller Sales Amount])} ON COLUMNS,
{
[Date].[Calendar].[Month].[January 2003]:
[Date].[Calendar].[Month].[December 2003]
} ON ROWS
FROM [Step-by-Step]
;

邀月工作室

下面我们求均值

例7-5

WITH
MEMBER [Date].[Calendar].[CY 2003 Monthly Avg Reseller Sales] AS
Avg(
{
[Date].[Calendar].[Month].[January 2003]:
[Date].[Calendar].[Month].[December 2003]
},
[Measures].CurrentMember
)
SELECT
{([Measures].[Reseller Sales Amount])} ON COLUMNS,
{([Date].[Calendar].[CY 2003 Monthly Avg Reseller Sales])} +
{
[Date].[Calendar].[Month].[January 2003]:
[Date].[Calendar].[Month].[December 2003]
} ON ROWS
FROM [Step-by-Step]
;

结果：

邀月工作室

加上季度数据

例7-6

WITH
MEMBER [Date].[Calendar].[CY 2003 Monthly Avg Reseller Sales] AS
Avg(
{
[Date].[Calendar].[Month].[January 2003]:
[Date].[Calendar].[Month].[December 2003]
},
[Measures].CurrentMember
)
SELECT
{([Measures].[Reseller Sales Amount])} ON COLUMNS,
{([Date].[Calendar].[CY 2003 Monthly Avg Reseller Sales])} +
Hierarchize(
{
[Date].[Calendar].[Month].[January 2003]:
[Date].[Calendar].[Month].[December 2003]
} +
{
[Date].[Calendar].[Calendar Quarter].[Q1 CY 2003]:
[Date].[Calendar].[Calendar Quarter].[Q4 CY 2003]
}
) ON ROWS
FROM [Step-by-Step]
;

邀月工作室

加上季度平均：

例7-7

WITH
MEMBER [Date].[Calendar].[CY 2003 Quarterly Avg Reseller Sales] AS
Avg(
{
[Date].[Calendar].[Calendar Quarter].[Q1 CY 2003]:
[Date].[Calendar].[Calendar Quarter].[Q4 CY 2003]
},
[Measures].CurrentMember
)
MEMBER [Date].[Calendar].[CY 2003 Monthly Avg Reseller Sales] AS
Avg(
{
[Date].[Calendar].[Month].[January 2003]:
[Date].[Calendar].[Month].[December 2003]
},
[Measures].CurrentMember
)
SELECT
{([Measures].[Reseller Sales Amount])} ON COLUMNS,
{
([Date].[Calendar].[CY 2003 Monthly Avg Reseller Sales]),
([Date].[Calendar].[CY 2003 Quarterly Avg Reseller Sales])
} +
Hierarchize(
{
[Date].[Calendar].[Month].[January 2003]:
[Date].[Calendar].[Month].[December 2003]
} +
{
[Date].[Calendar].[Calendar Quarter].[Q1 CY 2003]:
[Date].[Calendar].[Calendar Quarter].[Q4 CY 2003]
}
) ON ROWS
FROM [Step-by-Step]
;

注意：AVG是一个静态函数，那么，如果是动态的表达式求均值，用什么方法?

3、用表达式计算均值

例7-8

SELECT
{
([Measures].[Reseller Sales Amount]),
([Measures].[Reseller Order Count])
} ON COLUMNS,
{
[Date].[Calendar Year].[CY 2001]:
[Date].[Calendar Year].[CY 2004]
} ON ROWS
FROM [Step-by-Step]
;

邀月工作室

下面我们加上每年的月均值，这是动态计算的：

例7-9

WITH
MEMBER [Measures].[Monthly Avg Reseller Sales Amount] AS
Avg(
EXISTING [Date].[Calendar].[Month].Members,
[Measures].[Reseller Sales Amount]
)
SELECT
{
([Measures].[Reseller Sales Amount]),
([Measures].[Reseller Order Count]),
([Measures].[Monthly Avg Reseller Sales Amount])
} ON COLUMNS,
{
[Date].[Calendar Year].[CY 2001]:
[Date].[Calendar Year].[CY 2004]
} ON ROWS
FROM [Step-by-Step]
;

邀月工作室

下面我们再加上每年的每个订单的销售均值，这也是动态计算的：

例7-10

WITH
MEMBER [Measures].[Average Reseller Sales Amount] AS
([Measures].[Reseller Sales Amount]) / ([Measures].[Reseller Order Count])
,FORMAT_STRING="Currency"
MEMBER [Measures].[Monthly Avg Reseller Sales Amount] AS
Avg(
EXISTING [Date].[Calendar].[Month].Members,
[Measures].[Reseller Sales Amount]
)
SELECT
{
([Measures].[Reseller Sales Amount]),
([Measures].[Reseller Order Count]),
([Measures].[Monthly Avg Reseller Sales Amount]),
([Average Reseller Sales Amount])
} ON COLUMNS,
{
[Date].[Calendar Year].[CY 2001]:
[Date].[Calendar Year].[CY 2004]
} ON ROWS
FROM [Step-by-Step]
;

4、计算标准偏差

MDX使用函数Stdev（http://msdn.microsoft.com/zh-cn/library/ms146068.aspx），来计算标准差，它使用无偏差总体公式，而

对应地，StdevP 函数（http://msdn.microsoft.com/zh-cn/library/ms146019.aspx）使用有偏差总体公式。

看一个复杂点的例子：

例7-11

WITH
MEMBER [Measures].[Average Reseller Sales Amount] AS
([Measures].[Reseller Sales Amount])/ ([Measures].[Reseller Transaction Count])
,FORMAT_STRING="Currency"
MEMBER [Measures].[Variance Reseller Sales Amount] AS
(
([Measures].[Squared Reseller Sales Amount]) /
(([Measures].[Reseller Transaction Count])-1)
) -
(([Measures].[Average Reseller Sales Amount])^2)
,FORMAT_STRING="Currency"
MEMBER [Measures].[StDev Reseller Sales Amount] AS
([Measures].[Variance Reseller Sales Amount])^(0.5)
,FORMAT_STRING="Currency"
SELECT
{
([Measures].[Average Reseller Sales Amount]),
([Measures].[Variance Reseller Sales Amount]),
([Measures].[StDev Reseller Sales Amount])
} ON COLUMNS,
{
[Date].[Calendar Year].[CY 2001]:
[Date].[Calendar Year].[CY 2004]
} ON ROWS
FROM [Step-by-Step]
;