Ssas OLAP-计算径流三角形、样本数据和包含的立方体（PostgreSQL/Mondrian）

现实描述： 我们有一个项目清单。每个项目都有很多账户。你可以在每个帐户上执行许多操作。我确实定义了以下维度和事实表（简化）：

现在，我想使用径流三角形方法来分析数据（可能不是真正的径流三角形，但方法是相同的）。最简单的三角形如下所示：

              Distance in Months
Project name|     1    2    3    4    5    6    7    8    9    10
-------------------------------------------------------------------------
 Project1   |     5   10   15   20   25   30   35   40   45    50
 Project2   |     7   14   21   28   35   42   49   56   63
 Project3   |     2    5    8   11   14   20   25   30
 Project4   |     0    2    5   10   18   23   40
 Project5   |     5   12   18   20   21   30

有行操作数的运行总和。以月为单位的距离显示行动日期和项目开始日期之间的距离。显然，您可以使用四分之一的距离（或距离维度中定义的任何其他时段）创建类似的三角形

您还可以为项目维度中的不同层次创建三角形，例如行业（Project1-Project3=Industry1，Project4-Project5=Industry2）：

还有一个更高级的运行三角形，您可以将运行中的操作总和除以帐户数。假设我们的项目有以下数量的账户：

Project_name number_of_accounts  
-----------------------------
Project1     100
Project2     100
Project3     100
Project4     100
Project5     200

然后我想得到以下三角形：

              Distance in Months
Project    |     1    2    3    4    5    6    7    8    9    10
------------------------------------------------------------------------
 Project1  |   .05  .01  .15  .20  .25  .30  .35  .40  .45   .50
 Project2  |    .7  .14  .21  .28  .35  .42  .49  .56  .63
 Project3  |    .2   .5   .8  .11  .14  .20  .25  .30
 Project4  |    .0   .2   .5  .10  .18  .23  .40
 Project5  |   .05  .06  .09  .10 .105  .15

当您希望在项目中的帐户数对于所有项目都不相同的情况下比较项目及其操作时，这尤其有用

问题是是否有可能在OLAP中创建这样的计算。我想我可以在项目表中使用帐户的数量，但我想不出来。另一个选项是在account维度中聚合数据。我在谷歌上也找不到任何东西，可能是因为我问错了问题

这个问题的解决方案广泛适用于许多行业，尤其是在保险和银行业。它可以在流程有较长性能窗口的任何地方使用，并且可以通过定义良好、可比较的单元批次进行跟踪

（我们使用的是PostgreSQL、Saiku，多维数据集是在Schema Workbench中定义的）

测试数据（PostgreSQL语法，如果需要其他内容，请告诉我）

样本立方体（蒙德里安）：

---

两次悬赏却没有回答，我很惊讶。我找到了一个变通解决方案——使用SQL和BIRT引擎，我现在已经接近我想要的了。我仍然希望有人能为OLAP解决这个问题

---

为了实现这一目标，我必须：

• 用于返回动态选定列的自定义函数
• 基于选定列计算径流三角形数据的SQL
• 在BIRT 2.6.1中报告，以显示结果并提供参数选择界面

动态返回列

    CREATE or replace FUNCTION bizdata.getColumns(_column1 text, _column2 text, _column3 text, _column4 text, _table text, _rqdl text)
      RETURNS TABLE(cmf1 text, cmf2 text, cmf3 text, outval numeric, rqdl text) AS $$
    BEGIN
        RETURN QUERY EXECUTE 
            'SELECT ' 
                || case when _column1 = 'None' then quote_literal('None') else quote_ident(_column1) end || '::text as cmf1,' 
                || case when _column2 = 'None' then quote_literal('None') else quote_ident(_column2) end || '::text as cmf2,' 
                || case when _column3 = 'None' then quote_literal('None') else quote_ident(_column3) end || '::text as cmf3,'   
                || quote_ident(_column4) || '::numeric as baseline,'
                || case when _rqdl = 'None' then 0::text else quote_ident(_rqdl)::text end || '::text as rqdl'  
            ' FROM '
                || 'bizdata.' || _table; 
    END;
     $$ LANGUAGE plpgsql;

Thi function takes the following as input variables:

 - _column1 - common mapping field number 1
 - _column2 - common mapping field number 2
 - _column3 - common mapping field number 3
 - _column4 - column used for aggregation (sum)
 - _table - table used for getting data
 - _rqdl - requested distance level

计算数据

Using bizdata.getColumns() function I can calculate triangle data using the following statement:


with 

params as (
    select 'cmf1'::varchar as prm_name, 'project_owner_name_short'::varchar as prm_value union all 
    select 'cmf2'::varchar as prm_name, 'project_source_name_short'::varchar as prm_value union all
    select 'cmf3'::varchar as prm_name, 'None'::varchar as prm_value union all
    select 'fact'::varchar as prm_name, 'amount'::varchar as prm_value union all    
    select 'fact_table'::varchar as prm_name, 'dwv_daily_allocation_fact'::varchar as prm_value union all       
    select 'baseline'::varchar as prm_name, 'tmp_nominal_value'::varchar as prm_value union all 
    select 'baseline_table'::varchar as prm_name, 'dw_project'::varchar as prm_value union all
    select 'rqdl'::varchar as prm_name, 'year_distance'::varchar as prm_value 
)

,baseline_data as (
    select 
        cmf1,
        cmf2,
        cmf3,
        sum(coalesce(outval,0)) as baseline
    from 
        bizdata.getColumns(
            (select prm_value from params where prm_name = 'cmf1'::text),
            (select prm_value from params where prm_name = 'cmf2'::text),
            (select prm_value from params where prm_name = 'cmf3'::text),
            (select prm_value from params where prm_name = 'baseline'::text),
            (select prm_value from params where prm_name = 'baseline_table'::text), 
            'None'
            )
    group by
        cmf1,
        cmf2,
        cmf3

)




,fact_data as (
    select 
        cmf1,
        cmf2,
        cmf3,
        rqdl::int as rqdl,
        sum(coalesce(outval,0)) as fact
    from 
        bizdata.getColumns(
            (select prm_value from params where prm_name = 'cmf1'::text),
            (select prm_value from params where prm_name = 'cmf2'::text),
            (select prm_value from params where prm_name = 'cmf3'::text),
            (select prm_value from params where prm_name = 'fact'::text),
            (select prm_value from params where prm_name = 'fact_table'::text),
            (select prm_value from params where prm_name = 'rqdl'::text)
            )
    group by
        cmf1,
        cmf2,
        cmf3,
        rqdl

)

select 
    case when cmf1 = 'None' then null else cmf1 end as cmf1,
    case when cmf2 = 'None' then null else cmf1 end as cmf,
    case when cmf3 = 'None' then null else cmf1 end as cmf1,
    rqdl,
    fact,
    baseline,
    sum(fact) over (partition by cmf1, cmf2, cmf3 order by rqdl) as cfact,
    sum(fact) over (partition by cmf1, cmf2, cmf3 order by rqdl) / baseline as cfactpct
from 
    fact_data 
    join baseline_data using (cmf1, cmf2, cmf3)

您可以看到，我最多可以使用3个分组变量（cmf1、cmf2、cmf3）并选择任何距离属性（只要该属性在dwv_daily_allocation_fact中可用）。分组变量应在基线表和事实表中都可用（以获得公共组级别）

报告

最后一步是在BIRT（2.6.1）中创建报表，其中SQL参数部分的参数被数据集参数替换并链接到报表参数。使用BIRT的人可能理解，其他人必须找到其他方法

参数选择GUI

输出报告

我仍然需要找出表的正确排序（所以历史最长的组排在第一位）

编辑：我已经计算出BIRT交叉表中的排序，现在它看起来像真正的三角形：

              Distance in Months
Project    |     1    2    3    4    5    6    7    8    9    10
------------------------------------------------------------------------
 Project1  |   .05  .01  .15  .20  .25  .30  .35  .40  .45   .50
 Project2  |    .7  .14  .21  .28  .35  .42  .49  .56  .63
 Project3  |    .2   .5   .8  .11  .14  .20  .25  .30
 Project4  |    .0   .2   .5  .10  .18  .23  .40
 Project5  |   .05  .06  .09  .10 .105  .15

---

如果您需要更详细的说明，请联系我。

我已经创建了名为

pgvint

的R包，其中可以轻松计算复古曲线（径流三角形）。该包已打开，目前仅支持PostgreSQL作为数据源

示例输出：

此外，还有一款闪亮的应用程序，可在其中以不同的布局交互显示过时的数据：

---

很高兴听到你取得了进展。我怀疑没有回复是因为你处于一个有点模糊的技术堆栈的最前沿。@MikeHoney可能是模糊的技术，但问题是一般性的-如果你知道MS technology或任何其他技术的答案，请告诉我们。我可能会在SSRS中使用RunningValue函数（报告服务）

<Schema name="RunoffTriangleSchema">
  <Cube name="RunoffTriangleCube" visible="true" cache="true" enabled="true">
    <Table name="action_fact_table" schema="public">
    </Table>
    <Dimension type="StandardDimension" visible="true" foreignKey="project_key" name="Project">
      <Hierarchy name="Project" visible="true" hasAll="true">
        <Table name="project" schema="public" alias="">
        </Table>
        <Level name="Industry" visible="true" column="industry" uniqueMembers="false">
        </Level>
        <Level name="Project Name" visible="true" column="project_name" uniqueMembers="false">
        </Level>
      </Hierarchy>
    </Dimension>
    <Dimension type="StandardDimension" visible="true" foreignKey="distance_key" name="Distance">
      <Hierarchy name="Distance" visible="true" hasAll="true">
        <Table name="distance" schema="public" alias="">
        </Table>
        <Level name="Distance In Quarters" visible="true" column="distance_in_quarters" uniqueMembers="false">
        </Level>
        <Level name="Distance In Months" visible="true" column="distance_in_months" uniqueMembers="false">
        </Level>
      </Hierarchy>
    </Dimension>
    <Dimension type="StandardDimension" visible="true" foreignKey="account_key" name="Account">
      <Hierarchy name="Account" visible="true" hasAll="true">
        <Table name="account" schema="public">
        </Table>
        <Level name="Account Key" visible="true" column="account_key" uniqueMembers="false">
        </Level>
      </Hierarchy>
    </Dimension>
    <Measure name="CountActions" column="action_id" aggregator="count" visible="true">
    </Measure>
  </Cube>
</Schema>

    CREATE or replace FUNCTION bizdata.getColumns(_column1 text, _column2 text, _column3 text, _column4 text, _table text, _rqdl text)
      RETURNS TABLE(cmf1 text, cmf2 text, cmf3 text, outval numeric, rqdl text) AS $$
    BEGIN
        RETURN QUERY EXECUTE 
            'SELECT ' 
                || case when _column1 = 'None' then quote_literal('None') else quote_ident(_column1) end || '::text as cmf1,' 
                || case when _column2 = 'None' then quote_literal('None') else quote_ident(_column2) end || '::text as cmf2,' 
                || case when _column3 = 'None' then quote_literal('None') else quote_ident(_column3) end || '::text as cmf3,'   
                || quote_ident(_column4) || '::numeric as baseline,'
                || case when _rqdl = 'None' then 0::text else quote_ident(_rqdl)::text end || '::text as rqdl'  
            ' FROM '
                || 'bizdata.' || _table; 
    END;
     $$ LANGUAGE plpgsql;

Thi function takes the following as input variables:

 - _column1 - common mapping field number 1
 - _column2 - common mapping field number 2
 - _column3 - common mapping field number 3
 - _column4 - column used for aggregation (sum)
 - _table - table used for getting data
 - _rqdl - requested distance level

Using bizdata.getColumns() function I can calculate triangle data using the following statement:


with 

params as (
    select 'cmf1'::varchar as prm_name, 'project_owner_name_short'::varchar as prm_value union all 
    select 'cmf2'::varchar as prm_name, 'project_source_name_short'::varchar as prm_value union all
    select 'cmf3'::varchar as prm_name, 'None'::varchar as prm_value union all
    select 'fact'::varchar as prm_name, 'amount'::varchar as prm_value union all    
    select 'fact_table'::varchar as prm_name, 'dwv_daily_allocation_fact'::varchar as prm_value union all       
    select 'baseline'::varchar as prm_name, 'tmp_nominal_value'::varchar as prm_value union all 
    select 'baseline_table'::varchar as prm_name, 'dw_project'::varchar as prm_value union all
    select 'rqdl'::varchar as prm_name, 'year_distance'::varchar as prm_value 
)

,baseline_data as (
    select 
        cmf1,
        cmf2,
        cmf3,
        sum(coalesce(outval,0)) as baseline
    from 
        bizdata.getColumns(
            (select prm_value from params where prm_name = 'cmf1'::text),
            (select prm_value from params where prm_name = 'cmf2'::text),
            (select prm_value from params where prm_name = 'cmf3'::text),
            (select prm_value from params where prm_name = 'baseline'::text),
            (select prm_value from params where prm_name = 'baseline_table'::text), 
            'None'
            )
    group by
        cmf1,
        cmf2,
        cmf3

)




,fact_data as (
    select 
        cmf1,
        cmf2,
        cmf3,
        rqdl::int as rqdl,
        sum(coalesce(outval,0)) as fact
    from 
        bizdata.getColumns(
            (select prm_value from params where prm_name = 'cmf1'::text),
            (select prm_value from params where prm_name = 'cmf2'::text),
            (select prm_value from params where prm_name = 'cmf3'::text),
            (select prm_value from params where prm_name = 'fact'::text),
            (select prm_value from params where prm_name = 'fact_table'::text),
            (select prm_value from params where prm_name = 'rqdl'::text)
            )
    group by
        cmf1,
        cmf2,
        cmf3,
        rqdl

)

select 
    case when cmf1 = 'None' then null else cmf1 end as cmf1,
    case when cmf2 = 'None' then null else cmf1 end as cmf,
    case when cmf3 = 'None' then null else cmf1 end as cmf1,
    rqdl,
    fact,
    baseline,
    sum(fact) over (partition by cmf1, cmf2, cmf3 order by rqdl) as cfact,
    sum(fact) over (partition by cmf1, cmf2, cmf3 order by rqdl) / baseline as cfactpct
from 
    fact_data 
    join baseline_data using (cmf1, cmf2, cmf3)