After learning the painful data warehouse and OLAP basics, the data warehouse was created in the boss, and the universal analytical OLAP analysis interface was designed.
So I am in my heart, "Haha, you can share." Because the function items listed in the Demand Manual have been implemented. After holding this product, after the customer's demonstration, the customer said
A few "no" (how can this, don't give it a little face? I have spent a lot of time from the illiterate of a data warehouse. Next, the customer proposed
Several requirements, to intelligently analyze the data, the specific requirements may be just like you have experienced or being experienced. Ok, let me try to do intelligence analysis. First, let's complete the association analysis. By the way, I remember that the related analysis is introduced in the principle of data mining. After a period of hard reading and experiment, the example of this shopping basket still has not brought me.
How much inspiration? I use AS, only two data mining algorithms: decision tree and cluster. I don't know how these two algorithms are associated with association analysis. The next few days,
I am looking for reading data mining and related analysis (a little wasting time, but the boss is reluctant to spend money, please ask this project, there is no way, only to touch the stone
The river is On this day, I have finally put down books in data mining, reopen the AS help document, read the MDX section, and do test in the homemade MDX query analyzer
. Suddenly, the machine is moving, why not try MDX to achieve analysis? After talking about dry, after the test, he is really successful. Here is my success step: or first talk about the structure of the data warehouse: Samplecube - DIM1 ---- Dim1hier1 ------ Dim1LV1 ------ Dim1LV2 ------... ---- DIM1HIER2 - DIM2 ---- DIM2HIER1 ------- Dim2LEV1 ------- Dim2LEV2 ---- Dim2hier2 ...-- MeasureS ---- SUM1 Next, definition association The support - credibility indicators in the analysis; use MDX's WITH clause to implement the association analysis of DIM1.DIM1LV1 and DIM2.DIM2LEV1, defined as follows: with met [measures]. [Dim1LV1SUP] AS '([DIM1 ]. [DIM1LV1] .CurrentMember, [DIM2]. [DIM2HIER1]. [All DIM2],
[Sum1]) / ([DIM1]. [DIM1HIER1]. [DIM2], [DIM2]. [DIM2HIER1]. [All DIM2], [SUM1]) 'Member [MeasureS]. [DIM2LV1SUP] AS' ([DIM2 [Dim2hier1]. [Dim2lev1] .currentmember, [dim1]. [DIM1HIER1]. [All
DIM1], [SUM1]) / ([DIM2]. [DIM2HIER1]. [All DIM2], [DIM1]. [Dim1hier1]. [All DIM1], [Sum1]) 'Member [measure]. [Confidence] AS ([DIM1]. [Dim1Ler1]. [Dim1lev1] .currentmember,
[DIM2HIER1]. [Dim2lev1] .currentmember, [sum1]) / ([dim1]) / ([DIM1]. [DIM1], [DIM2]. [Dim2hier1]. [All dim2], [Sum1]) 'Improve the analysis with SELECT statement, set the minimum support degree of DIM1LV1SUP to 5%, and the minimum support degree of DIM2LEV1SUP is 1% to achieve strong correlation analysis (ie, the credit is greater than 1) .Select {[measure]. [Sum1], [Sum1], [Sum1], Recreation], [MeasureS]. [Dim1hier1sup], [measures]. [Dim2hier1sup]} on columns, order (filter ({[DIM1]. [Dim1hier1]. [Dim1lev1] .members * [dim2]. [Dim2hier1]. [Dim2lev1] .members}, [measure]. [Dim1LV1SUP]> 0.05
And [measures]. [dim2lev1sup]> 0.01 and ([MEASURES]. [dim1lev1sup] * [measures]. [dim2lev1sup])> 1)), [Sum1], BDESC)
On Rows from Samplecube
This strong correlation analysis algorithm is achieved. The above is personal experience, if there is a better implementation method, please reply.
