Slowly altering dimension (SCD) is a knowledge warehousing idea coined by the superb Ralph Kimball. The SCD idea offers with shifting a particular set of information from one state to a different. Think about we’ve a human assets (HR) system; Stephen Jiang is a Gross sales Supervisor, managing 10 gross sales representatives in his crew. The next screenshot exhibits the pattern information:
At present, Stephen Jiang obtained his promotion to the Vice President of Gross sales function, so his crew has grown in dimension from 10 to 17. Stephen is similar individual, however his function is now modified, as proven within the following picture:
One other instance is when a buyer’s handle modifications in a gross sales system. Once more, the client is similar, however their handle is now totally different. From a knowledge warehousing standpoint, we’ve totally different choices to take care of the information relying on the enterprise necessities, main us to several types of SDCs. It’s essential to notice that the information modifications within the transactional supply techniques (in our examples, the HR system or a gross sales system). We transfer and remodel the information from the transactional techniques through extract, remodel, and cargo (ETL) processes and land it in a knowledge warehouse, the place the SCD idea kicks in. SCD is about how modifications within the supply techniques replicate the information within the information warehouse. These sorts of modifications within the supply system don’t occur fairly often therefore the time period slowly altering. Many SCD sorts have been developed over time, which is out of the scope of this submit, however on your reference, we cowl the primary three sorts as follows.
SCD sort zero (SCD 0)
With any such SCD, we ignore all modifications in a dimension. So, when an individual’s residential handle modifications within the supply system (an HR system, in our instance), we don’t change the touchdown dimension in our information warehouse. In different phrases, we ignore the modifications inside the information supply. SCD 0 is additionally known as mounted dimensions.
SCD sort 1 (SCD 1)
With an SCD 1 sort, we overwrite the outdated information with the brand new. A wonderful instance of an SCD 1 sort is when the enterprise doesn’t want the client’s outdated handle and solely must preserve the client’s present handle.
SCD sort 2 (SCD 2)
With any such SCD, we preserve the historical past of information modifications within the information warehouse when the enterprise must preserve the client’s outdated and present addresses. In an SCD 2 situation, we have to keep historical past, so we insert a brand new row of information into the information warehouse every time a transactional system modifications. Inserting a brand new row of information causes information duplications within the information warehouse, which implies that we can not use the CustomerKey column as the first key of the dimension. Therefore, we have to introduce a brand new set of columns, as follows:
- A brand new key column that ensures rows’ uniqueness within the Prospects dimension. This new key column is just an index representing every row of information saved in a knowledge warehouse dimension. The brand new secret’s a so-called surrogate key. Whereas the Surrogate Key ensures every row within the dimension is exclusive, we nonetheless want to take care of the supply system’s main key. By definition, the supply system’s main keys at the moment are known as enterprise keys or alternate keys within the information warehousing world.
- A Begin Date and an Finish Date column signify the timeframe throughout which a row of information is in its present state.
- One other column exhibits the standing of every row of information.
SCD 2 is probably the most widespread sort of SCD.
Let’s revisit our earlier instance when Stephen Jiang was promoted from Gross sales Supervisor to Vice President of Gross sales. The next screenshot exhibits the information earlier than Stephen obtained the promotion:
The EmployeeKey column is the Surrogate Key of the dimension, and the EmployeeBusinessKey column is the Enterprise Key (the first key of the client within the supply system); the Begin Date column exhibits the date Stephen Jiang began his job as North American Gross sales Supervisor, the Finish Date column has been left clean (null), and the Standing column exhibits Present. Now, let’s take a look on the information after Stephen will get the promotion, which is illustrated within the following screenshot:
Because the above picture exhibits, Stephan Jiang began his new function as Vice President of Gross sales on 13/10/2012 and completed his job as North American Gross sales Supervisor on 12/10/2012.
Let’s see what SCD 2 means in relation to information modeling in Energy BI. The primary query is: Can we implement SCD 2 instantly in Energy BI Desktop with out having a knowledge warehouse? To reply this query, we should keep in mind that we create a semantic layer when constructing a knowledge mannequin in Energy BI. In a earlier submit, I defined totally different elements of a BI resolution, together with the semantic layer. However I repeat it right here. The semantic layer, by definition, is a view of the supply information (often a knowledge warehouse), optimised for reporting and analytical functions. The semantic layer doesn’t exchange the information warehouse or one other model of the information warehouse. So the reply isn’t any, we can not implement the SCD 2 performance in Energy BI. So we both want a knowledge warehouse, or the transactional system has a mechanism to assist sustaining the historic information, equivalent to a temporal mechanism. A temporal mechanism is a characteristic that some relational database administration techniques equivalent to SQL Server supply to offer details about the information stored in a desk at any time as an alternative of preserving the present information solely. To be taught extra about temporal tables in SQL Server, verify this out.
After we load the information into the information mannequin in Energy BI Desktop, we’ve all present and historic information within the dimension tables. Due to this fact, we’ve to watch out when coping with SCDs. As an illustration, the next screenshot exhibits reseller gross sales for workers:
At a primary look, the numbers appear to be appropriate. Effectively, they might be proper; they might be flawed. It is dependent upon what the enterprise expects to see on a report. Have a look at Picture 4, which exhibits Stephen’s modifications. Stephen had some gross sales values when he was a North American Gross sales Supervisor (EmployeeKey 272). However after his promotion (EmployeeKey 277), he’s not promoting anymore. We didn’t contemplate SCD once we created the previous desk, which implies we contemplate Stephen’s gross sales values (EmployeeKey 272). However is that this what the enterprise requires? Does the enterprise count on to see all staff’ gross sales with out contemplating their standing? For extra readability, let’s add the Standing column to the desk.
What if the enterprise must solely present gross sales values just for staff when their standing is Present? In that case, we must issue the SCD into the equation and filter out Stephen’s gross sales values. Relying on the enterprise necessities, we would want so as to add the Standing column as a filter within the visualizations, whereas in different circumstances, we would want to change the measures by including the Begin Date, Finish Date, and Standing columns to filter the outcomes. The next screenshot exhibits the outcomes once we use visible filters to take out Stephen’s gross sales:
Coping with SCDs shouldn’t be at all times so simple as this. Generally, we have to make some modifications to our information mannequin.
So, do all of the above imply we can not implement any forms of SCDs in Energy BI? The reply, as at all times, is “it relies upon.” In some eventualities, we will implement an answer just like the SCD 1 performance, which I clarify in one other weblog submit. However we’re out of luck in implementing the SCD 2 performance purely in Energy BI.
Have you ever used SCDs in Energy BI, I’m curious to know concerning the challenges you confronted. So please share you ideas within the feedback part beneath.
