Calculating Pearson Correlation Coefficient using DAX

The original request for this calculation came from one of my blog readers who dropped me a mail asking if it possible to calculated the Pearson Correlation Coefficient (PCC or PPMCC) in his PowerPivot model. In case you wonder what the Pearson Correlation Coefficient is and how it can be calculated – as I did in the beginning –  these links What is PCC, How to calculate PCC are very helpful and also offer some examples and videos explaining everything you need to know about it. I highly recommend to read the articles before you proceed here as I will not go into the mathematical details of the calculation again in this blog which is dedicated to the DAX implementation of the PCC.

Anyway, as I know your time is precious, I will try to sum up its purpose for you: “The Pearson Correlation Coefficient calculates the correlation between two variables over a given set of items. The result is a number between -1 and 1. A value higher than 0.5 (or lower than –0.5) indicate a strong relationship whereas numbers towards 0 imply weak to no relationship.”
Pearson_Graphic
The two values we want to correlate are our axes, whereas the single dots represent our set of items. The PCC calculates the trend within this chart represented as an arrow above.

The mathematical formula that defines the Pearson Correlation Coefficient is the following:
Pearson_Formula

The PCC can be used to calculate the correlation between two measures which can be associated with the same customer. A measure can be anything here, the age of a customer, it’s sales, the number of visits, etc. but also things like sales with red products vs. sales with blue products. As you can imagine, this can be a very powerful statistical KPI for any analytical data model. To demonstrate the calculation we will try to correlate the order quantity of a customer with it’s sales amount. The order quantity will be our [MeasureX] and the sales will be our [MeasureY], and the set that we will calculate the PCC over are our customers. To make the whole calculation more I split it up into separate measures:

  1. MeasureX := SUM(‘Internet Sales’[Order Quantity])
  2. MeasureY := SUM(‘Internet Sales’[Sales Amount])

Based on these measures we can define further measures which are necessary for the calculation of our PCC. The calculations are tied to a set if items, in our case the single customers:

  1. Sum_XY := SUMX(VALUES(Customer[Customer Id]), [MeasureX] * [MeasureY])
  2. Sum_X2 := SUMX(VALUES(Customer[Customer Id]), [MeasureX] * [MeasureX])
  3. Sum_Y2 := SUMX(VALUES(Customer[Customer Id]), [MeasureY] * [MeasureY])
  4. Count_Items := DISTINCTCOUNT(Customer[Customer Id])

Now that we have calculated the various summations over our base measures, it is time to create the numerator and denominator for our final calculation:

  1. Pearson_Numerator :=
  2.     ([Count_Items] * [Sum_XY]) – ([MeasureX] * [MeasureY])
  3. Pearson_Denominator_X :=
  4.     ([Count_Items] * [Sum_X2]) – ([MeasureX] * [MeasureX])
  5. Pearson_Denominator_Y :=
  6.     ([Count_Items] * [Sum_Y2]) – ([MeasureY] * [MeasureY])
  7. Pearson_Denominator :=
  8.     SQRT([Pearson_Denominator_X] * [Pearson_Denominator_Y])

Having these helper-measures in place the final calculation for our PCC is straight forward:

  1. Pearson := DIVIDE([Pearson_Numerator], [Pearson_Denominator])

This [Pearson]-measure can then be used together with any attribute in our model – e.g. the Calendar Year in order to track the changes of the Pearson Correlation Coefficient over years:
Pearson_by_Year
For those of you who are familiar with the Adventure Works sample DB, this numbers should not be surprising. In 2005 and 2006 the Adventure Works company only sold bikes and usually a customer only buys one bike – so we have a pretty strong correlation here. However, in 2007 they also started selling Clothing and Accessories which are in general cheaper than Bikes but are sold more often.
Pearson_Sales_Categories_Years

This has impact on our Pearson-value which is very obvious in the screenshots above.

As you probably also realized, the Grand Total of our Pearson calculation cannot be directly related to the single years and may also be the complete opposite of the single values. This effect is called Simpson’s Paradox and is the expected behavior here.

[MeasuresX] and [MeasureY] can be exchanged by any other DAX measures which makes this calculation really powerful. Also, the set of items over which we want to calculated the correlation can be exchanged quite easily. Below you can download the sample Excel workbook but also a DAX query which could be used in Reporting Services or any other tool that allows execution of DAX queries.

Sample Workbook (Excel 2013): Pearson.xlsx
DAX Query: Pearson_SSRS.dax

Recursive Calculations in PowerPivot using DAX

If you have ever tried to implement a recursive calculations in DAX similar to how you would have done it back in the good old days of MDX (see here) you would probably have come up with a DAX formula similar to the one below:

  1. Sales ForeCast :=
  2. IF (
  3.     NOT ( ISBLANK ( [Sales] ) ),
  4.     [Sales],
  5.     CALCULATE (
  6.         [Sales ForeCast],
  7.         DATEADD ( 'Date'[Calendar], 1, MONTH )
  8.     ) * 1.05
  9. )

However, in DAX you would end up with the following error:

A circular dependency was detected: ‘Sales'[Sales ForeCast],’Sales'[Sales ForeCast].

This makes sense as you cannot reference a variable within its own definition – e.g. X = X + 1 cannot be defined from a mathematical point of view (at least according to my limited math skills). MDX is somehow special here where the SSAS engine takes care of this recursion by taking the IF() into account.

So where could you possible need a recursive calculation like this? In my example I will do some very basic forecasting based on monthly growth rates. I have a table with my actual sales and another table for my expected monthly growth as percentages. If I do not have any actual sales I want to use my expected monthly growth to calculate my forecast starting with my last actual sales:

GeneralLogic

This is a very common requirement for finance applications, its is very easy to achieve in pure Excel but very though to do in DAX as you probably realized on your own what finally led you here Smile

In Excel we would simply add a calculation like this and propagate it down to all rows:
ExcelFormula
(assuming column C contains your Sales, D your Planned Growth Rate and M is the column where the formula itself resides)

In order to solve this in DAX we have to completely rewrite our calculation! The general approach that we are going to use was already explained by Mosha Pasumansky some years back, but for MDX. So I adopted the logic and changed it accordingly to also work with DAX. I split down the solution into several steps:
1) find the last actual sales – April 2015 with a value of 35
2) find out with which value we have to multiply our previous months value to get the current month’s Forecast
3) calculate the natural logarithm (DAX LN()-function) of the value in step 2)
4) Sum all values from the beginning of time until the current month
5) Raise our sum-value from step 4) to the power of [e] using DAX EXP()-function
6) do some cosmetic and display our new value if no actual sales exist and take care of aggregation into higher levels

Note: The new Office 2016 Preview introduces a couple of new DAX functions, including PRODUCTX() which can be used to combine the Steps 3) to 5) into one simple formula without using any complex LN() and EXP() combinations.

Step 1:
We can use this formula to get our last sales:

  1. Last Sales :=
  2. IF (
  3.     ISBLANK (
  4.         CALCULATE (
  5.             [Sales],
  6.             DATEADD ( 'Date'[DateValue], 1, MONTH )
  7.         )
  8.     ),
  9.     [Sales],
  10.     1
  11. )

It basically checks if there are no [Sales] next month. If yes, we use the current [Sales]-value as our [Last Sales], otherwise we use a fixed value of 1 as a multiplication with 1 has no impact on the final result.

Step 2:
Get our multiplier for each month:

  1. MultiplyBy :=
  2. IF (
  3.     ISBLANK ( [Last Sales] ),
  4.     1 + [Planned GrowthRate],
  5.     [Last Sales]
  6. )

If we do not have any [Last Sales], we use our [Planned GrowthRate] to for our later multiplication/summation, otherwise take our [Last Sales]-value.

Step 3 and 4:
As we cannot use “Multiply” as our aggregation we first need to calculate the LN and sum it up from the first month to the current month:

  1. Cumulated LN :=
  2. CALCULATE (
  3.     SUMX ( VALUES ( 'Date'[Month] ), LN ( [MultiplyBy] ) ),
  4.     DATESBETWEEN (
  5.         'Date'[DateValue],
  6.         BLANK (),
  7.         MAX ( 'Date'[DateValue] )
  8.     )
  9. )

 

Step 5 and 6:
If there are no actual sales, we display our calculated Forecast:

  1. Sales ForeCast :=
  2. SUMX (
  3.     VALUES ( 'Date'[Month] ),
  4.     IF ( ISBLANK ( [Sales] ), EXP ( [Cumulated LN] ), [Sales] )
  5. )

Note that we need to use SUMX over our Months here in order to also get correct subtotals on higher levels, e.g. Years. That’s all the SUMX is necessary for, the IF itself should be self-explaining here.

 

So here is the final result – check out the last column:
FinalPivot

The calculation is flexible enough to handle missing sales. So if for example we would only have sales for January, our recursion would start there and use the [Planned GrowthRate] already to calculate the February Forecast-value:
FinalPivot2

Quite handy, isn’t it?

The sample-workbook (Excel 365) can be downloaded here: RecursiveCalculations.xlsx

Excel CUBE-Functions and MDX UniqueNames

Two weeks ago at the German SQL Server Conference 2015 I was at Peter Myer’s session about Mastering the CUBE Functions in Excel. (PS: Peter is also speaking on our upcoming SQLSaturday #374 in Vienna next week and at PASS SQLRally in Copenhagen the week after). After his session we had a further discussion about this topic and our experiences on how to use Excels CUBE-functions in order to build nice Dashboards with native Excel functionalities that also work with e.g. Excel Services. Its always great to exchange with people that share the same passion on he same topic! One thing we both agreed on that is missing currently is a way to get the MDX UniqueName of something that is selected in a slicer, filter or simply in a cell using CUBEMEMBER-function. I once used a special Cube Measure which was created in MDX Script which returned the UniqueName of a given member that was selected together with this special measure. For this to work with Excel you need to know how Excel builds the MDX when querying cube values using CUBEVALUE-function. Here is a little example:
Excel_CubeValue_Formula
This produces the following MDX query:

  1. SELECT
  2. {
  3.     (
  4.         [Measures].[Internet Sales Amount],
  5.         [Product].[Category].&[1]
  6.     )
  7. } ON 0
  8. FROM [Adventure Works]
  9. CELL PROPERTIES VALUE, FORMAT_STRING, LANGUAGE, BACK_COLOR, FORE_COLOR, FONT_FLAGS

So it basically creates a tuple that contains everything you pass into the CUBEVALUE-Function as a parameter. Knowing this we can create a calculated measure to get the MDX UniqueName of this tuple using MDX StrToTuple()- and MDX AXIS()-function:

  1. MEMBER [Measures].[Excel TupleToStr] AS (
  2. TupleToStr(axis(0).item(0))
  3. )

Replacing the [Measures].[Internet Sales Amount] of our initial CUBEVALUE-function with this new measure would return this to Excel:

  1. ([Measures].[Internet Sales Amount],[Product].[Category].&[1])

 

Ok, so far so good but nothing really useful as you need to hardcode the member’s UniqueName into the CUBEVALUE-function anyway so you already know the UniqueName.
However, this is not the case if you are dealing with Pivot Table Page Filters and/or Slicers! You can simply refer to them within the CUBEVALUE-function but you never get the UniqueName of the selected item(s). Well, at least not directly! But you can use the approach described above, using an special MDX calculated measure, to achieve this as I will demonstrate on the next pages.

Calculated measures can only be created using the Pivot Table interface but can also be used in CUBE-functions. So first thing you need to do is to create a Pivot Table and add a new MDX Calculated Measure:
Excel_Create_MDX_calculated_measure

!Caution! some weird MDX coming !Caution!


Excel_Create_MDX_calculated_measure2

You may wonder, why such a complex MDX is necessary and what it actually does. What it does is the following: Based on the example MDX query that Excel generates (as shown above) this is a universal MDX that returns the MDX UniqueName of any other member that is selected together with our measure using the CUBEVALUE-function. It also removes the UniqueName of the measure itself so the result can be used again with any other measure, e.g. [Internet Sales Amount]
The reason why it is rather complex is that Excel may group similar queries and execute them as a batch/as one query to avoid too many executions which would slow down the overall performance. So we cannot just reference the first element of our query as it may belong to any other CUBEVALUE-function. This MDX deals with all this kinds of issues.

The MDX above allows you to specify only two additional filters but it may be extended to any number of filters that you pass in to the CUBEMEMBER-function. This would be the general pattern:

  1. MID(
  2.   IIf(axis(0).item(0).count > 0 AND
  3.         NOT(axis(0).item(0).item(0).hierarchy IS [Measures]),
  4.     "," + axis(0).item(0).item(0).hierarchy.currentmember.uniquename,
  5.     "")
  6. + IIf(axis(0).item(0).count > 1 AND
  7.         NOT(axis(0).item(0).item(1).hierarchy IS [Measures]),
  8.     "," + axis(0).item(0).item(1).hierarchy.currentmember.uniquename,
  9.     "")
  10. + IIf(axis(0).item(0).count > n AND
  11.         NOT(axis(0).item(0).item(n).hierarchy IS [Measures]),
  12.     "," + axis(0).item(0).item(n).hierarchy.currentmember.uniquename,
  13.     "")
  14. , 2)

After creating this measure we can now use it in our CUBE-functions in combination with our filters and slicers:
Excel_MDX_CUBEVALUE_UniqueNames_Filter
Excel_MDX_CUBEVALUE_UniqueNames_Slicer

You may noted that I had to use CUBERANKEDMEMBER here. This is because filters and slicers always return a set and if we would pass in a set to our CUBEVALUE function a different MDX query would be generated which would not allow us to extract the single UniqueNames of the selected items using the approach above (or any other MDX I could think of). So, this approach currently only works with single selections! I hope that the Excel team will implement a native function to extract the UniqueName(s) of the selected items in the future to make this workaround obsolete!

Once we have our UniqeName(s) we can now use them in e.g. a CUBESET-function to return the Top 10 days for a given group of product (filter) and the selected year (slicer):
Excel_MDX_CUBESET_TopCount

And that’s it!

So why is this so cool?

  • It works with SSAS (multidimensional and tabular) and Power Pivot as Excel still uses MDX to query all those sources. It may also work with SAP HANA’s ODBO connector but I have not tested this yet!
  • It does not require any VBA which would not work in Excel Services – this solution does!
  • The calculation is stored within the Excel Workbook so it can be easily shared with other users!
  • There is no native Excel functionality which would allow you to create a simple Top 10 report which works with filters and slicers as shown above or any more complex dynamic report/dashboard with any dynamic filtering.

So no more to say here – Have fun creating your interactive Excel web dashboards!

Download sample Workbook: Samples.xlsx

Note: You may also rewrite any TOPCOUNT expression and use the 4th and 5h parameter of the CUBESET-function instead. This is more native and does not require as much MDX knowledge:Excel_MDX_CUBESET_TopCount_Native
However, if you are not familiar with MDX, I highly recommend to learn it before you write any advanced calculations as show above as otherwise the results might be a bit confusing in the beginning! Especially if you filter and use TOPCOUNT on the same dimension!

SSAS Dynamic Named Sets in Calculated Members

Recently at one of my customers we were doing some performance tuning on a SSAS Multidimensional cube and I came across something I have not seen before yet in my lifetime as a SSAS developer. Even the simplest select queries where running for several seconds, even if executed on warm cache! So first I thought this may be related to some SCOPE assignments overwriting values etc. Using my MdxScriptDebugger I could easily identify that none of the SCOPE assignments had any impact on the actual result of the query. However, what the result of the MdxScriptDebugger trace also revealed was that the query-time increased after a calculated member was created. Important to note here is that this specific member was NOT used in the test-query at all! Investigating into the calculated member turned out that it was using a Dynamic Named Set.

Dynamic Named Sets are usually used if you want a set to be re-evaluated in the current context opposed to Static Named Sets which are only evaluated once during creation. For example you can create a Dynamic Named Set for your Top 10 customers. Changing the Year or Country would cause a re-evaluation of the set and different customers are returned for your filter selections. These type of calculated sets is usually not a problem.

Another reason to used Dynamic Named Sets is to deal with Sub-Selects. Some client tools, foremost Microsoft Excel Pivot Tables, use Sub-Selects to deal with multiple selections on the same attribute. Lets do a little example on how to use Dynamic Named Sets here. Assume you need to calculate the average yearly sales for the selected years. This may sound very trivial at first sight but can be very tricky. In order to calculated the yearly average we first need to calculated how many years are in the currently selected:

  1. CREATE MEMBER CURRENTCUBE.[Measures].[CountYears_EXISTING] AS (
  2. COUNT(EXISTING [Date].[Calendar Year].[Calendar Year].members)
  3. );

However, this does not work if Sub-Selects are used in the query:
PivotTable_SSAS_DynamicNamedSet_Wrong
The calculated member returns “6” (the overall number of years) instead of “2” (the actually selected number of years). The issue here is that the calculation is not aware of any Sub-Select or filters within the Sub-Select as it is executed only outside of the Sub-Select.

To work around this issue you can create a Dynamic Name Set and refer to it in your calculated member:

  1. CREATE DYNAMIC SET [ExistingYears] AS {
  2. EXISTING [Date].[Calendar Year].[Calendar Year].members
  3. };
  4.  
  5. CREATE MEMBER CURRENTCUBE.[Measures].[CountYears_DynamicSet] AS (
  6. COUNT([ExistingYears])
  7. );

PivotTable_SSAS_DynamicNamedSet_Working

Now we get the correct results for our Count of Years calculation and could simply divide our Sales Amount by this value to get average yearly sales. The reason for this is that Dynamic Named Sets are also evaluated within the Sub-Select and therefore a COUNT() on it returns the correct results here.

So this technique is quite powerful and is also the only feasible workaround to deal with this kind of problem. But as I initially said, this can also cause some performance issues!

To illustrate this issue on Adventure Works simply add these two calculations to your MDX Script:

  1. CREATE DYNAMIC SET [ExpensiveSet] AS {
  2. Exists(
  3.     [Product].[Category].[Category].members,
  4.     Filter(
  5.         Generate(
  6.             Head([Date].[Calendar].[Month].MEMBERS, 30),
  7.             CrossJoin(
  8.                 {[Date].[Calendar].CURRENTMEMBER},
  9.                 Head(
  10.                     Order(
  11.                         [Customer].[Customer].[Customer].MEMBERS,
  12.                         [Measures].[Internet Sales Amount],
  13.                         BDESC),
  14.                     10000))),
  15.         [Measures].[Internet Order Quantity] > -1
  16.     ),
  17.     'Internet Sales'
  18. )
  19. };
  20.  
  21. CREATE MEMBER CURRENTCUBE.[Measures].[UnusedCalc] AS (
  22. COUNT([ExpensiveSet])
  23. );

The [ExpensiveSet] is just a Dynamic Named Set which needs some time to be evaluated and the [UnusedCalc] measure does a simple count over the [ExpensiveSet]. Having these calculations in place you can now run any query against your cube and will notice that even the simplest query now takes some seconds to execute even if the new calculated member is not used:

  1. SELECT
  2. [Measures].[Internet Sales Amount] ON 0
  3. FROM [Adventure Works]

I am quite sure that this behavior is related to how Dynamic Named Sets are evaluated and why they also work for Sub-Selects. However, I also think that calculations that are not used in a query should not impact the results and/or performance of other queries!

I just created a bug on Connect in case you want to vote it:
https://connect.microsoft.com/SQLServer/feedback/details/1049569

I know that Dynamic Named Sets in combination with calculated members is a not a widely used technique as I guess most developers are not even aware of its power. For those who are, please keep in mind that these kind of calculations get evaluated for every query which can be crucial if your Dynamic Named Set is expensive to calculate! This has also impact on meta-data queries!

Applied Basket Analysis in Power Pivot using DAX

Basket Analysis is a very common analysis especially for online shops. Most online shops make use of it to make you buy products that “Others also bought …”. Whenever you view a specific product “Product A” from the online shop, basket analysis allows the shop to show you further products that other customers bought together with “Product A”. Its basically like taking a look into other customers shopping baskets. In this blog post I will show how this can be done using Power Pivot. Alberto Ferrari already blogged about it here some time ago and showed a solution for Power Pivot v1. There is also dedicated chapter in the whitepaper The Many-to-Many Revolution 2.0 which deals with Basket Analysis, already in Power Pivot v2. Power Pivot v2 already made the formula much more readable and also much faster in terms of performance. Though, there are still some things that I would like to add.

Lets take a look at the initial data model first:Model_Old

First of all we do not want to modify this model but just extend it so that all previously created measures, calculations and, most important, the reports still work. So the only thing we do is to add our Product-tables again but with a different name. Note that I also added the Subcategory and Category tables in order to allow Basket Analysis also by the Product-Category hierarchy. As we further do not want to break anything we only use an inactive relationship to our ‘Internet Sales’ fact table.

After adding the tables the model looks as follows:Model_New

The next and actually last thing to do is to add the following calculated measure:

Sold in same Order :=
CALCULATE (
COUNTROWS ( ‘Internet Sales’ ),
CALCULATETABLE (
SUMMARIZE (
‘Internet Sales’,
‘Internet Sales’[Sales Order Number]
),
ALL ( ‘Product’ ),
USERELATIONSHIP ( ‘Internet Sales’[ProductKey], ‘Filtered Product’[ProductKey] )
)
)

(formatted using DAX Formatter)

The inner CALCULATETABLE returns a list/table of all [Sales Order Numbers] where a ‘Filtered Product’ was sold and uses this table to extend the filter on the ‘Internet Sales’ table. It is also important to use ALL(‘Product’) here otherwise we would have two filters on the same column ([ProductKey]) which would always result in an empty table. Doing a COUNTROWS finally returns all for all baskets where the filtered product was sold.
We could also change ‘Internet Sales'[Sales Order Number] to ‘Internet Sales'[CustomerKey] in order to analyze what other customers bought also in different baskets (This was done for Example 3). The whole SUMMARIZE-function could also be replaced by VALUES(‘Internet Sales'[Sales Order Number]). I used SUMMARIZE here as I had better experiences with it in terms of performance in the past, though, this may depend on your data. The calculation itself also works with all kind of columns and hierarchies, regardless whether its from table ‘Product’, ‘Filtered Product’, or any other table!

So what can we finally do with this neat formula?

1) Classic Basket Analysis – “Others also bought …”:
Result_ClassicBasketAnalysis

As we can see Hydration Packs are more often sold together with Mountain Bikes opposed to Road Bikes and Touring Bikes. We could also use a slicer on ‘Filtered Product Subcategory’=”Accessories” in order to see how often Accessories are sold together with other products. You may analyze by Color and Product Category:
Result_ByColor
As we can see people that buy black bikes are more likely to buy red helmets than blue helmets.

2) Basket Analysis Matrix:
What may also be important for us is which products are sold together the most often? This can be achieved by pulling ‘Product’ on rows and ‘Filtered Product’ on columns. By further applying conditional formatting we can identify correlations pretty easy:
Result_BasketMatrix
Water Bottles are very often sold together with Bottle Cages – well, not really a surprise. Again, you can also use all kind of hierarchies here for your analysis.
This is what the whole matrix looks like:
Result_BasketMatrixFull
The big blank section in the middle are our Bikes. This tells us that there is no customer that bought two bikes in the same order/basket.

For this analysis I used an extended version of the calculation above to filter out values where ‘Product’ = ‘Filtered Product’ as of course every product is sold within its own basket:

BasketMatrix :=
IF (
MIN ( ‘Product’[ProductKey] )
<> MIN ( ‘Filtered Product’[ProductKey] ),
[Sold in same Order]
)

3) Find Customers that have not bough a common product yet
As we now know from the above analysis which products are very often bought together we can also analyze which customers do not fall in this pattern – e.g. customers who have bough a Water Bottle but have not bought a Bottle Cage yet. Again we can extend our base-calculation to achieve this:

Not Sold to same Customer :=
IF (
NOT ( ISBLANK ( [Sum SA] ) ) && NOT ( [Sold to same Customer] ),
“Not Sold Yet”
)

The first part checks if the selected ‘Product’ was sold to the customer at all and the second part checks if the ‘Filtered Product’ was not sold to the customer yet. In that case we return “Not Sold Yet”, and otherwise  BLANK() which is the default if the third parameter is omitted. That’s the result:
Result_NotSoldToCustomer
Aaron Phillips has bought a Water Bottle but no Mountain Bottle Cage nor a Road Bottle Cage – maybe we should send him some advertisement material on Bottle Cages? Smile

 

As you can see there are a lot of analyses possible on top of that little measure that we created originally. All work with any kind of grouping or hierarchy that you may have and no change to your data model is necessary, just a little extension.

And that’s it – Basket Analysis made easy using Power Pivot and DAX!

Downloads:

Sample Workbook with all Examples: BasketAnalysis.xlsx

Error-handling in Power Query

Data is the daily bread-and-butter for any analyst. In order to provide good results you also need good data. Sometimes this data is very well prepared beforehand and you can use it as it is but it is also very common that you need to prepare and transform the data on your own. To do this Microsoft has introduced Power Query (on tool of the Power BI suite). Power Query can be used to extract, transform and load data into Excel and/or Power Pivot directly.

When using any data you usually know what the data looks like and what to expect from certain columns – e.g. a value is delivered as a number, a text contains exactly 4 characters, etc.
Though, sometimes this does not apply for all rows of that dataset and your transformation logics may cause errors because of that. In order to avoid this and still have a clean data load you need to handle those errors. Power Query offers several options to this which I will elaborate in this post.

This is the sample data I will use for the following samples:

A B C
1 4 AXI23
2 5 TZ560
NA 6 UP945

we will perform simple transformations and type casts on this table to generate some errors:

Prepare_ChangeType

Error-handling on row-level

This is the easiest way of handling errors. Whenever a transformation causes an error, we can simply remove the whole row from the result set:

RowLevel_RemoveErrors

This will generate the following result table:

A B C
1 4 AX312
2 5 TZ560

As you can see, the third row was simply removed. This is the easiest way on how to remove errors from your result set but of course this may not be what you want as those removed rows may contain other important information in other columns! Assume you want to calculate the SUM over all values in column B. As we removed the third row we also removed a value from column B and the SUM is not the same as without the error-handling (9 vs. 15)!

 

Error-handling on cell-level

As we now know that column A may result in an error, we can handle this error during our transformation steps. As “NA” could not be converted to a number we see Error as result for that column. Clicking on it gives use some more details on the error itself. To handle this error we need to create a new calculated column where we first check if the value can be converted to a number and otherwise return a default numeric value:

RowLevel_RemoveErrors_CalcColumn

The M-function that we use is “try <expressions to try> otherwise <default if error>” which is very similar to a try-catch block in C#. If the expression causes an error, the default will be used instead. Details on the try-expression can be found in the Microsoft Power Query for Excel Formula Language Specification (PDF) which can be found here and is a must read for everyone interested in the M language.

CellLevel_HandleError

We could further replace our column A by the new column A_cleaned to hide this error handling transformation.

A B C A_cleaned
1 4 AXI23 1
2 5 TZ560 2
NA 6 UP945 0

 

Error-handling on cell-level with error details

There may also be cases where it is OK to have one of this errors but you need/want to display the cause of the error so that a power user may correct the source data beforehand. Again we can use the try-function, but this time without the otherwise-clause. This will return a record-object for each row:

RowLevel_ShowErrors_CalcColumn

RowLevel_ShowErrors_Result1

After expanding the A_Try column and also the A_Try.Error column we will get all available information on the error:

RowLevel_ShowErrors_Result2

A B C A_Try.HasError A_Try.Value A_Try.Error.Reason A_Try.Error.Message A_Try.Error.Detail
1 4 AXI23 FALSE 1
2 5 TZ560 FALSE 2
6 UP945 TRUE DataFormat.Error Could not convert to Number. NA

As you can see we get quite a lot of columns here. We could e.g. use A_Try.HasError to filter out error rows (similar to error-handling on row-level) or we could use it in a calculated column to mimic error-handling on cell-level. What you want to do with all the information is up to you, but in case you don’t need it you should remove all unnecessary columns.

 

Downloads:

Power Query Error Handling Workbook: Power Query Error Handling.xlsx

SAP HANA’s Big Data Scenario with Power BI

While browsing the web for any BI related topics I recently came across this blog post about SAP HANA and how it can be used to analyze Big Data. Its actually pretty cool, SAP together with Amazon Web Services (AWS) offer a free try out of their tools for 4 hours which you can use to rebuild a predefined demo. The demo itself is very well explained and document with videos and scripts and gives some good insights on how to deal with Big Data in SAP HANA. Basically it is divided into 3 steps:
1) Load Wikipedia data (Pagehits, etc.) from Hadoop/Hive (~2GB of flatfiles)
2) Create a data mart with SAP HANA
3) Analyze results with SAP Lumira

Having done a lot recently with Power BI and its tools I asked myself if this would also be possible with Power BI? So I first did the demo on SAP HANA and afterwards I was quite sure that I could do the same also with Power BI.

And this was the initiation of this blog post where we will do the same demo but instead of SAP HANA we will use only tools of the Power BI suite. Basically we only use 3 of our Power tools:
1) Power Query to load the data
2) Power Pivot to build the “data mart”
3) Power View to analyze the data

First of all we need to load the data. The demo uses data from Wikipedia where for each year, month, day and hour the number of pagehits and bytesdownloaded are monitored per page: http://dumps.wikimedia.org/other/pagecounts-raw/

This data was then moved to a Hadoop/Hive cluster on AWS S3 store which was made public available. The transformed files can be downloaded here (~250MB per file):
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=03/000000
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=03/000001
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=03/000002
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=04/000000
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=04/000001
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=04/000002
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=05/000000
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=05/000001
http://wikipedia-pagecounts-hive-results.s3.amazonaws.com/year=2013/month=05/000002

FileList

Once we have downloaded the files we can start loading them using Power Query’s “Load from Folder” source. It is basically textdata which needs to be split into several columns first. The delimiter used is 0x0001 which in Power Query’s M-language needs to be resolved to “#(0001)”:

= Table.SplitColumn(ImportedText ,"Column1",
    Splitter.SplitTextByDelimiter("#(0001)"),
    {"Column1.1", "Column1.2", "Column1.3", "Column1.4",
     "Column1.5", "Column1.6", "Column1.7", "Column1.8"})

The columns are defined as PROJECTCODE, PAGENAME, YEAR, MONTH, DAY, HOUR, PAGEHITCOUNTFORHOUR and BYTESDOWNLOADEDFORHOUR where PROJECTCODE can be further split into language code and the real project code. There is some more logic which I will not explain in detail like error handling, data conversions, etc., which are similar to what is done in the SAP HANA demo.

This is the final M-script I came up with:

let
    Source = Folder.Files("E:\SAP\Big Data"),
    FilteredRows = Table.SelectRows(Source, each ([Extension] = "")),
    CombinedBinaries = Binary.Combine(FilteredRows[Content]),
    ImportedText = Table.FromColumns({Lines.FromBinary(CombinedBinaries)}),
    SplitColumnDelimiter = Table.SplitColumn(ImportedText ,"Column1",Splitter.SplitTextByDelimiter("#(0001)"),{"Column1.1", "Column1.2", "Column1.3", "Column1.4", "Column1.5", "Column1.6", "Column1.7", "Column1.8"}),
    ChangedType = Table.TransformColumnTypes(SplitColumnDelimiter,{{"Column1.1", type text}, {"Column1.2", type text}, {"Column1.3", type number}, {"Column1.4", type number}, {"Column1.5", type number}, {"Column1.6", type number}, {"Column1.7", type number}, {"Column1.8", type number}}),
    SplitColumnDelimiter1 = Table.SplitColumn(ChangedType,"Column1.1",Splitter.SplitTextByDelimiter("."),{"Column1.1.1", "Column1.1.2"}),
    ChangedType1 = Table.TransformColumnTypes(SplitColumnDelimiter1,{{"Column1.1.1", type text}, {"Column1.1.2", type text}}),
    ReplacedValue = Table.ReplaceValue(ChangedType1,null,"wp",Replacer.ReplaceValue,{"Column1.1.2"}),
    InsertedCustom = Table.AddColumn(ReplacedValue, "PageHitsPerHour", each try Number.From([Column1.7]) otherwise 0),
    InsertedCustom1 = Table.AddColumn(InsertedCustom, "BytesDownloadedPerHour", each try Number.From([Column1.8]) otherwise 0),
    RemovedColumns = Table.RemoveColumns(InsertedCustom1,{"Column1.7", "Column1.8"}),
    RenamedColumns = Table.RenameColumns(RemovedColumns,{{"Column1.3", "Year"}, {"Column1.4", "Month"}, {"Column1.5", "Day"}, {"Column1.6", "Hour"}, {"Column1.2", "PageName"}, {"Column1.1.1", "LanguageCode"}, {"Column1.1.2", "ProjectCode"}}),
    InsertedCustom2 = Table.AddColumn(RenamedColumns, "Date", each #date([Year],[Month],[Day])),
    ChangedType2 = Table.TransformColumnTypes(InsertedCustom2,{{"PageHitsPerHour", type number}, {"BytesDownloadedPerHour", type number}, {"Date", type date}}),
    FilteredRows1 = Table.SelectRows(ChangedType2, each [PageHitsPerHour] < 53000000)
in
    FilteredRows1

If you also did the demo on SAP HANA you found some quality issues in the data why you needed to remove rows with more than 53,000,000 PageHitsPerHour as this 5 rows mess up the whole analysis. The above M-script already handles this (last statement).

There are some more important things to mention here. First of all we are dealing with about 37M rows, so loading to datasheet will not work. Instead we need to load the data directly into Power Pivot. I had several memory issues when using 32bit Excel but after switching to a 64bit Excel everything went just fine. The import itself takes about 15 minutes which I think is OK for roughly 2GB of data and 37M rows. 

Once the data is loaded into Power Pivot we need to load some additional data which are basically our lookup/dimension tables. They can also be imported from Wikipedia, in this case directly from the web:
http://wikipedia-proj-lang-codes.s3.amazonaws.com/UniqueProjectCodes.csv
http://wikipedia-proj-lang-codes.s3.amazonaws.com/UniqueLanguageCodes.csv

Both are very simple tables with only two columns. Again we can use Power Query to import them and add them to our Power Pivot data model.

The last thing to add is a time-dimension.SAP HANA has its predefined time-dimension. In the case of Power BI I simply used a linked table in Excel that holds all necessary days or actually all days of 2013. Adding all those tables and linking them we end up with this pretty simple Power Pivot model:

PowerPivot_Diagram

This is very similar to what you create during the demo as an analytical view in SAP HANA so both approaches are very similar here.

Now that the model is set up we can do our analysis using Power View and classic Excel Pivot Tables:

PowerView_Analysis

PivotTable_Report

All together it took me about 2 hours to build the whole solution of which it took about 1h to download and process the data. The final workbook containing all the data has ~500MB which is mainly because of the PageName column which contains a high number of unique values which cannot be compressed very well. After removing the PageName column and some further tuning of the datamodel for compression I could bring the size down to 148 MB. This is quite OK – in numbers this is a compression of 12 from originally 1.74 GB

I hope you understand that I could not attach the whole workbook, instead I created a smaller workbook with only 50k rows by adding a TOP filter in the Power Query. This workbook can be downloaded below.

As I showed in this post, Power BI is capable of handling this kind of data very well. Both, in terms of data volume and also in terms of the type of data (unstructured/semi-structured data). Once the data is loaded into Power Pivot it can be analyzed just like any other data source using all Excel reporting capabilities.

Downloads:
Sample Excel Sheet: Wikidata_small.xlsx
Demo Script SAP HANA: Demo Script SAP HANA.docx
Power Query M-Script: Power Query M-Script.docx

Trigger Cube-Processing from Excel using Cube Actions

Recently I faced the requirement to give specific users of my Analysis Services Multidimensional cube the opportunity to process the cube whenever they want. There are several ways to achieve this:
– start a SQL Agent that processes the cube
– create some kind of WebService that does the processing
– create some custom .NET application that does the processing
– run some script (PowerShell, Command Line, ascmd.exe, …)

 


NOTE:
The post describes a solution which requires SSAS Server Admin rights. If you are interested in a solution which triggers an asynchronous processing without requiring SSAS Server Admin Rights please scroll down to the very bottom and download VS_Solution.zip
It basically runs the same script but in a separate thread in the context of the service account so the calling user does not need to have any specific rights on the server.


From a usability point-of-view none of the above is really convenient as all options involve some kind of external process or application. In this post I will show how to process a cube using cube actions which can be triggered from Excel Pivot Tables natively. So this is what we want to achieve:

Excel_Action

 

This requires several steps:
1) create custom .Net assembly
2) add that custom .Net assembly to your Analysis Services database
3) add an action to your cube that calls the assembly

 

Creating a Custom .Net assembly:

First of all we need to create a new “Visual C#” project of type “Class Library”. To work with Analysis Services we need to add two references:
VS_References

“Microsoft.AnalysisServices” refers to Microsoft.AnalysisServices.dll which  can be found in the shared features folder of your installation (default is c:Program FilesMicrosoft SQL Server110SDKAssemblies)

“msmgdsrv” refers to msmgdsrv.dll which can found be in the OLAPbin-folder of your SSAS instance (default is c:Program FilesMicrosoft SQL ServerMSAS11.MSSQLSERVEROLAPbin)

Once these references are added, we can start to create our processing code:

using System;
using AdomdServer = Microsoft.AnalysisServices.AdomdServer;
using Microsoft.AnalysisServices;
using System.Data;

namespace ASSP_Processing
{
    public class Processing
    {
        public const string LoggingPrefix = “ASSP.ProcessObject: “;

        public enum ResultHandling
        {
            Datatable = 0,
            Exception = 1,
            NULL = 2
        }

        public static DataTable ProcessObject(string cubeName, ResultHandling resultHandling)
        {
            DataTable ret = new DataTable();
            ret.Columns.Add(“Result”);

            Server server = null;

            try
            {
                server = new Server();

                //connect to the current session…
                server.Connect(“*”);

                AdomdServer.Context.CheckCancelled(); //could be a bit long running, so allow user to cancel

                Database db = server.Databases.GetByName(AdomdServer.Context.CurrentDatabaseName);

                string objectToProcessName = “#N/A”;
                ProcessableMajorObject objectToProcess = db;

                AdomdServer.Context.TraceEvent(0, 0, LoggingPrefix + “Database <“ + db.Name + “> found!”);
                objectToProcessName = “DB[“ + db.Name + “]”;

                if (!string.IsNullOrEmpty(cubeName))
                {
                    Cube cube = db.Cubes.GetByName(cubeName);

                    if (cube != null)
                    {
                        objectToProcess = cube;
                        AdomdServer.Context.TraceEvent(0, 0, LoggingPrefix + “Cube <“ + cubeName + “> found!”);
                        objectToProcessName = objectToProcessName + ” > Cube[“ + cubeName + “]”;
                    }
                }

                if (objectToProcess != null)
                {
                    AdomdServer.Context.TraceEvent(0, 0, LoggingPrefix + “Processing Object “ + objectToProcessName + ” …”);

                    objectToProcess.Process(ProcessType.ProcessFull);

                    ret.Rows.Add(new object[] { “S U C C E S S:    Object “ + objectToProcessName + ” successfully processed!” });
                    AdomdServer.Context.TraceEvent(0, 0, LoggingPrefix + “Finished Processing Object “ + objectToProcessName + “!”);
                }
            }
            finally
            {
                try
                {
                    if (server != null)
                        server.Disconnect();
                }
                catch { }
            }

            // if processing was successful a row has been added beforehand
            if (ret.Rows.Count == 0)
            {
                ret.Rows.Add(new object[] { “F A I L U R E:    Error while processing an object!” });
            }

            switch (resultHandling)
            {
                case ResultHandling.Datatable:
                    return ret;

                case ResultHandling.Exception:
                    throw new Exception(Environment.NewLine + Environment.NewLine + Environment.NewLine + ret.Rows[0][0].ToString());

                case ResultHandling.NULL:
                    return null;
            }

            return null;
        }

        public static DataTable ProcessObject(string cubeName, int resultHandling)
        {
            return ProcessObject(cubeName, (ResultHandling)resultHandling);
        }

        public static DataTable ProcessObject(string cubeName)
        {
            return ProcessObject(cubeName, 1);
        }
    }
}

There are two things here that are worth to point out. First of all we need to somehow establish a connection to the server that hosts the SSAS database in order to process an object. The first thing that comes into mind would be to create a new connection to the server and run some processing XMLA. Unfortunately this does not work here because as this would result in a deadlock. The assembly will be called as a Rowset-Action which in terms of locking is similar to a query. So if we run separate processing command within our Rowset-Action this processing command can never be committed as there is always a query running on the database. The query (=Rowset-Action) will wait until the function is finished and the function waits until the processing is committed resulting in a classical deadlock!

To avoid this we need to connect to the current users sessions using “*” as our connection string:

//connect to the current session…
server.Connect(“*”);

 

The second thing to point out is the return value of our function. In order to be used in a Rowset-action, the function has to return a DataTable object. The results would then be displayed similar to a drill through and a separate worksheet would pop up in Excel showing the result:

Result_DataTable

From my point-of-view this can be a bit annoying as after each processing this new worksheet pops up and you loose the focus of the pivot table. So I investigated a bit and found another way to display the output of the processing.

When a drill through / action throws an exception, this will also be displayed in Excel without actually displaying the result of the drill through / action in a separate worksheet. By default it may look like this message:

Result_NULL

Well, not really much information here right?

To display more information, e.g. that the processing was successful (or not) we can throw our own exception in the code after processing is done:

Result_Exception

I added a separate parameter to control this behavior, in the code above the default would be option 3 – custom Exception.

 

Once this project is built the DLL can be added to the SSAS database. It is necessary to grant the assembly “Unrestricted” permission set:

AssemblySettings

 

The last step would be to create the appropriate cube actions that call the assembly:

CubeActionConfiguration

It is necessary that the action is of Type=Rowset, otherwise it would not be displayed in Excel! The Action Target is up the you, in this example it will be displayed on every cell. The Action expression is the actual call to our assembly passing the cube name as a parameter. The second parameter controls how the result is displayed and could also be omitted.  In this example I added 3 separate actions, one for each result-type (as described above).

And that’s it! This simple steps allow you and your users to process any cube (or any other database object) from within your favorite client application, assuming it supports cube actions.

This is the final result where the three cube actions can be triggered from an Excel pivot table:

Excel_Action_Final

 

The attached zip-file includes everything that is necessary to recreate this example:
– SSAS project
– ASSP_Processing project

CubeProcessingAction.zip

Solution for Asynchronous Processing without Admin-Rights:
VS_Solution.zip

Universal Quantiles Calculation for PowerPivot using DAX

In my last post I showed a new approach on how to calculate the median in PowerPivot using DAX. In the comments the question was raised whether it is possible to make that calculation universal for all kind of Quantiles like Median, Quartiles, Percentiles, etc. and that’s what this post is about.

Lets analyze the final Median calculation from the last post:

Median SA Months:=CALCULATE([SumSA],
TOPN(
2-MOD([Cnt_Months], 2),
TOPN(
([Cnt_Months] + 1) / 2,
VALUES(‘Date’[Month]),
[SumSA]),
[SumSA],
1))
/
(2-MOD([Cnt_Months], 2))

 

The Median defines the value in the middle of an ordered set. To get the first half (+1 to handle even and odd sets) of the whole we are using TOPN function:

TOPN(
([Cnt_Months] + 1) / 2,
VALUES(‘Date’[Month]),
[SumSA]),

The important part here is the “divide by 2” to split the set in the middle to start our Median calculation. Instead of dividing by 2 we could also multiply by 0.5 where 0.5 would be the separator for our quantile (in this case the Median). This expression can be made dynamic. For the first Quartile we would use 0.25 for the second Quartile (=Median) we would use 0.5 and for the third Quartile we would use 0.75.

I created a little linked table to demonstrate the dynamic approach:

Quantile SortOrder Percentage
Median 1 0.5000
Quartile 1 401 0.2500
Quartile 2 402 0.5000
Quartile 3 403 0.7500
Sextile 1 601 0.1667
Sextile 2 602 0.3333
Sextile 3 603 0.5000
Sextile 4 604 0.6667
Sextile 5 605 0.8333

We also need to add a calculated measure that returns the percentage-value of the currently selected Quantile:

SelectedQuantile:=IF(
HASONEVALUE(Quantiles[Percentage]),
VALUES(Quantiles[Percentage]))

 

Now we can change our old Median-calculation to be dynamic by using the measure defined above:

Quantile SA Months:=CALCULATE([SumSA],
TOPN(
2 – MOD([Cnt_Months], 2),
TOPN(
ROUNDUP(([Cnt_Months] + 1) * [SelectedQuantile], 0),
VALUES(‘Date’[Month]),
[SumSA]),
[SumSA],
1))
/
(2 – MOD([Cnt_Months], 2))

We also need to explicitly add ROUNDUP() as “([Cnt_Months] + 1) * [SelectedQuantile]” may return any decimal places whereas the previous divide by 2 could only result in a x.5 which was rounded up automatically. And well, that’s all we need to change in order to make the calculation universal for all different Quantiles! The rest is the same logic that I already described for Median calculation.

 

Download Final Model (Office 2013!)

Calculating Median in PowerPivot using DAX

I just came across this blog post by Bill Anton where he discusses several approaches to calculated the Median of a given set in T-SQL, MDX and DAX. In the end of his post when it comes to the DAX calculation, he references several post by Marco, Alberto and Javier (post1, post2) that already address that kind of calculation in DAX. But he also claims that non of the solutions is “elegant”. Well, reason enough for me to try it on my own and here is what I came up with. Its up to you to decide whether this solution is more elegant than the others or not 🙂

In general, the median calculation always varies depending on the number of items and whether this number is even or odd.
For an even population the median is the mean of the values in the middle:
the median of {3, 5, 7, 9} is is (5 + 7)/2 = 6
For an odd population, the median is the value in the middle:
the median of {3, 5, 9} is 5

In both cases the values have to be ordered before the calculation. Note that it does not make a difference whether the values are sorted in ascending or descending order.

OrderedValues

In this example, our set contains 12 items (=months) so we have to find the 2 items in the middle of the ordered set – December and February – and calculate the mean.

So, how can we address this problem using DAX? Most of the posts I mentioned above use some kind of combination of ranking – RANKX() – and filtering – FILTER(). For my approach I will use none of these but use TOPN instead (yes, I really like that function as you probably know if you followed my blog for some time).

In this special case, TOPN() can do both, ranking and filtering for us. But first of all we need to know how many items exist in our set:

Cnt_Months:=DISTINCTCOUNT(‘Date’[Month])

 

This value will be subsequently used in our next calculations.

To find the value(s) in the middle I use TOPN() twice, first to get the first half of the items (similar to TopCount) and then a second time to get the last values that we need for our median calculation (similar to BottomCount):

TopBottom

As the median calculation is different for even and odd sets, this also has to be considered in our calculation. For both calculations MOD()-function is used to distinguish both cases:

Items_TopCount:=IF(MOD([Cnt_Months],2) = 0,
([Cnt_Months] / 2) + 1,
([Cnt_Months] + 1) / 2)

For an even number of items (e.g. 12) we simply divide the count of items by 2 and add 1 which gives us a (12 / 2) + 1 = 7 for our sample.
For an odd number of items (e.g. 5) we first add 1 to our count of items and then divide by 2 which gives us (5 + 1) / 2 = 3

Items_BottomCount:=IF(MOD([Cnt_Months],2) = 0, 2, 1)

For an even number of items we have to consider the last 2 values whereas for an odd number of items we only have to consider the last value.

 

These calculations are then used in our median calculation:

Median SA Months:=CALCULATE([SumSA],
TOPN(
[Items_BottomCount],
TOPN(
[Items_TopCount],
VALUES(‘Date’[Month]),
[SumSA]),
[SumSA] * -1))
/
[Items_BottomCount]

As DAX has no built-in BOTTOMN()-function, we need to “abuse” the TOPN() function and multiply the OrderBy-value by “–1” to get the BOTTOMN() functionality. As you can see most of the logic is already handled by our [Items_TopCount] and [Items_BottomCount] measures and this pattern can be reused very easily.

 

Of course all these calculations can also be combined and the use of IF() can be avoided:

Median SA Months v2:=CALCULATE([SumSA],
TOPN(
2 – MOD([Cnt_Months], 2),
TOPN(
([Cnt_Months] + 1) / 2,
VALUES(‘Date’[Month]),
[SumSA]),
[SumSA] * -1))
/
(2 – MOD([Cnt_Months], 2))

Note: for an even population ([Cnt_Months] + 1) / 2 returns X.5 which is automatically rounded up when it is used in a function that expects a whole number. In our example this is what happens: (12 + 1) / 2 = 6.5 –> 7

These are the final results:

FinalResults

 

Additional content:

We could also use AVERAGEX() to calculate our median but I think that it is some kind of overhead to use AVERAGEX() just to divide by “1” or “2” depending on the number of items that our TOPN-functions return:

Median SA Months AvgX:=AVERAGEX(
TOPN(
2-MOD([Cnt_Months], 2),
TOPN(
([Cnt_Months] +1) / 2,
VALUES(‘Date’[Month]),
[SumSA]),
[SumSA] * -1),
[SumSA])

 

As you can see there are various approaches to calculate the median, its up to you which on you like most. I have not tested any of them in terms of performance over bigger sets – this may be topic for an upcoming post.

 

Download Final Model (Office 2013!)