Deriving Columns from What You Know in Power BI

Before I get started with the topic of the day, I want to remind you that Power BI is still being updated by Microsoft. In fact, there was an update just last week on October 20th that appears to have added some missing features that I mentioned before. So if you haven’t updated Power BI recently, be sure to do that before reading the rest of this blog.

A few weeks ago when I was looking at defining relations between tables in the Relationship dialog I was disappointed in the fact that when I display my tables in what I would have previously called a Dialog view in PowerPivot, I could not use drag and drop to define my relations. Well, my disappointment is over. The latest update, among other things, adds this capability.

Since last Wednesday, October 21, was ‘Back to the Future’ day, I want to go back to last week’s blog just after I added the DimProductCategory table. This time however, rather than using the Management Relationships dialog, I am just going to the Relationships view and drag the field ProductCategoryLabel to ProductCategoryKey as shown in the following image.

Now I must say that I am use to just identifying the fields from my two tables that I want to connect to form the relationship and expect PowerPivot to figure out which is the one and which is the many table. Unfortunately, Power Bi still is not quite this smart. Because when I attempt to define this relationship from the one site to the many side, the relationship definition fails as shown in the following figure.

Maybe the next update will automatically reverse the direction of the relationship for me. However, for now I can click the OK button in the message box shown above and then redefine the relationship correctly from the DimProductSubcategory to the DimProductCategory table.

Now the relation is created and I am ready to continue. By the way, when I click on the relationship, Power BI highlights the two tables as well as the relationship line. It also encloses the connecting fields in a box to make it easy to identify the relationship fields.

If instead of left clicking on the relationship, suppose I right click on the relationship. Now I get an expected option menu which in this case consists of only a single option, Delete. Clicking this option will of course delete the relationship.

On the other hand, double clicking on the relationship opens the Edit Relationship dialog shown below. I can use this dialog to view the relationship or a sampling of the data, or make changes to the relationship.

That’s great. But like an infomercial, there’s more! In the Relationship view (Diagram view) I can also now right click on a field and delete a field, hide a field from the Report view or rename the field.

As with PowerPivot, if I delete a field, it is gone for good. If I realize later that I need the field, I would need to delete the table and reload it to get the missing field. Of course there are other consequences to doing this like needing to redefine the relationships and possibly rebuilding reports (visualizations) that used that table. Because at this time, I cannot visually define the fields to include or exclude during my original data load from my data sources, it is important that I know my data and delete any fields that I know that I will not need before I begin creating new columns, measures, or reports. In PowerPivot, we called these columns Useless columns.

I can also hide some columns from the Report view. For example, I can often hide columns used to define relationships between tables because end users typically do not include these columns in reports. Hidden columns are referred to as Technical columns because they are required in the data model and cannot be deleted without destroying relationships or perhaps calculations of other fields. I never want to show more columns to a user than they know what to do with.

Finally, renaming columns can be very beneficial. Often column names in databases have cryptic or abbreviated names. End users may not be comfortable with these shorten names. Use the Rename feature to make names user-friendly and descriptive.

Not only can I delete, hide, or rename columns in a table, but by right clicking on the table header, I can perform these same actions on an entire table. For example, I may not like dimension tables that begin with the letters ‘Dim’ like many DBAs prefer but end users may have no idea why the table is Dim. Similar to columns, I don’t delete tables from my model unless I am positive that I do not need them. Hiding tables only makes sense if I want to use only one or two fields from a table. In this case, I may need a column from another table for a calculation, but I would never display those columns directly in reports.

Returning to the data view of my tables, I can also right click on any of the column headers and delete, hide or rename the column. There are also several other options ranging from sorting, to creating new columns or new measures.

I can also right click on the column names in the fields list along the right side of the Data view. The dropdown list of options shown below is similar to options in the context menu above. So I have several different ways to manage columns

Let’s try something new, a new column in fact. In my FactSales table, I can find several columns like SalesAmount, TotalCost, and several others, but there is not Total Profit column. I can easily calculate that value from other values in the table. To begin, I need to create a new column in my data model. That means clicking the New Column button in the Modeling ribbon.

New columns are created at the right end of the table. By default, the column name is cleverly called Column. Of course I can change that by entering a new name. Then after an equal sign which indicates that an expression will be used to define the column, I can begin entering the column definition using DAX expressions. Yes, DAX is still alive and well. If you need a review of using DAX, I’ve covered multiple DAX topics over the last couple of years of this blog.

As in PowerPivot, I can select a column from the current table by just typing the left square bracket. This action opens a dropdown of all column names in the current table listed alphabetically. I can scroll down through the list and select a column by double clicking on its name. I can also type a few characters of the column name to narrow down my list as shown below.

My full expression to calculate TotalProfit is shown below

When I click the Enter key, the entered expression is used to calculate the values for all the table rows.

Before moving off the column, I might want to define custom formatting for the values. The formatting definition here is carried forward to all reports generated with the data. In the above example, I might want to only display the dollar amounts to two decimal places. (Actually, because of the size of aggregated data, I might later decide to format the values with no decimal places.)

Suppose that I want to display some of this data using a standard table with Channel names for the rows and a few select columns from the table. Notice that the values displayed here obey the formatting definition set on the Data page.

That’s it for this week. Next time I cover creating New Measures and why you might want to do so.


Can You FILTER() That Down For Me

The last several weeks I have been looking at how PowerPivot in Excel works with Row Context and Filtered Context. I showed that most column expressions use a row context while measures use a filtered context although I could add and remove filters using certain expressions that allowed me to define a filter as a Boolean expression in one of the parameters. Last time we even looked at how to remove the filters by using the ALL() function. This time I will explore the FILTER() function which allows me to define a permanent filter condition to a measure no matter what dimensions or slicers the user chooses for the pivot table. In fact, in the case I am going to show you today, I need to do this because I need one measure to use all filters defined by the dimensions in the pivot table, and I need another measure to obey those filters plus one more.

Again I will use my Contoso data model that I’ve been using for all the examples in this set. I want to look at the number of orders that have returns and compare that to the total number of orders. I initially will want to show this information by sales channel and year/month. However, once I have my pivot table defined, I could of course change the dimensions I want to explore.

Let’s begin with a basic Sales pivot table as shown below.

I built this table using my basic data model with no additional calculated columns or measures except the calculated column in the date dimension that I use to order the name of the months correctly. I can use any of the columns in the FactSales table as my value field as long as I change the aggregate function from SUM to COUNT. By default, Pivot tables assume that numeric fields are summed and non-numeric fields are counted. But as long as I change the aggregate function for numeric fields to COUNT, I will get my desired results. I also modified the formatting to get rid of any decimal places and to add a thousands separator. Other than that, I did nothing special to build this table.

However, now I am going to return to the FactSales table and add a simple measure to count the total number of sales. The expression I will use is shown in the following figure.

I use the COUNT() function which has a single parameter, the name of the column I want to count. Again I could choose any column, but I chose the column [ReturnQuantity]. I will come back to format this measure in a moment, but you can see that the count is a little over two and a quarter million sales records. In fact, I know that this is correct by simply looking at the number of rows in my FactSales table.

Next, I want to count the number of sales records that have returns. This I can do by comparing either the [ReturnQuantity] column or the [ReturnAmount] column to 0. Only sales records which have values greater than 0 for these two columns represent orders which had returns. How can I do this?

One way I could do this is to use the SUMX() functionwith a second measure named [ReturnCount2]. This function has two parameters. The first parameter must be a table and second parameter is an expression of what I want to count. So I might think that I could do something like the following expression:

ReturnCount2:=SUMX(FactSales,IF(FactSales[ReturnAmount]>0, 1, 0))

The theory is that I want to compare the column [ReturnAmount] to 0 and if it is greater than zero to add one to my ReturnCount2 value. I cannot simply sum the [ReturnAmount] because this column represents the dollar value of the return. Nor can I use [ReturnQuantity] because the buyer may have returned more than one of the item from the order and summing the quantity would over count the total number of orders with returns.

I could also use the COUNTX() function. However, if I simply replace SUMX() with COUNTX(), I will get the total number of orders in the FactSales table because COUNTX() will could all non-blank rows. But I can trick the IF() into returning a blank for orders without returns by using the following expression:

ReturnCount2:=COUNTX(FactSales,IF(FactSales[ReturnAmount]>0, 1, BLANK()))

But both of these solutions used the entire FactSales table. There is one other way I want to show you today. I can use the FILTER() function to apply a filter to the FactSales table to return a subtable that only has rows with returns by using the following expression to return a table

FILTER(FactSales, FactSales[ReturnAmount]>0.0)

I can now replace the first parameter in COUNTX() with this FILTER() result which is a filtered table. I can then use any column in FactSales as the column I want to count. Well, almost any column. Actually, I cannot reuse the [ReturnAmount] column which is used in the FILTER() expression because this confuses DAX, but as I said before, I can count on any column in the table. Therefore, my [ReturnCount2] measure expression is shown below.

In this image you can see that I already formatted my measures as numbers without decimals but with thousands separators. Why do I format the numbers here? Simply because it saves time from having to format the numbers in each pivot table in which I use the measures. If I display these two measures in my pivot table side by side, I can see the total number of order by channel in each month along with the number of orders that had returns.

Suppose I wanted to show this information to management and rather than look at the raw counts which could take a bit of time to interpret, I decide to calculate the percent of orders that have returns. I can create a third measure as shown in the following figure that uses the results of the first two measures. I can then format this measure as a percentage prior to using it in my pivot table.

Returning to my pivot table, I remove the counts which I no longer need to display and replace them with the [Percent_Returns] so that management can quickly see that Catalog sales result in the most returns and Store sales in the least returns. Returns do not vary greatly by month, something that I will leave up to you to explore with a Pivot Chart.

Well, I hoped you learned some new ways to apply different filters in your measures from this discussion. C’ya next time.

Getting the Right Context – Part 2

Last time, I introduced the concept of context within DAX expressions using by PowerPivot to calculate columns and measures. We saw that the default context for column calculations was row context while the default context for measures was filtered context. However, I ended the discussion by showing that within a measure calculation, I could use column context with certain aggregate functions like SUMX which can be used to apply an expression across all the rows of a table. However, before that expression is evaluated, that table is automatically filtered since I used it in a measure which begins by applying the filtered context of the pivot table to the rows used by the expression.

So this time, let’s dive a little deeper. Let me begin by going to the FactSales table of my Contoso data model and calculate a measure for total sales. This is not difficult and can be achieved by using the SUM() function as shown in the following figure.

Now let’s assume that I want a measure that shows only the sales made through the store channel. At first, I might try to use the SUMX() function along with the FILTER() function. The FILTER() function also has two parameters. The first parameter is the name of the table that I want to apply a filter on. The second parameter is a filter expression. I may at first assume that because I created a relationship between the FactSales table and the DimChannel table, that I can simply reference the column [ChannelName] and compare it to the string “Store” to filter the FactSales table. However, as you can see in the following figure, this expression would result in an error.

The reason for the error is that while the FILTER() function references the table FactSales, there is no context to link records in FactSales to DimChannel. I know you might ask, “Doesn’t the relation between these two tables define that context?” The answer is that the relationship between tables while defining the ‘mechanism’ of how to connect the two tables, it does not activate a context between the rows in FactSales with a row in DimChannel. When pointing from the many side of a relationship to the one side of the relationship, we must use the RELATED() function to activate the context within the expression. I show this in the following figure.

You can see that now I have a total sales for the just the stores as a measure. If I use this measure in a pivot table that displays sales by month and by product category, I will have the additional filter of sales by store in each of the pivot table cells.

What if, however, I wanted to create a calculated column in the table DimChannel that displayed the total sales for that channel? Again you might start with the SUMX() function because you want to calculate an expression from another table. In this case, the first parameter of the SUMX() function would be the FactSales table and the column that we want to sum would be the [SalesAmount] column. However, if we were to create this column, we might be surprised by the result, shown in the following figure.

All of the values in the column are exactly the same. Furthermore, if I refer to the image earlier in this blog for the total sales across all channels, I would see that the value displayed here in each cell of the column is actually the total sales. Again the problem is context. There is no context to refer back to FactSales from DimChannel. Therefore, when SUMX() evaluates the [SalesAmount] in table FactSales, it pulls values from all the sales records, not just the sales from the channel represented by the current row in DimChannel.

In this case, because I am going from the one side of the DimChannel relationship to the many side of FactSales, I need to return a table that contains just the rows from FactSales that represent sales from the channel in the context of the current row. I can do this by using the RELATEDTABLE() function which uses a single parameter, the name of the table on the many side of the relationship. I must also have the relationship explicitly defined between DimChannel and FactSales. I have already done this. So Power Pivot can use the relationship to create a subset of rows from FactSales for the current channel. I can then use this resulting table in the SUMX() function to sum the [SalesAmount] column as shown in the following figure.

As you can see in the figure, the first row which represents the store channel displays the same sales total as we calculated from the measure in FactSales earlier.

So again, you can see that there are many different ways to define the context of an expression. When dealing with multiple tables, it is important to understand whether you can perform a row context calculation by using the RELATED() function to extend the row context to the related table on the lookup side of the relationship or whether you need to use the RELATEDTABLE() function to filter the rows used in an aggregate function like SUMX() to calculate values for a column which uses the row context to define the link to the many side table.

Next time, I’ll look at some functions that let you turn off a filtered context and show where you might use it.

C’ya next time.

Can You Put That Into Context For Me?

Over the past several years, I’ve conducted many sessions on how to use DAX in PowerPivot to create columns and measures. Throughout this time, the one thing that has confused many people with whom I have talked with is the concept of context for an expression. Many do not understand why some equations appear to work across for defining new columns while other equations only appear to work when calculating measures within a pivot table. This week, and perhaps for another week or two, I am going to take a shot at explaining the concept of context. Until you get a good understanding of context, many of the DAX expressions for columns and measures may appear to be magic that somehow get the correct answer but you may just not understand why.

Let me start by defining the two major context types that DAX expressions have to deal with:

  1. Row Context – This is probably the easiest context to understand. When you define a formula to calculate the value for a new column, the formula is applied to each row individually. In fact, the value calculated can depend only on the value of other columns in the current row. This is much like the way you calculate a new column value in Excel. However, unlike Excel, it is more restrictive. Using Row Context, it is not possible to directly access the values in a prior or successive row because those terms do not really have meaning within a table in the data model of a pivot table.
  2. Filtered Context – This is probably easiest to understand when you think of the entire set of rows in a table. If you sum a numeric column or simply count any column in the table, you are applying a filtered context. Of course, in its simplest form, the filter is really the same as all of the rows in the table, or no filter at all. But let’s assume for a moment that you have a sales table that includes sales from each of your stores in each of the states of United States. Furthermore, you only want to know the sales for stores in the state of Florida. In database terms, you apply a filter using a WHERE clause to include only those rows from the sales table in which the sales came from a store in Florida. Of course you can have more than one filter. You might filter not only on the state, but also on a specific product that was sold during a specific month of a specific year. That would require four filters to be applied to the table to reduce the total number of rows of sales to only those you wanted to sum. Filtered context is often used in expressions that involve summing, counting, averaging, finding a minimum or finding a maximum. These are also referred to as aggregate functions.

So let’s see how this works in DAX. Using my Contoso sales table, suppose I want to calculate the profit for each sale. Since I want to perform the calculation per sale and because my sales table has one row for each sale, I can use a calculated column. The formula for my calculated column begins by taking the value of the column [SalesAmount] which is the money received from the sale. This is not the profit. To calculate profit, I need to subtract costs from [SalesAmount]. To do this, I can subtract the value from the column [TotalCost]. This gives me a pretty good idea of what my potential profit might be. However, in the Contoso data, I also have to consider any returned item. These are items returned because they are damaged or do not work. The policy is to refund the customer their money. However, the returned product it not worth anything and is considered trash. Therefore, I need to also subtract from [SalesAmount] the [ReturnAmount]. If I define a new column in my sales table, I can use the formula shown in the following figure to calculate the profit for each of the sales.

As you know, the PowerPivot data model applies the column formula to each record in the table. I can then use this column in pivot tables to display profit for sales by any of the appropriate dimensions such as time, location, product, etc.

Fortunately, in PowerPivot for Excel, displaying profit in the pivot table is as simple as moving the column [Profit] to the Values area in the field list. PowerPivot lets us use any numeric field in the values area without any additional work. However, if I attempt to create the same model in Analysis Services, I must create a measure as shown below.

The SUM() function is an aggregate function and therefore defines a measure, never a column. Note however, I must still first create the column [Profit] as previously defined. I cannot simply create a measure that sums the result of the expression as shown in the following figure.

This is because the calculation: [SalesAmount] – [TotalCost] – [ReturnAmount] is a row context expression since it must be calculated on every row of the table. However the SUM() function is a filter context expression (even if the filter is all the rows). I cannot combine the two in this expression without getting the Semantic Error shown above. The SUM() function can only aggregate the values of a single column, not aggregate the values of a calculation.

What I can do however, is use one of the other DAX functions, SUMX(). This function has two parameters, the first parameter is the name of the table I want to perform a calculation on. In this case, the table name is FactSales. The second parameter is an expression that can include one or more columns from the designated table. In this case, I can use the expression I previously used to calculate the [Profit] column as shown in the following figure.

As you can see in this figure, the resulting total profit across all the records in the FactSales table is exactly the same whether I use the expression SUM([Profit]) which sums the calculated column or if I use the expressions SUMX(FactSales, [SalesAmount] – [TotalCost] – [ReturnAmount]). This second expression essentially calculates the row context expression for profit on each row in the table FactSales. However, as a measure rather a calculated column, the expression is filtered by any dimensions used in the pivot table.

In the following figure, the only dimension is sales channel. Therefore, with four sales channels, there are four rows. Note that both calculations of total profit result in the same values for each channel.

What is the difference between the two? Since all of the sales records were used to build this table since all four sales channels were included, the same number of calculations were effectively performed in both cases, calculating the profit for each sales row and then summing the profit for each channel. The only real difference is that the first one required the prior creation of a [Profit] column which takes some of the precious data model memory while the second method using the SUMX() function does not require that column.

But what if we applied other dimensions any perhaps by using a slicer displayed the profit from only some of the sales. In the following example, the row dimension displays countries. However, only countries in Asia appear based on a slicer that limits which continent I want to see profits. I also included a slicer on year to limit the years in my analysis to only two years. Therefore, in the following figure, the profit amounts represent only a subset of the total profits for the Contoso organization.

How does this affect the calculation? To calculate Total_Profit, I still need the column [Profit] to calculate value for every sales row in the FactSales table, even rows that I will not use. On the other hand, to calculate Total_Profit_2, the filter on continent and year is applied first limited the total number of rows that are needed to be calculated from the FactSales table. The calculation of profit is then only performed on this filtered subset of rows and then aggregated appropriately based on the dimensions shown.

Thus, you might say for now that column calculations are generally row context and measures are filter context expressions. However some functions when used in measures let you perform a row context expression after applying a filter context like the SUMX() function and other aggregate functions that end in ‘X’ like COUNTX(), AVERAGEX(), MAXX(), and MINX().

Next time, I will dive a little further into understanding context within DAX.

C’ya next time.

It’s a Fiscal Thing

In past blog articles, I’ve talked about the need for a date table, not just for BI/PowerPivot analysis, but for any application. You might be hard pressed to come up with many applications you have written or have used in your organizations that do not include some aspect of time as one of the dimensions by which you collect data and/or report on it. Therefore, building a Date table once that you can use for many different applications can save you a great deal of time when trying to determine how to group your data for different reporting period or to display different date labels along one of your dimensions.

One of the easiest ways to build a date table that you can use anywhere is to begin with Excel. It is a simple matter to create a column called DateKey and enter the first two dates of a large general purpose date range as shown in the following figure.

Then to create any size table, simply select the first two cells and drag the bottom right corner down to any number of rows you want. Given Excel’s limit of a million rows, this should cover you for most applications.

Once you have the first column, you can add columns for other ways of displaying date information. The following table shows a few of the more common expressions I have used to create additional columns.

<insert table here>

If you use a standard calendar, January 1 through December 31, the functions provided work well to calculate the names of months, days of the week, quarters, years, etc. However, if your organization uses a fiscal calendar, some of these calculations are not as easy. For example, suppose your organization uses a fiscal year that begins October 1 and goes through September 30th. The calculation for month number is not as simple as using the MONTH() function. Nor is the calculation of year simple either because the fiscal year for 2015 may in that case go from October 1, 2014 through September 30, 2015. So I thought you might like to see a few functions that could help you calculate fiscal columns in your data table.

Let’s start with the calculation of the fiscal year. Using the example date range above, how can I calculated the fiscal year from any given date. If I were to add 3 months to any date and then calculate the year using the YEAR() function, I would get the fiscal year. Why three months? Because there are three months in the prior calendar year. If my fiscal year started July 1, 2014 and went to June 30, 2015, I would add 6 months to the date before calculating the fiscal year using the YEAR() function. The following figure shows the column expression I need to create a CompanyFiscalYear column.

Calculating the year was not that difficult. However, calculating the month number of the fiscal year is a bit more complex. Why do I need a fiscal month number? Remember how I used the month number column as the sort by column for the month names so that the months appear in the correct order. I again have a similar problem when I am dealing with fiscal calendars. In the above example fiscal year beginning October 1, 2014, the order of the month names must be:

  • October
  • November
  • December
  • January
  • February
  • March
  • April
  • May
  • June
  • July
  • August
  • September

I can calculate the fiscal month number using the expression shown in the following figure. Note that it is a bit more complex than the expression to calculate the fiscal year. Where I’m using the number ‘9’, I’m really using ’12 – 3′ in which ’12’ is the number of months in the year and ‘3’ is the number of fiscal months in the first calendar year which in this case is October, November, and December. Therefore, if my fiscal calendar started in July, I would use ’12-6′ or ‘6’. A similar argument would explain why I need a ‘3’ at the end of argument for months in the second calendar year to increase the fiscal month number of these months appropriately. Again I would use a ‘6’ for a fiscal year starting in July because there are 6 months in the first calendar year that are part of the current fiscal year.

The final column I’m going to show today is the calculation of the fiscal quarter. In my October-September scenario, the first quarter of the fiscal year includes the months October, November, and December. Since I already have calculated fiscal month numbers, I can simplify my calculation of the quarter by simply dividing the fiscal month number by ‘3’ and rounding the result up to the next whole number if necessary using the ROUNDUP() function. Thus the following figure shows an easy calculation of the fiscal quarter.

That’s all for this time. I hope this helps you create useful date tables with not just traditional calendar dates and calculated fields, but also fiscal calendar fields that you can use in your pivot tables and reports.

C’ya next time.

Populating Your Data Model with A Query

Over the last two weeks we talked about normalizing your logical data model in PowerPivot to help make it easier for users to understand and work with (https://sharepointmike.wordpress.com/2015/03/21/your-logical-data-model-is-not-normal/) and (https://sharepointmike.wordpress.com/2015/03/28/dealing-with-blanks-in-your-data-model/). However, in each of these models, we did all the data manipulation inside the data model. That left a lot of technical fields and tables that we could not remove from the model, but which took extra memory which could limit the size of our data model. So how do we solve that problem? We could do much of the data manipulation outside of the data model or as part of the import step. Let’s see how.

I’m going to start by importing product data from the Access database version of Contoso into the Power Pivot data model.

In the Table Import Wizard, I’m going to select the option to write a query that will specify the data to import. This option allows me to combine data from the dimProduct, dimProductCategory, and dimProductSubCategory tables into a single file. I can also embed the calculation to fix the blank Class fields we visited last time.

Clicking Next, I see the Specify a SQL Query dialog. This is an intimidating dialog. I could manually enter a SQL query here. However, I would have to know all the field names from all the tables that I want to bring together. Not impossible, but there has got to be an easier way. Wait. What is this Design button in the bottom right of the dialog?

When I click the Design button, I get an equally intimidating dialog that looks like it has an area at the top to again manually enter a SQL statement and a grid area at the bottom to display the results of the SQL statement if I were to test run it by clicking the exclamation point (!). Obviously, this screen is not very friendly toward building Access database queries. Nowhere near as friendly as the query builder inside Access itself.

If I go back into Access and build my query there using the query builder, I can visually design my query. First, I add the tables I want to use from the Show Table dialog that appears when I open the Query Designer from the Create ribbon. Then I can add the fields I want to include from any of the included tables by double clicking on their names in the tables at the top of the dialog.

I can even create my calculated column here by entering a new name in the row of the design grid followed by a colon (:). Then I can use the same expression as I did last time to test the value in the ClassName field.

NewClassName: IIF(([ClassName]=””),”[No Class]”, [ClassName])

Note that the IIF() function in Access has a double ‘I’. Otherwise, it works the same. Also note that I’m using a different name for the new column here to keep it separate from the field in the table dimProduct.

After I have finished creating the query, I can test it by clicking the Run button in the Query Tools Design window. If the query executes correctly, I am ready to copy my query to my Power Pivot data load screen. If I open the dropdown menu under SQL View in the Query Tools Design window, I see an option: SQL View as well as some other views.

This option displays the SQL statement which I can copy and then paste back in the Excel Data Model Table Import Wizard as shown below.

If I were to try to run this query directly, it would fail. Admittedly, it took me awhile to figure out why. I finally found that it did not like the one field definition: dimProduct.Size. To test what was going on, I clicked on the Design button to take me to the screen which displays both the SQL query and a grid of results from testing the query. After some testing, I found that if I redefined this field as: dimProduct.[Size], the parsing engine was happy and gave me the results I expected as shown below. The square brackets are used to clarify to the browser that the text within the brackets represent a field name, not a command or reserve word.

I then clicked OK for this screen and Finish on the next screen. I was rewarded with the data engine reading and loading my Product data. Note that it is a single table now. By including the name of the product category and the product subcategory in my query, I no longer need those two additional tables thus reducing the overall size of my data model a bit.

Next I still need to load the Sales data from Contoso along with the date and channel tables. The follow screen shows that I can open the Access database a second time to extract the other tables leaving out the product and related tables.

After the FactSales table is loaded, I still need to create a relationship with the new Product table I pulled in with the query. Even though FactSales and dimProduct were linked in the original Access database, by loading the tables separately, the relationship is not automatically discovered. Therefore, I have to use the Create Relationship dialog to link these two tables.

If I were to display the diagram view of my Power Pivot data model, you can see that the model has been simplified down to only four tables rather than six as we used before.

If I now build the same Pivot table that I did last week to shows sales by class, you can see in the following figure that the results are the same.

I can even create a hierarchy in my product table to go from category to subcategory to product. I covered building hierarchies before (). After building the hierarchy, I strongly recommend hiding the supporting columns so as to not confuse the end-users. (See: https://sharepointmike.wordpress.com/2012/07/21/power-pivot-hierarchies-part-2-of-powerpivot-2012-series/)

Denormalizing your data as you import it may sound a bit more complex, but it can save you memory and Power Pivot’s main limitation is that it must hold the entire data model in memory. So the smaller you can make that model, the more data you can load into it.

Some people might ask how far do I denormalize the model. Of course, in theory, you could denormalize to a perfectly flat single table. However this is not practical as it can serious limit the flexibility of the model and it can even cause other data errors. I tend to think in terms of model objects. For example, Products are an object in the model. Therefore, collapsing the Product Subcategory and Product Category tables into Product makes sense to me. If I had tables Stores, Cities, Countries, and Geography, I might be tempted to collapse them together as well since they represent a Location object for sales. On the other hand, I would not denormalize all the product information into the FactSales table.

So if you are thinking that there is no hard and fast answer to when should you denormalize a model, that is somewhat true. It is a bit of an art. However, if you have a table that is used just as a lookup for a name or perhaps one or two other associated fields and it is not a role playing table (See: https://sharepointmike.wordpress.com/2012/07/21/power-pivot-hierarchies-part-2-of-powerpivot-2012-series/), denormalize that table.

That’s it for this series. C’ya next time.

Dealing with Blanks in Your Data Model

Last time, I discussed some reasons why you might want to denormalize your data model when using PowerPivot (https://sharepointmike.wordpress.com/2015/03/21/your-logical-data-model-is-not-normal/). As you may remember, one reason is to make the model easier for the end user to understand. A second reason is that if you denormalize the data outside of PowerPivot, you can substantially decrease the model size by eliminating many if not most of the lookup tables and just adding the lookup information you need in the entity table such as the Product table. I will talk more about denormalizing the data outside of PowerPivot, or at least during the process of importing the data next time. This time, I want to look at another issue that sometimes occurs in your data that could make your data harder to interpret by your end-users.

I’m talking about blank or null fields in the data. Often times a table will have one or more columns which are optional. Take for example a product table like the one we are using from Contoso. Not every product may be sold by color. Some products may not belong to a style or a class. Some products may not have a measured size or a weight. In some databases, the corresponding lookup tables have a special entry for no color, no style, or no class so that every product can point to one and only one value in the lookup. However, you may have a lookup table that only contains specific values. If a product does not exactly match one of those values, the reference from the product table to the lookup table may be left blank. When the user sees a blank value, they may wonder if the user just forgot to select a value or if no value from the lookup table applied. In other words, they do not really know the reason why the field is blank.

Suppose we start with the data model using Contoso data in the following figure. Note to anyone trying to reproduce this example, I went into the Product table ahead of time and removed the ClassID and ClassName from all products in which the color was ‘Blue’ just to create some records with blank values.

After bringing this data into an Excel PowerPivot data model, I create a simple pivot table to show the sales amount and sales counts by class. Notice that the three product classes that are defined include:

  • Economy
  • Regular
  • Deluxe

But you can see from the report that 87,597 of the over 2 million sales records had no class defined. From a user’s viewpoint, they do not know if someone forgot to enter the class for some of the products or if the definition of class does not apply to some products.

As you build your data model in Excel, you can fix this problem by replacing the blank values with something like the string: “[No Class]”. The difficulty is, however, after you load your data into the PowerPivot data model, you cannot edit individual cells. You can, on the other hand, create a new calculated column.

Since I want the new calculated column to have the name ClassName and because I cannot have two columns in the same table with the same name, I begin by renaming the existing ClassName column to ClassNameOriginal. The actual new name I use does not matter as long as it is different. Also note that changing the column names in the Excel data model has absolutely no effect on the name of the column in the original data source.

After renaming the original ClassName column, I add a new column to the end of the table called ClassName. For the calculation to replace the blanks from the original column, I need to know if those blanks are Null values or if they contain an empty string. If the original column used empty strings when a class value was not defined, I can use the following DAX expression to create my new class name value.

= IF(([ClassNameOriginal]=””),”[No Class]”, [ClassNameOriginal])

This expression uses a simple IF() function to see if the value in the column ClassNameOriginal contains the empty string. If it does, the function outputs the string: [No Class]. Otherwise, the function outputs the current row value of the column ClassNameOriginal. This is shown in the following figure.

On the other hand, if the original ClassName column uses a NULL value when there is no corresponding class, the expression needed here must be defined with the ISBLANK() function to test for NULL values as in:

= IF(ISBLANK([ClassNameOriginal]),”[No Class]”, [ClassNameOriginal])

Whichever method is appropriate for your table, you now have a column that you can use for the PivotTable. Before proceeding however, be sure to go back and hide the original class name column: ClassNameOriginal so as not to confuse users with two similar columns that they could use as dimensions in their tables.

Since I still have my original pivot table open, I can just switch over to the Pivot table and it should automatically update to show that the blank class now has the name [No Class]. (Which goes to show you that having no class is better than just being a blank. J )

That’s it for this week. By the time you read this I will be at Code Camp Orlando giving my presentation: Calculated Columns, Measures, and KPIs, Oh My!

C’ya next time when I will show you how to use the query designer while loading data to the PowerPivot model to fix some of the issues from this week and last week.

Your Logical Data Model is not Normal

If you have been reading this blog over the last several years, you know that I have been a strong supporter of using PowerPivot and the Tabular model to perform data analysis, even for power users, not just DBAs. What you may not have realized is that I’ve been secretly teaching you a little about data modeling. What is data modeling? It is really nothing more than the collection of all the data in tables and the relationships between those tables in a database. Did you know that there are at least two major ‘types’ of data models when it comes to how your structure your data into tables? Each one serves a different purpose. Therefore, it can reasonably be argued that neither one is more correct than the other. But you need to understand when to use each type. That is what I want to discuss today.

Most DBAs and developers who work with databases like SQL Server, Oracle, and many common database are comfortable with the relational model for creating tables and defining the relationships that connect them. In fact, they immediately start to normalize a denomalized database in their heads within seconds of seeing the data schema. The relational model relies on the application of the rules of data normalization introduced by Edgar F. Codd in 1970. In essence, the goal of the rules of data normalization is to minimize the data redundancy which also has the effect of decreasing the overall size of the database while at the same time making it easier to maintain information that would otherwise be repeated through many records.

There are three primary rules that are used to determine whether a database, a collection of tables, has been normalized. These rules are:

First Normal Form: No two rows of data may contain repeating data or groups of data. Such repeating data must be split into a separate but connected table. For example, a sales order may contain the purchase of one or more items. Because the number of items included in a sales order is not predefined, it must be split into a separate table with one row for each item on the sales order. These two tables are then typically connected by an order id.

Second Normal Form: This rule only applies to tables which have a compound primary index, an index built from two or more fields. In this rule, all other fields in the table must depend on the entire compound index value, not only a portion of it. A possible example of this might be a table that includes students at a school in which the primary index combined the school name (or id) along with the student name (or id). Imagine that the table also included the address and phone number of the school. This information is not depended on the combination of the school and the student. It only depends on the school. Therefore, this violation of the second normal form requires that the data related only to the school be split into a second table that includes only school information.

Third Normal Form: This rule requires that every field not part of the primary index be depended on the primary index. Going back to my favorite Contoso database, one could argue that in the FactSales table, the Sales Amount field is redundant since this table also includes the sales quantity, sales price, and any sales discounts or returns. Why is it redundant? Because it can be calculated based on the other values of other columns in the record. Therefore, to fully achieve, third normal form, this field should be removed.

While there are other special case rules that can be applied to normalization of a database, most DBAs will be satisfied with a database that satisfies these three rules. They will then build the tables in their database corresponding to these rules and thus create the physical data model. It is called the physical data model because it defines the physical schema of the tables and the relationships between them.

However, business users of the data don’t look at the database that way. In fact most business users would be baffled by the large number of tables required and the relationships between them. In fact, they will not understand why they need to combine data from a half dozen to a dozen tables just to answer a single query question. As an example, let’s look at how a business user might think of just the products within the Contoso database.

The figure below shows the normalized data model for products and their subcategories and categories. As you can see from the schema, the only important piece of information in the subcategories table is the name of the subcategory. Similarly, the only important piece of information in the categories table is the name of the category.

Most business users would not think of category and subcategory names as part of separate tables, but as attributes of the product itself. In fact they would think of product information more like the following figure.

Imagine a database model with dozens of normalized tables and then try to envision how the typical business user sees the data with perhaps only a half dozen tables after denormalization.

In PowerPivot, we can address this issue in one of two ways. The first way would be to add the category and subcategory names to the product table in SQL Server before loading the data into PowerPivot. This would essentially make the physical data model the same as the logical data model used by the end-users. However, this data model would no longer be normalized.

Physically denomalizing the data tables is not the only solution. I could, as shown before in https://sharepointmike.wordpress.com/2012/07/21/power-pivot-hierarchies-part-2-of-powerpivot-2012-series/, build two calculated fields in the product table that use the RELATED() function to get the category and subcategory names and display them in the product table. I could then hide the two tables, dimCategory and dimSubcategory, from the user so that they would see a structure similar to their expected denormalized logical data model even though the data still is physically stored in a normalized structure.

The advantage of making the changes to the data model outside of PowerPivot is that it effectively reduces the amount of data that PowerPivot must store in memory thus potentially allowing a larger model to be created. The advantage of making the changes inside PowerPivot by using the RELATED() function and then hiding the technical tables that are still needed in the model but do not need to be seen by the client is that it preserves the sanity of those DBAs who cringe every time someone tries to their denormalize their data. Both methods will allow me to create similar Pivot tables in my Excel spreadsheet. Thus both methods can be considered correct.

The thing to remember is this. Data Normalization was developed to make data storage and maintenance easier and more efficient. However, data analysis and reporting often requires a fair amount of data denormalization. This is especially true when performing data analysis using reports, pivot tables and charts, and even third party data analysis tools. In fact, the logical view of the data schema aids in the performance of most data analysis. Just remember that your logical model is not a normalized model.

C’ya next time.

It’s Only the Role I’m Playing

This week I’m going to return to my favorite sample database, Contoso. If you remember, Contoso consists of sales data for a company that produces several different lines of customer electronics, from computers to phones to TVs. The data spans several years of sales. In past examples, I related the FactSales table (the table containing all of the sales records) to several tables including Channel (DimChannel), Date (DimDate), Product (DimProduct) and Product Sub-Category (DimProductSubCategory). In fact, the data source pre-defined these relations so that when I imported the data into my PowerPivot model in Excel, these relationships appeared by default as shown in the following figure.

Visually, I could switch to the Diagram view of the data model to see these relationships represented by solid lines connecting these five tables as shown below.

However, suppose I have additional information in my FactSales table. The DateKey field that I have used in the past identifies the sale date for each of the sales records. In the real world, there may be additional dates associated with each sale. For example, it is not hard to imagine that each sale would also have a Delivery Date and an Invoice Due Date. Now I might ask, does it make sense to report on sales based on the sale date or does it make more sense to report on sales based on the invoice due date or even the date of payment. Well, you might say, ‘That depends on who is asking the question.’ Very true. My Marketing/Sales manager might want to see sales by the sales date. He or she does not care about when the invoice is paid off. They just need to know if they hit their sales quotas for each month. Therefore, they might need a report that uses the default relationship and looks something like the following:

On the other hand, the Chief Financial Officer is not as concerned about when the sale took place, but when the invoice for the sale is due because only then is the income truly realized for the company. Thus the above report does not meet their needs.

Now suppose that I had a few additional columns, as mentioned earlier, that told me some other dates such as when the items were delivered and/or when the invoice for the sale was due. Let’s further suppose that these columns exist in the FactSales table and might appear like the following figure.

With these columns in my FactSales table, I would want to create relations between them and my date table (DimDate) so that I could generate reports using them. The following figure shows the Create Relationship screen in which I define a new relationship between the delivery date (DeliveryDate) column in FactSales and the date (DateKey) column in DimDate. When I click the Create button, the Power Pivot engine creates a second relationship between these two tables. (Remember the first relationship was between the sales date (Datekey) column in FactSales and the (Datekey) column in DimDate.

In a similar fashion, I create a third relation between these two tables to connect the invoice due date (InvDue) column in FactSales and the date (Datekey) column in DimDate. For the purposes of this demo, I am going to stop there. However, I could create additional relationships between any other date fields in the FactSales table and the date (DateKey) column in DimDate. Switching to the Diagram view, I could now see something like the following between these two tables.

Notice that there are three lines between FactSales and DimDate. One of these lines is solid and the other two lines are dashed. You can only have one active relationship between any two tables. That active relationship is represented by the solid line. The other two relations are inactive at the moment. These three relationships are collectively called role playing relations and DimDate is a role playing dimension because only one relation at a time can be active from the DimDate dimension. Think of it this way, the date column (DateKey) in DimDate can play one of three different roles. It can either play the part of the sales date, the delivery date, or the Invoice date. However, it can only play one role at a time. Unless all three dates were exactly the same, I have to choose which role I want DateKey in DimDate to play with the FactSales table by selecting one of the relationships.

The initial data import associate the DateKey column in DimDate to the DateKey value in FactSales which represented the date the item was sold. As long as the report requested wants to allocate sales to the sales date, I need to do nothing other than generate the Pivot Table report as shown below.

However, if I now have to generate a report for my CFO, I would have to go into the model and change the role that DateKey in DimDate plays. I might try to simply right click on the dashed line representing the connection between the DateKey in DimDate with the InvDue column in FactSales and select: Mark as Active.

This action would generate the following error message.

The problem is clearly stated. You can only have a single active relationship between the two referenced tables. Therefore, I must first right click on the current active relationship to deactivate it before activating a different relationship between the two tables.

Once the original relationship is deactivated, I can go back and activate the new relationship between DateKey and InvDue.

Knowing that only one role (relationship) can be active at a time, I am a little surprised that Microsoft did not automatically deactivate the original relationship when a new one is set active. However, like many things, they did not ask me. Plus they may have had other reasons for not automatically deactivating the current relationship. In any case, after making this change and returning to my pivot table (without making any changes to the pivot table itself), the data automatically updates using the new relationship to show the sum of sales by month of the invoice due date rather than the sales date.

If you examine the sales totals for either the years or individual months, you can quickly see that this role change for the DimDate dimension makes a significant change in the sales numbers reported each month.

That’s all for this week. I hope that now you have a better idea how you can use a single dimension to play different roles. (You could also have added the DimDate table three times to the model, once for each date column in FactSales so that each instance of the DimDate table could have a single relationship to FactSales. Then by changing which dimension you select to display in the Pivot Table, you can achieve essentially the same result. This may be easier for the end-user who does not have access to the data model to make the changes listed above. However, your power-users may prefer a simpler model with fewer tables and can change the role played by those dimensions as needed.)

C’ya next time.

A PowerPivot Sort By Any Other Name

The past several weeks I have been looking at the new features of PowerPivot 2012.  I also mentioned this past Tuesday, that the new Microsoft Office 2013 version of Excel includes both PowerPivot and Power View as part of the install.  They just need to turned on.  So this week I’m continuing my Contoso example with Excel 2013 and PowerPivot running on Windows 8.  Therefore, you may notice some differences in the look of the Excel spreadsheet.  Don’t fear.  The techniques  I discuss here work equally well if you are using Vista with Excel 2010 and PowerPivot 2012.

Today I am going to look at a problem that we solved a different way several months ago.  That problem is how to sort month labels chronologically rather than alphabetically.  Back on November 6, 2011 (Wow, was it really that long ago?) I described a way to create a custom sort order for months.  While that method still works in PowerPivot 2012, there is a new way to sort a column by using another column.  Let’s see how that works.

I begin by opening my Contoso 2012 sample data and create a simple pivot table that shows total sales by month.  As shown in the figure below, when I drag the calendar month label field into the rows, the rows appear by default sorted alphabetically.  Unfortunately, that is not a good sort order for months.

To fix this sort order, return to the PowerPivot window and open the tab (table) containing the dates.  In Contoso, this is the dimDate table.  In the Home ribbon find the Sort and Filter option group.  A new button in this group is labeled Sort by Column.    You could immediately click on this button to open the sort dialog.  However, I recommend first clicking on the column for which you want to define the new column sort, then click the Sort by Column button.

This action opens the following dialog and pre-populates the Sort column with the column that you selected.  This saves you a step.  However, if you forget, you can always use the dropdown arrow to display all fields in the current table and select the column from the list.

Next select the column you want to sort by and click the OK button.  You may be tempted to sort by the existing column CalendarMonth.  However, this will generate an error because there are multiple values for CalendarMonth for each value of CalendarMonthLabel because CalendarMonth has the format YYYYMM.  In other words, July would match not just to 07 or 7, but rather to 200507, 200607, 200707, and 200807.  You cannot match the sort column to multiple values in the sort by column.

Therefore, I created a simple calculated column named Month which only consists of the month number as shown in the next figure.

Now I can return to the Sort By Column dialog and create a sort relationship between CalendarMonthLabel and Month as shown in the following dialog.

When I click OK and return to my pivot table in the Excel spreadsheet, I will see that the table automatically refreshes and now displays the month labels in the expected chronological order that makes sense for most users.

That’s it for this week.  A short, simple, but very useful way to sort any column by another column in PowerPivot 2012 or Excel 2013.

C’ya next time.