No CACULATE challenge

[AlexisOlson]

For those haters of CALCULATE like @Greg_Deckler, I challenge you to create a measure without CALCULATE that performs anywhere close to the same speed as this very simple measure that switches to an inactive relationship. It must also not break if a simple filter from any of the dimension tables is applied.

Sales Amount (Delivered) = 
CALCULATE (
    SUM ( Sales[SalesAmount] ),
    USERELATIONSHIP ( Sales[DeliveredDateKey], 'Calendar'[DateKey] )
)

Please use the attached file.

Report preview:

As expected, we can see that sales during periods without a discount promotion have products delivered in promotion months.

[tamerj1 · Accepted solution]

Hi @AlexisOlson  & @Greg_Deckler 
This one uses neither CALCULATE nor CALCULATETABLE and performs faster than the original one

Sales Amount (Delivered) 2 = 
VAR SelectedSales = ALLSELECTED ( Sales )
VAR SummarySales = 
    SUMMARIZE ( 
        SelectedSales, 
        Sales[DeliveredDateKey],
        "@SalesAnount", SUM ( Sales[SalesAmount] ) 
    )
RETURN
    SUMX ( 
        VALUES ( 'Calendar'[DateKey] ),
        VAR FilteredSales = FILTER ( SummarySales, [DeliveredDateKey] = 'Calendar'[DateKey] ) 
        RETURN
            SUMX ( FilteredSales, [@SalesAnount] )
    )

[AlexisOlson]

I'm awarding the win to @tamerj1 for the original problem. It's about 20ms in testing on my machine instead of 10ms for mine but that's still super fast.

I'm awarding the win to @AntrikshSharma for the cumulative challenge since that solution works for subtotals as well.

[AntrikshSharma · Accepted solution]

@AlexisOlson Good challenge! I am a Pro CALCULATE but recommend users not to stick with a single side. 
Here are my solutions:

Solution 1 using INNERJOIN, performs in 40ms without any filter

Delivered Amount AS = 
VAR Dates = 
    SELECTCOLUMNS ( 
        VALUES ( 'Calendar'[DateKey] ),
        "DeliveredDateKey", 'Calendar'[DateKey] & ""
    )
VAR SalesByDelivery = 
    SELECTCOLUMNS ( 
        SUMMARIZE ( 
            ALLSELECTED ( Sales ),
            Sales[DeliveredDateKey],
            "Sum", SUM ( Sales[SalesAmount] )
        ),
        "DeliveredDateKey", [DeliveredDateKey] & "",
        "Sales", [Sum]
    )
VAR Result = 
    SUMX ( 
        NATURALINNERJOIN ( Dates, SalesByDelivery ),
        [Sales]
    )
RETURN 
    Result

Solution 2 using CONTAINSROW or the IN performs in 23 ms

Delivered Amount AS 2 = 
VAR SalesByDeliveryDate = 
    SUMMARIZE ( 
        ALL ( Sales ),
        Sales[DeliveredDateKey],
        "TotalSales", SUM ( Sales[SalesAmount] )
    )
VAR CurrentYearRows = 
    FILTER ( 
        SalesByDeliveryDate,
        CONTAINSROW (
            VALUES ( 'Calendar'[DateKey] ),
            Sales[DeliveredDateKey]
        )
    )
VAR Result = 
    SUMX ( CurrentYearRows, [TotalSales] )
RETURN 
    Result

Cumulative Total - I added a YearMonth column in the Dates table, since we are not showing date level I decided to reduce granularity, takes about 50ms.

CT DeliveredAmount AS = 
VAR SalesByDelivery = 
    SUMMARIZE ( 
        ALL ( Sales ),
        Sales[DeliveredDateKey],
        "@TotalSales", SUM ( Sales[SalesAmount] ),
        "@YearMonth", YEAR ( Sales[DeliveredDateKey] ) * 100 + MONTH ( Sales[DeliveredDateKey] ) 
    )
VAR Result = 
    SUMX ( 
        FILTER ( 
            SalesByDelivery, 
            [@YearMonth] <= MAX ( 'Calendar'[YearMonth] ) 
        ),
        [@TotalSales]
    )
RETURN 
    Result

@tamerj1 I tried and your code and it returns incorrect subtotals.

[tamerj1]

@AntrikshSharma 
Yes you are right I've over looked that but this is an easy fix. The outer SUMX should have been a MAXX

Cumulative Sales Amount (Delivered) 2 = 
VAR SelectedSales = ALLSELECTED ( Sales )
VAR SummarySales = 
    SUMMARIZE ( 
        SelectedSales, 
        Sales[DeliveredDateKey],
        "@SalesAnount", SUM ( Sales[SalesAmount] ) 
    )
VAR SummaryDates =
    SUMMARIZE ( 
        'Calendar',
        'Calendar'[Year],
        'Calendar'[MonthName],
        "@MaxDate", MAX ( 'Calendar'[DateKey] )
    )
RETURN
    MAXX ( 
        SummaryDates,
        VAR FilteredSales = FILTER ( SummarySales, [DeliveredDateKey] <= [@MaxDate] ) 
        RETURN
            SUMX ( FilteredSales, [@SalesAnount] )
    )

[Greg_Deckler]

@tamerj1 Slick, way better than my revision in my video I made last night.

[tamerj1]

@Greg_Deckler 

Wow!

Thank you for mentioning me in your video ☺️

My name is Tamer Juma by the way and we are connected in LinkedIn actually you replied to my comments there couple of times. 

Note: I fixed the wrong total by simply replacing the outer SUMX with MAXX

[Greg_Deckler]

@tamerj1 Yep, I saw that after the video went up. I posted the PBIX out on MicrosoftHatesGreg on github with the revised measure that corrects the subtotals. Absolutely, of course I'm going to mention you! You're one of the stars of the show! And nice to meet you Tamer Juma! Sorry, never put the two together before or my memory is failing me! But, it did let me work in the SwizzleStick420 joke!

[Greg_Deckler]

@AntrikshSharma Really impressive! If there is a restriction that you can't change the data model, then this slight tweak works just as well:

Cumulative Sales Amount (Delivered) AS rev = 
VAR SalesByDelivery = 
    SUMMARIZE ( 
        ALLSELECTED ( Sales ),
        [DeliveredDateKey],
        "@TotalSales", SUM ( [SalesAmount] ),
        "@YearMonth", YEAR ( [DeliveredDateKey] ) * 100 + MONTH ( [DeliveredDateKey] ) 
    )
VAR Result = 
    SUMX ( 
        FILTER ( 
            SalesByDelivery, 
            [@YearMonth] <= MAX ( 'Calendar'[Year] ) * 100 + MAX( 'Calendar'[MonthOfYear])
        ),
        [@TotalSales]
    )
RETURN
    Result

[tamerj1 · Accepted solution]

Hi @AlexisOlson  & @Greg_Deckler 
This one uses neither CALCULATE nor CALCULATETABLE and performs faster than the original one

Sales Amount (Delivered) 2 = 
VAR SelectedSales = ALLSELECTED ( Sales )
VAR SummarySales = 
    SUMMARIZE ( 
        SelectedSales, 
        Sales[DeliveredDateKey],
        "@SalesAnount", SUM ( Sales[SalesAmount] ) 
    )
RETURN
    SUMX ( 
        VALUES ( 'Calendar'[DateKey] ),
        VAR FilteredSales = FILTER ( SummarySales, [DeliveredDateKey] = 'Calendar'[DateKey] ) 
        RETURN
            SUMX ( FilteredSales, [@SalesAnount] )
    )

[Greg_Deckler]

@AlexisOlson 

Sales Amount (Delivered) NC = 
    VAR __Dates = SELECTCOLUMNS('Calendar',"DateKey",[DateKey])
    VAR __Table = FILTER(ALLSELECTED('Sales'), [DeliveredDateKey] IN __Dates)
    VAR __Result = SUMX(__Table, [SalesAmount])
RETURN
    __Result

In my testing the measures return within 50 milliseconds of one another.

[AlexisOlson]

Nice!

Now try it without the slicer filtering and see how the performance compares. 🙂

[Greg_Deckler]

@AlexisOlson This one doesn't use CALCULATE either (LOL!) and is actually 3 times faster than the original measure from a DAX Query perspective.

NC3 = 
 VAR __Table = CALCULATETABLE('Sales', USERELATIONSHIP ( Sales[DeliveredDateKey], 'Calendar'[DateKey] ))
 VAR __Result = SUMX(__Table, [SalesAmount])
RETURN
 __Result

[AlexisOlson]

@Greg_Deckler Ha! The performance of that is the same as my CALCULATE version in my testing.

For a final challenge, see if you can write a cumulative version without CALCULATE or CALCULATETABLE that refreshes in under 10 seconds with no slicer filtering. The CALCULATE version refreshes in about 0.2 sec, including all the overhead:

Cumulative Sales Amount (Delivered) = 
VAR _MaxDate = MAX ( 'Calendar'[DateKey] )
VAR _Result =
    CALCULATE (
        SUM ( Sales[SalesAmount] ),
        USERELATIONSHIP ( Sales[DeliveredDateKey], 'Calendar'[DateKey] ),
        ALL ( 'Calendar' ),
        'Calendar'[DateKey] <= _MaxDate
    )
RETURN
    _Result

[tamerj1]

@AlexisOlson 
Here you go 

Cumulative Sales Amount (Delivered) 2 = 
VAR SelectedSales = ALLSELECTED ( Sales )
VAR SummarySales = 
    SUMMARIZE ( 
        SelectedSales, 
        Sales[DeliveredDateKey],
        "@SalesAnount", SUM ( Sales[SalesAmount] ) 
    )
VAR SummaryDates =
    SUMMARIZE ( 
        'Calendar',
        'Calendar'[Year],
        'Calendar'[MonthName],
        "@MaxDate", MAX ( 'Calendar'[DateKey] )
    )
RETURN
    SUMX ( 
        SummaryDates,
        VAR FilteredSales = FILTER ( SummarySales, [DeliveredDateKey] <= [@MaxDate] ) 
        RETURN
            SUMX ( FilteredSales, [@SalesAnount] )
    )

This is faster than yours. I guess @Greg_Deckler has a very good point 😉

[Greg_Deckler]

@tamerj1 One other thing, on a whim decided to replace SUMMARIZE with GROUPBY. SUMMARIZE is waaaaaayyyy more performant than GROUPBY. Like orders of magnitude more performant.

[tamerj1]

@Greg_Deckler 

To be honest somehow I cheated 😅

I couldn't do that without context transition so I thought ok, SUMMARIZE can do that as well. Yes indeed GROUPBY does not perform context transition therefore would perform slower in this case. 

[Greg_Deckler]

@tamerj1 Wow! Super impressive! That is lightning fast!! I ran 10-15 tests and your version has DAX query timing that is on average 2-3 times faster than CALCULATE ( ~75 milliseconds versus ~200 milliseconds.  Overall time was consistently 100 milliseconds faster. @BrianJulius 

The technique also works for the original problem and performs almost identically to the CALCULATE version.

NC4 = 
    VAR __SelectedSales = ALLSELECTED( 'Sales' )
    VAR __SummarySales = 
        SUMMARIZE(
            __SelectedSales,
            [DeliveredDateKey],
            "__SalesAmount", SUM( [SalesAmount] )
        )
    VAR __Dates = DISTINCT('Calendar'[DateKey])
    VAR __Result = 
        SUMX( FILTER( __SummarySales, [DeliveredDateKey] IN __Dates), [__SalesAmount])
RETURN

Once again, CALCULATE goes down in defeat.

[tamerj1]

@Greg_Deckler 

Yes I actually placed the solution of the first problem in a separate reply right before the the solution of the 2nd problem. 

[Greg_Deckler]

@tamerj1 Somehow I missed that, the way these forums do threading is mildly infuriating at times.

[Greg_Deckler]

@AlexisOlson It's a good more or less single use case where CALCULATE is beneficial but really just because USERELATIONSHIP was coded to work with CALCULATE. USERELATIONSHIP could have been coded to support FILTER as well. It's not really CALCULATE, it's the additional functions that were coded to work with CALCULATE. But, there are plenty of examples of the reverse. Try writing a version of these with only using CALCULATE and no X aggregator that A. Actually works and B. Performs significantly faster.

1. Write a running total measure that uses a single table data model only using only CALCULATE and no X aggregator
2. Write a working measure total for a semi-additive measure using only CALCULATE and no X aggregator. Measure Totals, The Final Word - Microsoft Power BI Community
3. Write a version of Open Tickets measure using only CALCULATE and no X aggregator
4. Write a Deseasonalized Correlation Coefficient measure using only CALCULATE and no X aggregator
5. Write a Net Work Days measure using only CALCULATE and no X aggregator
6. Write a Simple Linear Regression measure formula using only CALULATE and n X aggregator
7. Write Cthulhu using only CALCULATE and no X aggregator
8. Write an MTBF measure using CALCULATE and no X aggregator
9. Write a While loop measure using CALCULATE
10. Write a For loop measure using CALCULATE
11. Write a Days of Supply measure using only CALCULATE and no X aggregator
12. Write Hour Breakdown measure using only CALCULATE and no X aggregator
13. Write Lookup Value measure using only CALCULATE and no X aggregator
14. Write Overworked measure using only CALCULATE and no X aggregator

I could go on, I only got through 3 pages of 11 of the Quick Measure Gallery.

The point here is the the No CALCULATE approach is a far better and more flexible approach to writing DAX that allows you to solve real-world problems, problems that CALCULATE could never hope to touch. Yes, there are times when CALCULATE is a good idea for one reason or another. But, this fixation the DAX community has on CALCULATE is unhealthy. It makes DAX harder to learn and breeds this crazy culture where people want to use it everywhere when it is simply making their lives harder and is absolutely unnecessary the vast majority of the time when the simple approach of:

1. Create some VAR's
2. Create a Table VAR
3. X Aggregator

Will solve the VAST majority of problems in DAX without ever needing CALCULATE. Simple. No reason to worry about the internal workings of CALCULATE or context transition or pretty much any of the stuff that people find "hard" about DAX.

[AlexisOlson]

@Greg_Deckler X aggregators are absolutely essential. I have no interest in comparing No CALCULATE vs No X aggregator since I'd rather not handicap myself at all.