Forum Discussion

mleepin's avatar
mleepin
Helper I
8 years ago
Solved

Iterative solution replacement in powerbi

I am interested in replacing an iterative xls solution and I would be grateful for some advice please.   Requirement is to:   To calculate the average 'initial score' - this is the average "earli...
Need Help
Tips and Tricks
  • OwenAuger's avatar
    OwenAuger
    8 years ago

    Hi mleepin

     

    The logic you've described sounds exactly right - the question is how to implement this with DAX in Power BI. In this case we need to filter both players and dates per player as you've described.

     

    Here is how I would do it (pbix link).

     

    AVG_INITIAL_SCORE = 
VAR PlayersToInclude =
    FILTER ( VALUES ( Scores[PLAYER] ), CALCULATE ( COUNTROWS ( Scores ) ) > 1 )
RETURN
    CALCULATE (
        AVERAGE ( Scores[SCORE] ),
        GENERATE ( PlayersToInclude, FIRSTDATE ( Scores[DATE] ) )
    )
    AVG_PROCEEDING_SCORE = 
VAR PlayersToInclude =
    FILTER ( VALUES ( Scores[PLAYER] ), CALCULATE ( COUNTROWS ( Scores ) ) > 1 )
RETURN
    CALCULATE (
        AVERAGE ( Scores[SCORE] ),
        GENERATE (
            PlayersToInclude,
            EXCEPT (
                CALCULATETABLE ( VALUES ( Scores[DATE] ) ),
                FIRSTDATE ( Scores[DATE] )
            )
        )
    )

    EDIT: After some thought, the AVG_PROCEEDING_SCORE measure can be written a little more briefly:

    AVG_PROCEEDING_SCORE v2 =
CALCULATE (
    AVERAGE ( Scores[SCORE] ),
    GENERATE (
        VALUES ( Scores[PLAYER] ),
        EXCEPT (
            CALCULATETABLE ( VALUES ( Scores[DATE] ) ),
            FIRSTDATE ( Scores[DATE] )
        )
    )
)

     

    The initial FILTER produces a list of Players with more than 1 row.

    The GENERATE function used as a filter argument within CALCULATE gives us either combinations of players and their first dates, or combinations of players and all dates but their first date.

     

    There may be other approaches but this is how I would do it.

     

    Regards,

    Owen

  • OwenAuger's avatar
    OwenAuger
    8 years ago

    Hi again mleepin

     

    Yes, I'd forgotten that FIRSTDATE and LASTDATE don't always like duplicate dates.

    We can fix this by using FIRSTNONBLANK and LASTNONBLANK instead.

     

    Edited pbix file here.

     

    Within your pbix file, I created these modifiedmeasures:

    AVG_INITIAL_SCORE FIX = 
VAR PlayersToInclude =
    FILTER ( VALUES ( Sheet3[Player]), CALCULATE ( COUNTROWS ( Sheet3) ) > 1 )
RETURN
    CALCULATE (
        AVERAGE ( Sheet3[Score]),
        GENERATE ( PlayersToInclude, FIRSTNONBLANK ( Sheet3[Date], 0 ) )
    )
    AVG_PROCEEDING_SCORE FIX = 
VAR PlayersToInclude =
    FILTER ( VALUES ( Sheet3[Player]), CALCULATE ( COUNTROWS ( Sheet3) ) > 1 )
RETURN
    CALCULATE (
        AVERAGE ( Sheet3[Score] ),
        GENERATE (
            PlayersToInclude,
            EXCEPT (
                CALCULATETABLE ( VALUES ( Sheet3[Date] ) ),
                FIRSTNONBLANK ( Sheet3[Date], 0 )
            )
        )
    )

    These gives the same results as your measures at a total level.

     

    One thing with having MIN_DATE_PER_PLAYER as a calculated column is that the min date won't respond to filter context.

     

    Anyhow, hopefully that's useful and will leave it to you to compare the measures :)

     

    Best regards,

    Owen

  • mleepin's avatar
    mleepin
    8 years ago

    Thank You - following a test these calcs. that you have confirmed both work as expected.  

     

    Regards

     

    ML.

  • OwenAuger's avatar
    OwenAuger
    8 years ago

    Hi mleepin

    Apologies I missed your follow-up query.

    I take it you've solved it already with that measure?

    Regards,

    Owen

mleepin's avatar
mleepin
Helper I
8 years ago

HI Owen,

 

I have another question regarding this solution that you kindly presented a little while ago.

 

I had tested this further with a different range of dates i.e. where there is  a different minimum date per player.  I have found that the use of  FIRSTDATE with GENERATE seems to restrict or filter the result to where a player has a minimum date that is equal to the minimum from the whole result set, rather than the minimum per player ( or from the window).

 

Would there be a way around this?

 

OwenAuger's avatar
OwenAuger
Super User
8 years ago

Hi mleepin

 

From what I can tell, the measures should be behaving correctly even when players have different minimum dates, taking into account any date filters applied as well.

 

Could you post some sample data or a PBIX containing unexpected results, ideally with a visual showing the context where you are using the measures? There could be something I didn't consider when the measures are used in a certain context.

 

In particular FIRSTDATE ( Scores[DATE] ) is internally converted to FIRSTDATE ( CALCULATETABLE ( VALUES ( Scores[DATE] ) ), so context transition should give the first date per Player being iterated over inside GENERATE.

(an interesting article on this here).

 

How I tested the measures: 

In your original sample dataset, there were a few different minimum dates among the players (dates shown as d/mm/yyyy):

To test that the AVERAGE_INITIAL_SCORE measure was working correctly, I created this DAX query in DAX Studio that includes the same Player/Date filter present in the measure.

You could equally create a calculated table in Power BI, leaving out the EVALUATE:

 

-- Test Average Initial Score
EVALUATE

VAR PlayersToInclude =
    FILTER ( VALUES ( Scores[PLAYER] ), CALCULATE ( COUNTROWS ( Scores ) ) > 1 )
VAR PlayerDateFilter =
        GENERATE ( PlayersToInclude, FIRSTDATE ( Scores[DATE] ) )
        
VAR FilteredTable = 
   CALCULATETABLE(
       Scores,
       PlayerDateFilter
       )
RETURN
    FilteredTable

This returns the following table (with different min dates per Player), whose scores average to 2.25 which is what the AVERAGE_INITIAL_SCORE measure returned:

 

DatePlayerScore
1/01/2018A2
1/01/2018B2
1/01/2018C3
5/01/2018D2

 

 Similarly I used this DAX query to test AVERAGE_PROCEEDING_SCORE:

-- Test Average Proceeding Score
EVALUATE
VAR PlayerDateFilter =
    GENERATE (
        VALUES ( Scores[PLAYER] ),
        EXCEPT (
            CALCULATETABLE ( VALUES ( Scores[DATE] ) ),
            FIRSTDATE ( Scores[DATE] )
        )
    )
VAR FilteredTable =
    CALCULATETABLE ( Scores, PlayerDateFilter )
RETURN
    FilteredTable

This returns the following table, whose scores average to 2.875 which is what the AVERAGE_PROCEEDING_SCORE measure returned:

DatePlayerScore
2/01/2018A3
4/01/2018A2
6/01/2018A4
2/01/2018B3
4/01/2018B1
2/01/2018C2
4/01/2018C4
6/01/2018D4

 

Best regards,

Owen

  • mleepin's avatar
    mleepin
    Helper I
    8 years ago

    Hello Owen,

     

    Thank you again much appreciated and thank you for the link as well which I have been looking over.  I did encoutner an error when using the variable based  calcs,   with a different data set. The error was -  Mdx script model (9,38), a date column containing duplicate dates was specifiedin a call to function FIRSTDATE.

     

    I did attempt another longer winded approach with more calcs -  this is perhaps not best practice but this ended up providing desired results:

     

    ---Maximum date per player (column) ---

    MAX_DATE_PER_PLAYER =
    MAXX(
        FILTER ( 'Sheet3', EARLIER ( 'Sheet3'[Player]) = 'Sheet3'[Player]),
        Sheet3[Date]
        )

     

    ---Minimum date per player (column)---

    MIN_DATE_PER_PLAYER =
    MINX(
        FILTER ( 'Sheet3', EARLIER ( 'Sheet3'[Player]) = 'Sheet3'[Player]),
        Sheet3[Date]
        )

     

    ---Count of player occurrence per player (column)---

    COUNT_PLAYERS_PER_PLAYER = COUNTX(FILTER(Sheet3, EARLIER(Sheet3[Player]) = Sheet3[Player]), Sheet3[Player]) 

     

    ---Average initial / earliest score for players that have >1 score (calc)---

    AVG_INITIAL =
    AVERAGEX(
        FILTER ( 'Sheet3', Sheet3[Date] = Sheet3[MIN_DATE_PER_PLAYER] && Sheet3[COUNT_PLAYERS_PER_PLAYER] > 1),
        Sheet3[Score]
    )

     

    ---Average proceeding score for players that have >1 score (Calc)---

    AVG_PROCEEDING =
    AVERAGEX(
        FILTER ( 'Sheet3', Sheet3[Date] <> Sheet3[MIN_DATE_PER_PLAYER] && Sheet3[COUNT_PLAYERS_PER_PLAYER] > 1),
        Sheet3[Score]
    )

     

    A pbix is  uploaded  as well for reference, this demonstrates the above and the error I had mentioned,  this link is here - 

     

    https://1drv.ms/u/s!ApNjWROrbni7j3YEWV24gzT5aLGr

     

    Regards,

     

    ML.

    • OwenAuger's avatar
      OwenAuger
      Super User
      8 years ago

      Hi again mleepin

       

      Yes, I'd forgotten that FIRSTDATE and LASTDATE don't always like duplicate dates.

      We can fix this by using FIRSTNONBLANK and LASTNONBLANK instead.

       

      Edited pbix file here.

       

      Within your pbix file, I created these modifiedmeasures:

      AVG_INITIAL_SCORE FIX = 
VAR PlayersToInclude =
    FILTER ( VALUES ( Sheet3[Player]), CALCULATE ( COUNTROWS ( Sheet3) ) > 1 )
RETURN
    CALCULATE (
        AVERAGE ( Sheet3[Score]),
        GENERATE ( PlayersToInclude, FIRSTNONBLANK ( Sheet3[Date], 0 ) )
    )
      AVG_PROCEEDING_SCORE FIX = 
VAR PlayersToInclude =
    FILTER ( VALUES ( Sheet3[Player]), CALCULATE ( COUNTROWS ( Sheet3) ) > 1 )
RETURN
    CALCULATE (
        AVERAGE ( Sheet3[Score] ),
        GENERATE (
            PlayersToInclude,
            EXCEPT (
                CALCULATETABLE ( VALUES ( Sheet3[Date] ) ),
                FIRSTNONBLANK ( Sheet3[Date], 0 )
            )
        )
    )

      These gives the same results as your measures at a total level.

       

      One thing with having MIN_DATE_PER_PLAYER as a calculated column is that the min date won't respond to filter context.

       

      Anyhow, hopefully that's useful and will leave it to you to compare the measures :)

       

      Best regards,

      Owen

      • mleepin's avatar
        mleepin
        Helper I
        8 years ago

        Thank You - following a test these calcs. that you have confirmed both work as expected.  

         

        Regards

         

        ML.