How to do custom rounding in DAX?

[PunchBird]

Hi all,

Here is a Christmas brain cracker for our DAX specialists!

I want to create a DAX measure that calculates the percentage of values per category in a table to the total of that table, with the following custom rounding:

- All percentages need to be rounded down to the nearest integer.

- Then it needs to be identified how much percentage points are missing to reach the 100% in total.

- Next, these percentage points need to be added each to the percentages with the largest decimal part in descending order.

Please note the dataset is huge so the impact of the DAX calculation should be limited as much as possible

So for example:

Category A: 26.98%,
Category B: 34.78%,
Category C: 38.24%

Rounded down this becomes:

Category A: 26
Category B: 34
Category C: 38
26 + 34 + 38 = 98 -> 2 missing percentage points

These missing percentage points need to be assigned to the 2 percentages with the highest decimal part:
Category A: 27
Category B: 35
Category C: 38
27 + 35 + 38 = 100%

How can I do this with DAX? I think some iteration and /or buffering is required, but how do I do this?

@parry2k maybe you know?

[Greg_Deckler · Accepted solution]

@PunchBird I have modified things to be more in line with your data and requirements. See if this works. Updated PBIX is attached below signature.

Measure 2 = 
    VAR __Cat2 = MAX('Table'[Category2])
    VAR __Table = 
        GENERATE(
            SUMMARIZE(
                FILTER(ALLSELECTED('Table'), [Category2] = __Cat2),
                'Table'[Category1],'Table'[Category2],"Value",MAX([Percentage])),
                VAR __Value = [Value]
                VAR __RD = ROUNDDOWN(__Value,0)
                VAR __Decimal = __Value - __RD
            RETURN
                ROW(
                    "RD", __RD,
                    "Decimal", __Decimal
                )
        )
    VAR __MaxDecimal = MAXX(__Table,[Decimal])
    VAR __MaxCategory = MAXX(FILTER(__Table, [Decimal] = __MaxDecimal),[Category1])
    VAR __2ndMaxDecimal = MAXX(FILTER(__Table, [Category1] <> __MaxCategory), [Decimal])
    VAR __2ndMaxCategory = MAXX(FILTER(__Table, [Category1] <> __MaxCategory && [Decimal] = __2ndMaxDecimal),[Category1])
    VAR __Category = MAX('Table'[Category1])
    VAR __Result = 
        IF(
            __Category = __MaxCategory || __Category = __2ndMaxCategory, 
            ROUNDUP(MAX([Percentage]),0), 
            ROUNDDOWN(MAX([Percentage]),0)
        )
RETURN
    __Result //CONCATENATEX(__Table, [Category1]&":"&[Category2]&":"&[Value]&":"&[RD]&":"&[Decimal],UNICHAR(10)&UNICHAR(13))

[Greg_Deckler · Accepted solution]

@PunchBird Easy fix for that, see below and attached PBIX.

Measure 2 = 
    VAR __Cat2 = MAX('Table'[Category2])
    VAR __Table = 
        GENERATE(
            SUMMARIZE(
                FILTER(ALLSELECTED('Table'), [Category2] = __Cat2),
                'Table'[Category1],'Table'[Category2],"Value",MAX([Percentage])),
                VAR __Value = [Value]
                VAR __RD = ROUNDDOWN(__Value,0)
                VAR __Decimal = __Value - __RD
            RETURN
                ROW(
                    "RD", __RD,
                    "Decimal", __Decimal
                )
        )
    VAR __MaxDecimal = MAXX(__Table,[Decimal])
    VAR __MaxCategory = MAXX(FILTER(__Table, [Decimal] = __MaxDecimal),[Category1])
    VAR __2ndMaxDecimal = MAXX(FILTER(__Table, [Category1] <> __MaxCategory), [Decimal])
    VAR __2ndMaxCategory = MAXX(FILTER(__Table, [Category1] <> __MaxCategory && [Decimal] = __2ndMaxDecimal),[Category1])
    VAR __Category = MAX('Table'[Category1])
    VAR __SumDown = SUMX(__Table, [RD])
    VAR __Result = 
        SWITCH(__SumDown,
            99, 
                IF(
                    __Category = __MaxCategory, 
                    ROUNDUP(MAX([Percentage]),0), 
                    ROUNDDOWN(MAX([Percentage]),0)
                ),
            98, 
                IF(
                    __Category = __MaxCategory || __Category = __2ndMaxCategory, 
                    ROUNDUP(MAX([Percentage]),0), 
                    ROUNDDOWN(MAX([Percentage]),0)
                )
        )
RETURN
    __Result //CONCATENATEX(__Table, [Category1]&":"&[Category2]&":"&[Value]&":"&[RD]&":"&[Decimal],UNICHAR(10)&UNICHAR(13))

[bolfri]

I can prepare Power Query M for rounddown or roundup solution, but without this additional req to split this % value between Category 2 or 3. Sorry. 🙂

[ppm1]

I see you worked a tie into the example data. Please see this updated measure expression that uses RAND to break the tie.

AdjValue = 
VAR vThisVal = [AvgVal]
VAR vThisCategory1 =
    MIN ( T1[Category1] )
VAR tNew =
    ADDCOLUMNS (
        CALCULATETABLE ( DISTINCT ( T1[Category1] ), REMOVEFILTERS ( T1[Category1] ) ),
        "cOrigVal", [AvgVal]
    )
VAR tRoundMod =
    ADDCOLUMNS (
        tNew,
        "cRD", ROUNDDOWN ( [cOrigVal], 0 ),
        "cMod", MOD ( [cOrigVal], 1 ),
        "cRand", RAND()/1000
    )
VAR vThisModRand = SUMX(FILTER(tRoundMod, T1[Category1] = vThisCategory1), [cMod] + [cRand])
VAR vGapTo100 =
    100 - SUMX ( tRoundMod, [cRD] )
VAR vModRank =
    RANKX ( tRoundMod, [cMod] + [cRand], vThisModRand, DESC )
VAR vResult =
    IF ( vModRank <= vGapTo100, ROUNDUP ( vThisVal, 0 ), ROUNDDOWN ( vThisVal, 0 ) )
RETURN
    vResult

Pat

[PunchBird]

Hi Pat @ppm1 I tested this on my simple test dataset, and noticed the results are correct! However, it seems that the values assigned to the ties (B and C in Category1) are very instable, i.e. when I refresh the report the values change every time, see screenshots below in column TEST. Maybe this has something to do with the results stored in a virtual table?

After refresh:

 

[parry2k]

@PunchBird Ah! so remainder distribution is dynamic. 

[PunchBird]

yes indeed!

[ppm1]

Here's a measure expression that shows one way to do it. You'll have to watch out for ties in the MOD values, and for when the values add up to >100 (but you can adapt this measure to do that).

AdjValue =
VAR vThisCategory =
    MIN ( T1[Category] )
VAR vThisVal = [AvgVal]
VAR vThisRD =
    ROUNDDOWN ( vThisVal, 0 )
VAR vThisMod =
    MOD ( vThisVal, 1 )
VAR tNew =
    ADDCOLUMNS ( ALL ( T1[Category] ), "cOrigVal", [AvgVal] )
VAR tRoundMod =
    ADDCOLUMNS (
        tNew,
        "cRD", ROUNDDOWN ( [cOrigVal], 0 ),
        "cMod", MOD ( [cOrigVal], 1 )
    )
VAR vGapTo100 =
    100 - SUMX ( tRoundMod, [cRD] )
VAR vModRank =
    RANKX ( SELECTCOLUMNS ( tRoundMod, "cMod2", [cMod] ), [cMod2], vThisMod, DESC )
VAR vResult =
    IF ( vModRank <= vGapTo100, ROUNDUP ( vThisVal, 0 ), ROUNDDOWN ( vThisVal, 0 ) )
RETURN
    vResult

Pat

[Greg_Deckler]

@PunchBird I did this (see below). PBIX is attached.

Measure = 
    VAR __Table = 
        GENERATE(
            SUMMARIZE(ALLSELECTED('Table'),'Table'[Category]),
                VAR __Value = [Percentage]
                VAR __RD = ROUNDDOWN(__Value,2)
                VAR __Decimal = __Value - __RD
            RETURN
                ROW(
                    "Value", __Value,
                    "RD", __RD,
                    "Decimal", __Decimal
                )
        )
    VAR __MaxDecimal = MAXX(__Table,[Decimal])
    VAR __MaxCategory = MAXX(FILTER(__Table, [Decimal] = __MaxDecimal),[Category])
    VAR __Category = MAX('Table'[Category])
    VAR __Result = IF(__Category = __MaxCategory, ROUNDUP([Percentage],2), ROUNDDOWN([Percentage],2))
RETURN
    __Result

[PunchBird]

@Greg_Deckler thanks for your calculation. I tested this on my testdata (see below for more elaborate example), but apparently your calculation only works when there is one percent point to be added. If there are two percent points that need to be added, the total is 99%, see also screenshot below. Maybe I could use a for- or while loop, like you described here?

[parry2k]

@PunchBird Interesting question. Does the remainder always get equally distributed to the top two contributors? 

[PunchBird]

@parry2k the remainder needs to be distributed in descending order. So if the remainder is 1, it is assigned to the percentage that had largest decimal part, and  if the remainder is 2, it is assigned to the two top percentages that had largest decimal part