Controlling filter context in percentile calculation

[cashewNut]

I am struggling to control different filter context transitions within a complex dax measure. I am trying to compute the median duration for a certain percentile of events, where that percentile is partitioned within a table. My table is structured like this:

index     opportunity_id  event_order   duration  status   date

109750	a	1	120	Active (Not Recruiting)	24/01/2023 00:00
86263	b	1	361	Active (Not Recruiting)	12/07/2022 00:00
109752	a	2	134	Active (Not Recruiting)	07/02/2023 00:00
86264	b	2	361	Active (Not Recruiting)	12/07/2022 00:00

My strategy has been to calculate the target event order (e.g. the one which is the 25th percentile) partitioned by the opportunity_id, and then calculate the median duration for a table filtered to only include records where the event_order = target:

25% Duration (Static) = 

VAR _percentile = 0.25

// compute the 25% percentile event order partitioned by opportunity id
VAR _tbl =

        ADDCOLUMNS (
            'Table',
                "@target",
                    CEILING(
                        CALCULATE (
                            MAX ( 'Table'[event_order]),
                            FILTER (
                                ALL('Table'),
                                'Table'[opportunity_id] = EARLIER ( 'Table'[opportunity_id] )
                            ))
                    * _percentile, 
                1)
            )


VAR  _result =

        MEDIANX(
            FILTER (
            _tbl,
            [event_order] = [@target]
        ), 
        [duration])

RETURN _result

This returns the correct results for all visual types. However, the use of ALL() makes it insensitive to external filtering in the right order, which is desired. For example, if applying a certain filter will remove those records from the _tbl result, when what I need is for the new event order and target to be recalculated dynamically.

To enable dynamic filtering I have tried to re-calculate the event_order dynamically within the measure like so

25% Duration (dynamic) = 

VAR _percentile = 0.25

// compute the 25% percentile event order partitioned by opportunity id
VAR _tbl =

        ADDCOLUMNS (
            'Table',
                "@target",
                    CEILING(
                        CALCULATE (
                            MAX ( 'Table'[event_order]),
                            FILTER (
                                ALL('Table'),
                                'Table'[opportunity_id] = EARLIER ( 'Table'[opportunity_id] )
                            ))
                    * _percentile, 
                1),
                "@event_order",
                ROWNUMBER(
                    ALLSELECTED('Table'), 
                    ORDERBY([event_order], ASC, [index], ASC), 
                    LAST, 
                    PARTITIONBY([opportunity_id])
)
            )


VAR  _result =

        MEDIANX(
            FILTER (
            _tbl,
            [@event_order] = [@target]
        ), 
        [duration])

RETURN _result

This returns the correct results in visuals partitioned by opportunty_id (and allows for external filtering) but as soon as I add a date range it becomes inaccurate because it partitions for each date range in the visual. I.e. it returns the 25th percentile for each date range in which it occurs. 

Is there any way to control the filter context in such a way that it computes accurate results against timelines (i.e. just one 25th percentile event per opportunity) while still respecting filtering that happens at the reporting layer? 

[v-kongfanf-msft]

Hi @cashewNut ,

Try to modify your formula like below:

25% Duration (Adjusted) = 
VAR _percentile = 0.25
VAR _filteredTable = FILTER(ALLSELECTED('Table'), [status] = "Active (Not Recruiting)")
VAR _partitionedTable = 
    ADDCOLUMNS (
        _filteredTable,
        "@target",
            CEILING(
                CALCULATE (
                    MAX ( 'Table'[event_order]),
                    ALLEXCEPT('Table', 'Table'[opportunity_id])
                ) * _percentile, 
            1)
    )
VAR _result = 
    MEDIANX(
        FILTER (
            _partitionedTable,
            [event_order] = [@target]
        ), 
        [duration]
    )
RETURN _result

Best Regards,
Adamk Kong

If this post helps, then please consider Accept it as the solution to help the other members find it more quickly.

[cashewNut · Accepted solution]

Thanks @v-kongfanf-msft ! I am trying to avoid specifying filters directly in the query. I've ended up getting the query to work in production where I have a relationship to the filtering tables ('Table1') in this example, using a combination of PERCENTILE.INC(), KEEPFILTERS(), and ALLACCEPT()

VAR _percentile =
ADDCOLUMNS (
FILTER (
'Table',
"@target",
CALCULATE (
MINX (
FILTER (
'Table',
[event_order]
>= PERCENTILEX.INC (
VALUES ( 'Table'[event_order] ),
'Table'[event_order],
_percentile
)
),
[event_order]
),
ALLEXCEPT ( 'Table', 'Table[opportunity_id] ),
KEEPFILTERS ( 'TABLE1' ),
)
)

//calculate the median duration for records in a result set limited to the ordering targets
VAR _result =
MEDIANX (
FILTER (
_percentile ,
[_event_order] = [@target]
),
[Duration]
)
RETURN
_result