FILTER vs CALCULATETABLE

[Anonymous]

I'm about to lose some serious street cred, but so be it... 🙂

I use FILTER( ) all the time and CALCULATETABLE( ) almost never.  Is there some time I *should* be using CALCULATETABLE?  What is this difference in these 2 functions?

If possible, would love a crisp example of them returning different results...

[slap33]

just found the answer p.364 of the Definitive Guide to Dax ! ✅ Thanks Marco&Alberto !!

[slap33]

7 years later....  😂

i was wondering why CALCULETABLE is advised against FILTER although CALCULETABLE is actually built on... FILTER !
saying it another way : if i wanna avoid using FILTER, by using CALCULATETABLE, i end up using FILTER because i did use CALCULATETABLE 😅😅

[mattbrice]

How about this from the best:

https://www.sqlbi.com/articles/filter-vs-calculatetable-optimization-using-cardinality-estimation/

[Anonymous]

Ya, of course I went to see what @marcorusso had to say! 🙂   But ... somehow it didn't help me.  At least from a pure functionality perspective -- as that seems to do more w/ performance.

And yet, I'm not convinced they are functionally equivalent.

[v-jiascu-msft]

Hi @Anonymous

"Filter" can generate a row context while "Calculatetable" can't. In the example below, the formulas are very similar. Both work but bring us different results. Hope this would help.

Filter_NumOfCities =
COUNTROWS ( FILTER ( 'Table15', 'Table15'[CITY] = EARLIER ( Table15[CITY] ) ) )

Calculatetable_NumOfCities =
COUNTROWS (
    CALCULATETABLE ( 'Table15', 'Table15'[CITY] = EARLIER ( 'Table15'[CITY] ) )
)

Best Regards!

Dale

[mattbrice]

@v-jiascu-msft I think the reason for the different results is because of the context transition caused by CALCULATETABLE, not because of the creation of the row context.  Just because you don't write an explict FILTER doesn't mean it isn't being used by Dax.   Internally I believe the CALCULATETABLE expression you wrote as: 

Calculatetable_NumOfCities =
COUNTROWS (
    CALCULATETABLE ( 'Table15', 'Table15'[CITY] = EARLIER ( 'Table15'[CITY] ) )
)

internally gets rewritten by Dax engine as:

Calculatetable_NumOfCities =
COUNTROWS (
    CALCULATETABLE (
        'Table15',
        FILTER (
            ALL ( 'Table15'[CITY] ),
            'Table15'[CITY] = EARLIER ( 'Table15'[CITY] )
        )
    )
)

the difference being, as calucated columns, the filter context that gets transitioned.

your 'Filter_NumOfCities' calc column has a row context but no filter context so filters the entire table down to rows where Table15[City] = "Compton" (for first and second row, LOS ANGELES for next 3, and so on).

your 'Calculatetable_NumOfCities' calc column transitions in (because of CALCULATETABLE)  all the values for all the columns for the current row except for 'City' which is being blocked by the ALL generated by the Dax engine.  So the columns you masked are transitioned into the filter contex.  The masked and not seen columns are filtering the table down to where only one row is left, except for SANTA MONICA which must have two identical rows. 

per my understanding, this is how the Dax engine works...any comments are welcome!

[ahmedaldafaae]

Hi, I think your next measure should be

Filter_NumOfCities =
COUNTROWS (
   FILTER (
            ALL ( 'Table15'[CITY] ),
            'Table15'[CITY] = EARLIER ( 'Table15'[CITY] )
        )
    )

[Anonymous]

---

@mattbrice wrote:

@v-jiascu-msft I think the reason for the different results is because of the context transition caused by CALCULATETABLE, not because of the creation of the row context.  Just because you don't write an explict FILTER doesn't mean it isn't being used by Dax.   Internally I believe the CALCULATETABLE expression you wrote as: 

Calculatetable_NumOfCities =
COUNTROWS (
    CALCULATETABLE ( 'Table15', 'Table15'[CITY] = EARLIER ( 'Table15'[CITY] ) )
)

internally gets rewritten by Dax engine as:

Calculatetable_NumOfCities =
COUNTROWS (
    CALCULATETABLE (
        'Table15',
        FILTER (
            ALL ( 'Table15'[CITY] ),
            'Table15'[CITY] = EARLIER ( 'Table15'[CITY] )
        )
    )
)

the difference being, as calucated columns, the filter context that gets transitioned.

your 'Filter_NumOfCities' calc column has a row context but no filter context so filters the entire table down to rows where Table15[City] = "Compton" (for first and second row, LOS ANGELES for next 3, and so on).

your 'Calculatetable_NumOfCities' calc column transitions in (because of CALCULATETABLE)  all the values for all the columns for the current row except for 'City' which is being blocked by the ALL generated by the Dax engine.  So the columns you masked are transitioned into the filter contex.  The masked and not seen columns are filtering the table down to where only one row is left, except for SANTA MONICA which must have two identical rows. 

per my understanding, this is how the Dax engine works...any comments are welcome!

---

Could you give a short quick example of Filter used inside Calculatetable and Calculatetable used inside of Filter, and state the key points?  I am trying to grasp the best use cases.

Thanks in advance!

[Anonymous]

I find your resonse very helpful in understanding the concept.  Thank you!

[Anonymous]

Thanks Dale, that *is* interesting.. though, I wish I could better wrap my head around what is happening in your example 🙂

* Since the formulas "work"... there must be an EARLIER( ) context, which suggets... a context was created!? 

* Unclear why Santa Monica is different than the others (I assume related to the hidding data?)

[mattbrice]

There is row context because both look to be calculated columns,...not measures. And of course calculated columns have row context, but only transitioned to filter context with use of CALCULATE or CALCULATETABLE or some other table function.

[Anonymous]

What if I remove CALCULATETABLE from the language.  What do we lose?

[mattbrice · Accepted solution]

---

@Anonymous wrote:

What if I remove CALCULATETABLE from the language.  What do we lose?

---

CALCULATETABLE triggers context transition whereas FILTER does not.  Andy by itself, FILTER creates a row context whereas CALCULATETABLE does not.  But other than these, it is a question for Marco if he is lurking around out there...potential performance issues per his article I referenced would be one thing I would think...