Forum Discussion
Advocate ISUMX returning blanks (BonusPoints: avoiding circular logic)
Super User
a way to do that more efficiently, without multiple measures and without running into circular logic
- not really
- no
- heck no.
This is a closed loop control system that will require multiple runs to calm down and level in. There is no chance at all that a deterministic single loop system like Power BI (which doesn't even have the concept of global variables) can handle this very complex process.
Can you add a timing component? Datetime stamps on when the referrals happened? So you can build a sequence?
I guess you could build a solution out of hydraulics, or with mycelium.
nickc_innovaI don't believe it is as hopeless a situation as you describe.
I was able to make some progress - though I admit I still don't have the full solution.
After calculating the demand referred out, and calculating the demand remaining at the PSA, I have a measure for the demand referred in for each PSA as follows:REFERRED_IN =VAR __currentPSA = MAX(SDP_OUTPUT[PSA_NAME])VAR __currentSTOP = MAX(SDP_OUTPUT[STOP_CODE])RETURNCALCULATE(SUMX(SDP_OUTPUT, [REFERRED_OUT]),REMOVEFILTERS(SDP_OUTPUT), SDP_OUTPUT[REFERRAL_LOCATION] = __currentPSA && SDP_OUTPUT[STOP_CODE] = __currentSTOP)
This resolves the first issue I originally posted about. I then took it a step further, by simply writing another measure that adds the Referred In demand and the "Adjusted Demand" - which probably isn't the best name for the field but it represents the demand originated at the PSA minus the portion that was referred out.
I think of this as the first cycle of referral. For any PSA that receives no Referred In demand, this is the end of the calculation but the problem is that some PSAs continue to refer their new adjusted demand onto another location. And the important part to remember is that it may not always be sending on 100% of it's demand.
To calculate this, I wrote a measure that subtracts the demand that the PSA already referred out from its Adjusted Demand, and then multiplie by the referral percentage. I subtract the referred out demand from the first cycle first so that I don't double dip on the referral percentage. Also, some PSAs will have already referred out 100% of their demand, so the subtraction will result in a negative number. I handle that with a conditional statement that just returns zero.
The measure looks like this:2nd Reff =VAR __remainder = ([1st_ADJUSTED_DEMAND] - [REFERRED_OUT]) * SUM(SDP_OUTPUT[REFERRAL_PERC])VAR __solution = IF(__remainder > 0, __remainder, 0)RETURN__solution
This is the new demand that each PSA has to send on to another PSA for the second cycle.
To get to the demand at the end of the second cycle of referral, I have a measure like this:2nd_ADJUSTED_DEMAND =VAR __currentPSA = MAX(SDP_OUTPUT[PSA_NAME])VAR __currentSTOP = MAX(SDP_OUTPUT[STOP_CODE])VAR __currentRefL = MAX(SDP_OUTPUT[REFERRAL_LOCATION])VAR __solution =CALCULATE(SUMX(SDP_OUTPUT, ([2nd Reff])),REMOVEFILTERS(SDP_OUTPUT), SDP_OUTPUT[REFERRAL_LOCATION] = __currentPSA && SDP_OUTPUT[STOP_CODE] = __currentSTOP)RETURNIF(__currentPSA = __currentRefL,[1st_ADJUSTED_DEMAND] + __solution,[1st_ADJUSTED_DEMAND] - [2nd Reff])
I get the sum of all of the demand I calculated for the second cycle, and add it to the demand calculated in at the end of the first cycle, but only for each PSA where it is programmed not to refer demand on to any location other than its-self (this is the default setting for PSA referral locations if they are not referring demand on). If the PSA dose have a referral location, then I subtract this second cycle of referred out demand from the first cycle of demand.
This solution with a series of measures works. However, it is predicated on the fact that at most there are 2 cycles of referral. Either the demand goes from A to B, or from A to B to C.
I am certain there has to be a way to combine all of these measures into a single calculation - someone better than me in writing DAX could figure out... Then, if I could just have the measure understand the referral pattern first... If I could calculate the number of referrals/relationships... that would be the final solution.- nickc_innova
lbendlin, I made some progress yesterday.
I didn't know DAX had functions to measure parent/child relationships. I can leverage these to detect exactly how many PSAs away a given PSA may be on the chain of referral. PSAs at the beginning of the chain with no children will have the largest path length. PSAs at the end of the chain with the most children and/or are only referring to themselves have the shortest path length.
So the larger the path length, the earlier its demand should be calculated. The PSA(s) with the smallest path length are calculated last.
Next, I need to write a looping statement... which I guess isn't possible in DAX but thankfully Greg_Deckler provided us with a write-up on how to work around that. I don't have the code written out yet but I think I have a map for how to get there.- lbendlin
PATH only works if each element appears in only one place of the hierarchy. From your original description it sounded more like you have circular references between PSAs?
The FOR and WHILE constructs you reference will only work for independent computations in each of the loops. As soon as one loop iteration depends on the result of another iteration this will break down. There is a Power Query function List.Accumulate that is a bit more powerful but even that cannot mask the fact that Power BI does not have a concept of global variables.
I think your best chance is still when you provide additional sequencing information (ideally timestamps, or index columns).