Hi – a question that has some similarities to one asked in the previous post “Addressing cells programmatically”.
I am writing a simple script to replicate an excel-style goal-seek functionality. The script is to work on a matrix with a filter category (‘PhaseName’). The idea is to change the value of a particular item ‘Decline Rate’ (incrementing it up or down), depending on the value returned by the model. Only the value of ‘decline rate’ for the ‘PhaseName’ currently selected in the filter should be changed, however.
I can use the .itemIndex property of the .getFilterItem method to obtain the index number of the currently selected ‘PhaseName’, and can get the particular value for the ‘Decline Rate’ of the currently selected ‘PhaseName’ as follows:
int current = |Phase Details|.getFilterItem(|Phase Details::PhaseName|).itemIndex
|Phase Details::Combined Peak.Decline Rate|.values[current].value
However, it appears that using values with an index number in this way is a read-only property – one can’t set this to a new value.
In the earlier post “Addressing cells programmatically”, I found a useful code snippet from dom, achieving something similar by creating an ArrayList based on all of the different values of the item being changed, changing only that value corresponding to the currently selected filter category, and then writing the ArrayList back to the item. In my case, this would work as follows:
def ArrayList<Item> declines = |Phase Details::Combined Peak.Decline Rate|.values
declines.set(current, |Phase Details::Combined Peak.Decline Rate|.values[current].value-increment)
|Phase Details::Combined Peak.Decline Rate|.values = declines
This does in fact work – but it seems an inordinately complicated way of doing something that users must surely want to do all the time. Is anyone able to suggest a better way of doing this?
On a related note, since when the model is complete, the filter category “PhaseName” will have several hundred ‘phases’ in it, I have a related question on the model.ensureCalculated() method. With only a few phases, calculation times for this model are generally under a second – so if the goal-seek algorithm I have put together takes 6-10 iterations to find an answer, and must calcuate each time, while it is slow, it is not unbearable. Once several hundered phases have been added to the model, a full recalc will almost certainly take well over 10 seconds, meaning 6-10 iterations will become painfully slow – frustrating since in this case, only one of those several hundred phases actually needs to be calculated. Is there any method that can be called that will calculate the model only for the currently selected phase, rather than for all of them? This would seem like an essential ability for anyone with a large, complex model…
Just to add my experience here:
I tried to implement an iterative algorithm of a similar nature to the one that Janek describes, and hit the same issue with EnsureCalculated causing unacceptable runtimes. I ended up dropping the use of Quantrix for this situation.
I’m less convinced that using separate models is a good solution to this requirement, more granular control over recalc is what is required. I don’t understand why Quantrix cannot use the same dirty flags for recalc from within the script engine as it would use from the UI, but I guess for some reason those are not available.
However, whatever the back story, at present I don’t believe Quantrix scripting is viable to implement iterative algorithms on any realistic size of data set.
It’s just as well it has lots of other uses
BTW, many thanks to Luca for pointing out that the range object returned by [font=”Courier New”]<matrix>.getSelection[/font] has a settable [font=”Courier New”]value [/font]property. That is not visible via autocomplete and I hadn’t found reference to it anywhere (so I wonder how he knew… ?) – great feature (pity about the documentation )