Yesterday i worked to prepare example of Correlation Matrix without loading mirror table.
With mirror table You loading your data twice. Then it is easy to make cartesian in the chart, which is input for correl function.
But, if your input data set is big, you allocate more memory.
I foud a way to calculate correlation loading data one time and making additional table with cartesian only for coordinates (dimension data).
But in this data model i couldnt use built-in correl function, because it has 2 input parameters. So, only way i foud, it was to make calculation using correlation formula, with aggr, set analysis, calculated dimension. It works and results are correct, but my formulas are much more complex than using built-in correl and mirror table.
I tried, but didnt find a way to input into correl function (it has 2 input params) data from "different" rows. So, if it is possible to "join" two records from the same table and use result as correl input? I also considered to use rangecorrel, but dont know how to eventually sort data and "transpond it from rows into columns" (range correl expect one row data in turns from two input sets.
Maybe somebody will find better way to resolve problem of correl matrix without mirror table?
I attached simple example, with 3 tables and 2 charts. Top chart is calculated without mirror table (so, using only tables: matrix and cartesian). Bottom chart uses tables matrix and matrix1 (mirror). You may see, that selecting something in "dim" listobox, both dimensions in top charts changes. In bottom chart, to have the same effect, you have to select in both, dim and dim1 listboxes (of corse it is possible to automate selection, but it was not my goal).
Take a look on expressions and dimensions in both charts
It may be useful for somebody, who wants to use calculated dimension, aggr and set analysis.