Parameter Correlation

You can use the Analysis > Parameter Correlation menu option to check if the parameters are correlated. A strong correlation (+1 or -1) means that two parameters do not independently affect the function node. To determine the correlation, the following happens.

ASCMO-MOCA calculates the gradient matrix G regarding all parameters:

with

F - the optimization function to be minimized
x - training data
p - parameter

ASCMO-MOCA then calculates the covariance matrix C:

with

G^T - transpose of G
I - identity matrix

Then the correlation coefficients c between parameters a and b are calculated. C_ab, C_aa, and C_bb are elements of the covariance matrix.

The results are displayed in the "Parameter Correlation" window.