Parameter Sensitivity

You can use the Analysis > Normalized Parameter Sensitivity menu option to check the influence of parameters on function nodes.

ASCMO-MOCA calculates the gradients G of a node regarding the parameters for all parameters p_j for all training data points x_i:

with

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

The gradient is normalized to the range of the parameter:

with

• u_j - upper limit of parameter p_j
• l_j - lower limit of parameter p_j

The results are displayed in the "Normalized Parameter Sensitivity" window as follows:

• dark gray area: maximum gradient regarding one parameter p_j over all training data points
• red line: mean gradient regarding one parameter p_j over all training data points
• light gray area: mean gradient ± 1 σ regarding one parameter p_j over all training data points

Note
Smaller values indicate less influence of the parameters on a node.

See also

Normalized Parameter Sensitivity