Model analysis

Buttons at the bottom of the Base + Delta Model Grid can be used to show validation feedback about regressed models:

• - click to perform a Model Analysis

• - click to highlight zero values.

Model Analysis compares the results of regression to a set of validation rules to ensure that the regressed model matches typical expected behaviour. Problems in the regressed data are highlighted, and further information on the problem can be found by hovering the mouse cursor over the indicated cell.

The Preferences dialog contains several settings which control the thresholds for use in showing analysis feedback (Base + delta settings).

The difference in delta values between different bases should not be too small, as the model is non-linear.

The change in property value is expected to be monotonic across the the bases.

The difference in values between bases should not be too big; if the difference is large it may indicate a problem with the data used for regression

If the average size of the delta value relative to the base value is too high there may be a problem with the model, as large changes in step size could result in unrealistic behaviour such as negative values where only positives are expected.

If the average size of the delta value relative to the base value is zero (as defined by the minimum threshold in the Preferences (Base + delta settings) then zeroing of the delta may be considered to remove any possible adverse consequences from curvature or large relative differences.