Prediction error vs. Power

In any AR model, the prediction error variance $\sigma^2$ (See Section 10.4) and the total power (equivalently the zero-th ACF lag) $r_0 = {\cal E}(x_i^2)$ are interchangeable (one can be gotten from the other). But, in a classifier, they may produce different results. We therefore facilitate using one or the other.

Module software/module_acf2rc.m, software/module_ar_mlx.m, and software/module_ar_ml.m allow choosing (argument e0flag default is $r_0$). But, we can get $r_0$ from $\sigma^2$ and $a_1,a_2 \ldots a_P$ with the rlevinson function. The module software/module_as2r0.m is designed for that purpose.