Add additional performance metrics

[dsmits]

The original paper has a couple of performance metrics where the output is compared to its central counterpart.

• Mean squared error
• summation of the absolute difference
• maximum difference ratio
• maximum absolute difference

[dsmits]

The maximum difference ratio as described in the paper is confusing. It doesn't take the maximum but is described as

$\sum_{m=1}^{M} \left| \hat{\beta}(m) - \beta(m) \right| /\beta(m)$

I can't find an official definition of MDR so I might skip that one