The current R^2 coefficient of determination method in eval_metrics produces confusing and unpredictable results. This PR simplifies this by simply squaring the linear regression correlation co-efficient, as recommended by the stats.linregress function:
https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.linregress.html
Proposed changes
The current R^2 coefficient of determination method in
eval_metricsproduces confusing and unpredictable results. This PR simplifies this by simply squaring the linear regression correlation co-efficient, as recommended by thestats.linregressfunction: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.linregress.html