Plotting pred vs true can look arbitrarily good as long as the range of labels is large enough. Everything gets smushed on the diagonal.
I changed the Node to plot the signed error instead and added an optional dict for supplying desired y limits for the plots to facilitate easy model comparisons.
Plotting pred vs true can look arbitrarily good as long as the range of labels is large enough. Everything gets smushed on the diagonal. I changed the Node to plot the signed error instead and added an optional dict for supplying desired y limits for the plots to facilitate easy model comparisons.