1.discovering single quantity analysis metric so to know in case your thought works, the aim is so you may shortly iterate experiements.
2.recall, precission, F1 rating(common of recall and precission)
3.Optimizing metric(the one I care most) and Satisficing metric(2nd precedence, simply good is ok).
4.Select val(dev) set and check check from identical distribution
5.Understanding Human degree accuracy(bias/variance evaluation)
that is based mostly on ML accuracy is decrease than human’s
if if coaching accuracy means too low from human’s degree accuracy: specializing in bias
if if coaching accuracy is analogous from human’s degree accuracy: specializing in variance.
6.Error evaluation is essential, handbook step.
