We've approximately addressed ANA with an ensemble where we drop a single attribute level at random. Now we need to create a ensemble where we are jointing inducing clever randomization on the training data WRT attribute non-attendance and screening.
Let's apply the idea (in aggregate) from the joint strategy for the preliminary ensemble and part-worth coefficients from a single HMNL:
marcdotson
commented
5 years ago
We've approximately addressed ANA with an ensemble where we drop a single attribute level at random. Now we need to create a ensemble where we are jointing inducing clever randomization on the training data WRT attribute non-attendance and screening.
Let's apply the idea (in aggregate) from the joint strategy for the preliminary ensemble and part-worth coefficients from a single HMNL: