Thanks Hamel for documenting your experiences in a wonderful real world use case.
Pradeep Reddy Raamana

Pradeep Reddy Raamana I cannot too many specifics as that will reveal more information than I am allowed to share. What I can share is the validation and holdout sets are time-partitioned such that the training set is from a different time period than the validation and holdout sets. The unit of observation for this modeling exercise is a (person, date) pair.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.