By week 5 we want to have clusters and some stories to tell.
Plot the hell out of each feature that we want to use
dump into hive
pick the features we want
write some hive queries to get histograms
get a first Hive example running by the end of Friday
Normalize / Preprocess features
fit to [-1,1]?
log, sqrt?
throw away outliers?
ready to meet with Jeff and Andreas by 4/19
Ask Jeff about these strategies on Friday, 4/13
Bucket engagements by session
should be a simple hive job
Decay sessions by timestamp
look into how to decay values?
trial and error
Cluster!
k-means?
hierarchical?
EM
Review clusters (mit deutschem Bier (Bitburger is AndreasÕs favorite))
present in week 5