Hi all,
a customer is asking, how to run large interaction models (400*400 predictors), without your computer collapsing.
Thanks for your help,
Valerie
Have they tried Predictor Screening to pare down the number of variables?
If they are looking for 1000s of one-way analyses, the response screening platform works well here and also employs the false discovery rate adjustment. 400 x's x 400 y's on my machine (2000 observations per column random data) takes about ~ 10 seconds to run in response screening.
But if they want to run a model with lots of terms, @Bill_Worley's strategy to do variable selection to prune down the number of potential predictors is sound.
