I like the (hidden) RF in JMP 13.2.1 in the Predictor Screening platform. I think you folks need to ponder Boruta, and make at least 2 iterations based on the sample counts.
I have 300 rows and 10k columns. The default settings don't traverse my space. I have to re-run it about 100 times to get a decent traversal.
for(i=1, i<=100, i++,
Predictor Screening( ... there are 10k columns in here, so you don't want it enumerated ...);
)
So I manually converted to data tables, concatenated into a single large table, the summarized by summing per column name.
It would be nice to do this in JSL. I'm able to make the predictor, and even put it into a loop. I don't know how to make it into a data table, or how to make those data tables to be concatenated.
I have looked at these, but they are so specific to a different platform (distribution, linear fit) that they can't apply, and I don't get the way assignment and piping are working.
If you are offended that I'm not great at JSL, could you point me to the online manuals that I should have read? If not, please feel to walk me through this so I can understand it and make it myself.
