K***a
发帖数: 72 | 1 我在做一个Response model，有40k responders，也就是 “1's”，我应该放多少“0
”（non-responders）呢？另外是不是应该用所有的40k呢？我有100多variables。大
家一般会怎么做？ | s*r
发帖数: 2757 | 2 from the viewpoint of a case-control study, people usually have at most 4 '0
' for every '1'. A ratio beyond that gives very little power advantage | K***a
发帖数: 72 | 3 Thanks. How about the overall size then? if I take all 40k 1's, 3x40k 0's,
total will be 160k, split to training and validation, each will have 80k,
with over 100 variables, is that too large? Is there a rule about the size
of the data? | s*r
发帖数: 2757 | 4 agresti's book categorical data analysis 2nd, page 212 | K***a
发帖数: 72 | 5 Thanks! I'll check that.
【在 s*r 的大作中提到】
: agresti's book categorical data analysis 2nd, page 212
|
|
