The n ext crackpot iteration [Two-Stage / GS Designs]

posted by Helmut Homepage – Vienna, Austria, 2017-08-20 02:20  – Posting: # 17717
Views: 23,598

Capt’n,

some remarks. More maybe tomorrow.

» The polynomial provides a fantastic approximation within our entire interval of interest.

Played around a little. Based on the AIC the 4th degree is the winner indeed.

» ##note: We might not want to write blah^3 etc if we optimize for speed, not sure.

Old wisdom. Here with my coefficients (total sample sizes [not N/seq], exact method for GRM 0.95, 80% power, CV 0.1–1.0).

library(microbenchmark)
a <- c(5.897943, -40.988390, 603.109578, -338.281351, 70.43138)
old.school <- function(a, CV) {
  x <- a[1] + a[2]*CV + a[3]*CV*CV + a[4]*CV*CV*CV + a[5]*CV*CV*CV*CV
  x + (2 - x %% 2)
}
lazy <- function(a, CV) {
  x <- a[1] + a[2]*CV + a[3]*CV^2 + a[4]*CV^3 + a[5]*CV^4
  x + (2 - x %% 2)
}
res <- microbenchmark(old.school(a, CV), lazy(a, CV), times=500L,
                      control=list("random", warmup=10))
boxplot(res, boxwex=0.25, las=1)
options(microbenchmark.unit="us")
print(res)


» ## perhaps ceil would be better?

I would round up to the next even (as above).

» You see, the sample size estimates are sort of almost perfect already. If you want to remove the very few 1's and -1's then just increase the polynomial degree above.

That doesn’t help. With CV <- seq(0.1, 1, 0.01) I got a match in 46/91 and +2 in 45/91. OK, conservative.

Cheers,
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

Complete thread:

Activity
 Admin contact
20,250 posts in 4,262 threads, 1,398 registered users;
online 15 (0 registered, 15 guests [including 4 identified bots]).
Forum time (Europe/Vienna): 19:19 CET

No written law has ever been more binding than
unwritten custom supported by popular opinion.    Carrie Chapman Catt

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5