Problem with contour(): typo

Dear all,

sorry, my last message contained a typo. Correct:

library(PowerTOST)
x??? <- 0.90
y??? <- 0.35
res? <- as.numeric(sampleN.TOST(theta0 = x, CV = y, design = "2x2x4",
   ??????????????????????????????? method = "central", details = FALSE,
   ??????????????????????????????? print = FALSE)[7:8])
mesh <- 28
ys?? <- unique(sort(c(y, seq(y*.8, y*1.2, length.out = mesh))))
xs?? <- unique(sort(c(x, seq(x*0.95, 1, length.out = mesh))))
z??? <- matrix(nrow = length(ys), ncol = length(xs),
   ?????????????? dimnames = list(paste0("y.", signif(ys, 5)),
   ?????????????????????????????? paste0("x.", signif(xs, 5))))
for (i in seq_along(ys)) {
   ? for (j in seq_along(xs)) {
   ??? z[i, j] <- suppressMessages(
   ???????????????? power.TOST(CV = ys[i], theta0 = xs[j], design = "2x2x4",
   ??????????????????????????? method = "central", n = res[1]))
   ? }
}
z??? <- z[nrow(z):1, ncol(z):1]?????????      # reverse rows & columns
z[paste0("y.", y), paste0("x.", x)] == res[2] # should be TRUE
contour(xs, ys, z, nlevels = 20, las = 1, labcex = 1,
   ??????? xlab = "x", ylab = "y", main = paste("n =", res[1]))
abline(h = y, v = x, lty = 2)
points(x, y, col = "red", cex = 1.5)
text(x, y, labels = signif(z[paste0("y.", y), paste0("x.", x)], 6),
   ???? col = "red", adj = c(-0.1, 1.5))


Helmut

You're doing a lot of manipulation of the z matrix; I haven't followed 
all of it, but that's where I'd look for problems.? Generally if you 
keep your calculation of the z matrix very simple you are better off. 
For example, once you have xs and ys in the form you want, calculate z as

z <- outer(x,y, function(x,y) ...)

and then plot it using contour(x, y, z, ...)

The only tricky issue here is that the function needs to be 
vectorized, so if your power.TOST function doesn't accept vectors for 
CV and theta0,
you'll need to put it in a wrapper that does (perhaps using the 
Vectorize function).

Dear Duncan,

Duncan Murdoch wrote on 2020-09-28 21:47:

You're doing a lot of manipulation of the z matrix; I haven't followed
all of it, but that's where I'd look for problems.? Generally if you
keep your calculation of the z matrix very simple you are better off.
For example, once you have xs and ys in the form you want, calculate z as

z <- outer(x,y, function(x,y) ...)

The only tricky issue here is that the function needs to be
vectorized, so if your power.TOST function doesn't accept vectors for
CV and theta0,
you'll need to put it in a wrapper that does (perhaps using the
Vectorize function).

Here I'm lost. power.TOST(theta0, CV, ...) vectorizes properly for
theta0 _or_ CV but no _both_. Hence
library(PowerTOST)
power.TOST(theta0 = c(0.9, 0.95, 1), CV = 0.25, n = 28)
and
power.TOST(theta0 = 0.95, CV = c(0.2, 0.25, 0.3), n = 28)
work, whereas
power.TOST(theta0 = c(0.9, 0.95, 1), 0.95, CV = c(0.2, 0.25, 0.3), n = 28)
not. Of note, we will throw an error in the next release if both
arguments are vectors.

I tried
f <- function(x, y) {
  ? power.TOST(theta0 = x, CV = y, n = 28)
}
x <- unique(sort(c(0.95, seq(0.95*0.95, 1, length.out = 28))))
y <- unique(sort(c(0.25, seq(0.25*0.8, 0.25*1.2, length.out = 28))))
Vectorize(f, c("x, y"), SIMPLIFY = "array")
which is obviously not correct.

On 29/09/2020 5:37 a.m., Helmut Sch?tz wrote:

Here I'm lost. power.TOST(theta0, CV, ...) vectorizes properly for
theta0 _or_ CV but no _both_. Hence
library(PowerTOST)
power.TOST(theta0 = c(0.9, 0.95, 1), CV = 0.25, n = 28)
and
power.TOST(theta0 = 0.95, CV = c(0.2, 0.25, 0.3), n = 28)
work, whereas
power.TOST(theta0 = c(0.9, 0.95, 1), 0.95, CV = c(0.2, 0.25, 0.3), n 
= 28)
not. Of note, we will throw an error in the next release if both
arguments are vectors.

I wouldn't do that, because it doesn't fit the usual R style. It's 
very common for functions to allow vector inputs in several arguments, 
and match up corresponding values to form a vector result.

If you want to use Vectorize, the command would be

? power.TOST.vectorized <- Vectorize(power.TOST, c("theta0", "CV"))

On 29. Sep 2020, at 12:43, Helmut Sch?tz <helmut.schuetz at bebac.at> wrote:

Dear Duncan,

Duncan Murdoch wrote on 2020-09-29 11:57:

On 29/09/2020 5:37 a.m., Helmut Sch?tz wrote:

Here I'm lost. power.TOST(theta0, CV, ...) vectorizes properly for
theta0 _or_ CV but no _both_. Hence
library(PowerTOST)
power.TOST(theta0 = c(0.9, 0.95, 1), CV = 0.25, n = 28)
and
power.TOST(theta0 = 0.95, CV = c(0.2, 0.25, 0.3), n = 28)
work, whereas
power.TOST(theta0 = c(0.9, 0.95, 1), 0.95, CV = c(0.2, 0.25, 0.3), n = 28)
not. Of note, we will throw an error in the next release if both
arguments are vectors.

I wouldn't do that, because it doesn't fit the usual R style. It's very common for functions to allow vector inputs in several arguments, and match up corresponding values to form a vector result.

I see. Here it would require to give the result as a matrix, data.frame,... Substantial change in the code though doable.

If you want to use Vectorize, the command would be

  power.TOST.vectorized <- Vectorize(power.TOST, c("theta0", "CV"))

library(PowerTOST)
theta0 <- unique(sort(c(0.95, seq(0.95*0.95, 1, length.out = 10))))
CV     <- unique(sort(c(0.25, seq(0.25*0.8, 0.25*1.2, length.out = 10))))
power.TOST.vectorized <- Vectorize(power.TOST, c("theta0", "CV", "n"))
and
power.TOST.vectorized(theta0 = theta0, CV = CV, n = 28)
gives the diagonal elements of the desired 11*11 matrix:
z <- matrix(ncol = length(theta0), nrow = length(CV))
for (i in seq_along(CV)) {
  z[i, ] <- power.TOST(theta0 = theta0, CV = CV[i], n = 28)
}

Helmut

-- 
Ing. Helmut Sch?tz
BEBAC ? Consultancy Services for
Bioequivalence and Bioavailability Studies
Neubaugasse 36/11
1070 Vienna, Austria
E helmut.schuetz at bebac.at
W https://bebac.at/
F https://forum.bebac.at/

I would assume the following snippet does what you want - note the use 
of outer with anonymous function wrapping powerTOST:

z <- outer(xs, ys, function(x, y)power.TOST(CV = y, theta0 = x, design 
= "2x2x4", method = "central", n = res[1]))
contour(xs, ys, z)

Dear Benno,

THX, you made my day! Case closed.

Helmut

Puetz, Benno wrote on 2020-09-29 13:14:

I would assume the following snippet does what you want - note the use
of outer with anonymous function wrapping powerTOST:

z <- outer(xs, ys, function(x, y)power.TOST(CV = y, theta0 = x, design
= "2x2x4", method = "central", n = res[1]))
contour(xs, ys, z)

On 29. Sep 2020, at 15:12, Duncan Murdoch <murdoch.duncan at gmail.com> wrote:

That won't work unless power.TOST is vectorized.  outer() will pass it vectors of x and y values.

Duncan Murdoch

On 29/09/2020 8:11 a.m., Helmut Sch?tz wrote:

Dear Benno,
THX, you made my day! Case closed.
Helmut
Puetz, Benno wrote on 2020-09-29 13:14:

I would assume the following snippet does what you want - note the use
of outer with anonymous function wrapping powerTOST:

z <- outer(xs, ys, function(x, y)power.TOST(CV = y, theta0 = x, design
= "2x2x4", method = "central", n = res[1]))
contour(xs, ys, z)

As I noted in my earlier post, it does - had checked that ;-)

It works by taking corresponding pair fo the input vectors (after possible recycling, as eluded by Helmut in his remark on working on only one vector) as needed for outer.

Thanks for the reminder, though,

	Benno

On 29. Sep 2020, at 15:12, Duncan Murdoch <murdoch.duncan at gmail.com> wrote:

That won't work unless power.TOST is vectorized.  outer() will pass it vectors of x and y values.

Duncan Murdoch

On 29/09/2020 8:11 a.m., Helmut Sch?tz wrote:

Dear Benno,
THX, you made my day! Case closed.
Helmut
Puetz, Benno wrote on 2020-09-29 13:14:

I would assume the following snippet does what you want - note the use
of outer with anonymous function wrapping powerTOST:

z <- outer(xs, ys, function(x, y)power.TOST(CV = y, theta0 = x, design
= "2x2x4", method = "central", n = res[1]))
contour(xs, ys, z)

On 29/09/2020 9:16 a.m., Puetz, Benno wrote:

As I noted in my earlier post, it does - had checked that ;-)

Great!