Setting my_lambda and nlambda on MADMMplasso()

[wleoncio]

Quick question: when calling MADMMplasso(), does it make sense to set nlambda != ncol(y)? There doesn't seem to be anything in the code that prevents this, but this bit of code implies that if my_lambda = NULL, lambda_i is created with length ncol(y):

MADMMplasso/R/MADMMplasso.R

Lines 97 to 113 in 081383c

	if (is.null(my_lambda)) {
	r <- y
	lammax <- lapply(
	seq_len(ncol(y)),
	function(g) {
	max(abs(t(X) %*% (r - colMeans(r))) / length(r[, 1])) / ((1 - alpha) + (max(gg[1, ]) * max(CW) + max(gg[2, ])))
	}
	)
	big_lambda <- lammax
	lambda_i <- lapply(
	seq_len(ncol(y)),
	function(g) {
	exp(seq(log(big_lambda[[g]]), log(big_lambda[[g]] * rat), length = maxgrid))
	}
	)

This then breaks this later bit of the code, since the vectors lam[, i] will have a different size than the vectors lambda_i[[i]]:

MADMMplasso/R/MADMMplasso.R

Lines 173 to 177 in 081383c

	if (is.null(my_lambda)) {
	lam <- matrix(0, nlambda, ncol(y))
	for (i in seq_len(ncol(y))) {
	lam[, i] <- lambda_i[[i]]
	}

[Theo-qua]

Hello Waldir,
nlambda cannot be equal to ncol(y). nlambda is for the number of rows of the lambda values and ncol(y) is the number of columns for lambda values (lam).

[wleoncio]

OK, so that means one can't freely assign nlambda to whatever integer they want, right? Isn't nlambda determined by the number of rows on my_lambda (whether passed to MADMMplasso() or calculated inside it)?

[Theo-qua]

nlambda can be any positive integer so it can be freely specified. I was only saying that it not the same as ncol(y).

[wleoncio]

OK, so why does this code give an error (main branch)?

[wleoncio]

The code above gives an error of Error in gg[hh, ] : subscript out of bounds, implying a connection between gg and nlambda (through hh), but according to the documentation gg must be a 2x2 matrix, is that so?

[Theo-qua]

Yes, it gave error because of maxgrid = 1L, nlambda = 2L. According to the documentation, nlambda should be less than or equal to maxgrid.

[Theo-qua]

gg is regenerated in the function with a length equal to maxgrid before the loop.

[wleoncio]

Great, thanks! I read that part, but didn't catch the exception. We should validate that nlambda <= maxgrid to avoid unclear error messages (and unnecessary computation), agreed? I can open a PR with a quick fix.

Regarding gg, I'm confused now. I can't find it being created inside MADMMplasso(), only used to calculate lammax, gg1, gg2 and gg3 (see here). gg even seems like a mandatory argument, as not passing it (or passing NULL, which is the default) results in error.

[Theo-qua]

I agree with you. I reassigned gg to gg3 in the MADMMplasso() after calculating gg3. I did not consider to cause any confusion since this is internal assignment. So gg is changed from the initial gg in the call to gg3.