nonworking 'nls' method frmo base 'stats' package

[donarus]

The 'stats' package is one of the base packages preinstalled with R/FastR

there is a method NLS (non-linear least squares method)

In GNU R the following example work perfectly fine

regressionData <- data.frame(
  x = c(1, 2, 3, 4),
  y = c(1.5, 3.3, 99, 16)
)
startingParams <- list(
  a = 1,
  q = 2
)
mNLS <- nls(formula = "y ~ a*x^q",
            data = regressionData,
            start = startingParams)

but in FastR it produces an error, although the input is valid.

Error in nlsModel(formula, mf, start, wts) :
  singular gradient matrix at initial parameter estimates

PS: If you want, I have multiple other examples of nonworking chunks from base packages, but I'm unsure where to track them

[steve-s]

Hi Tadeas,

thank you for the report. It boils down to issue with numericDeriv:

env <- new.env()
env$a <- 1
env$q <- 2
env$x <- c(1,2,3,4)
env$y <- c(1.5, 3.3, 99.0, 16.0)
numericDeriv(quote(a * x^q), c("a", "q"), env)

gives a wrong result on FastR (gradient matrix is all zeroes). We are working on a fix.

If you want, I have multiple other examples of nonworking chunks from base packages, but I'm unsure where to track them.

yes please, open GitHub issues or you can put them in one issue to make it simpler for you.

It may be useful to distinguish issues that are blockers for your use-case and those that have a workaround. The most useful bug reports for us are those that point to an R builtin/primitive function or native function (called via .Call/.Fortran etc.) that gives different result on FastR, but any bug reports are welcome.

[steve-s]

Should be fixed in 86fafa2