[Rd] stats::fft produces inconsistent results
bbo|ker @end|ng |rom gm@||@com
Wed Oct 20 03:27:29 CEST 2021
This is a long shot, but here's a plausible scenario:
as part of its pipeline, ravetools::mvfftw computes the mean of the
input vector **and then centers it to a mean of zero** (intentionally or
because variables are passed to compiled code by reference (someone
can feel free to correct my terminology), this means that the original
vector in R now has a mean of zero
the first element of fft() is mean(x)*length(x), so if mean(x) has
been forced to zero, that would explain your issue.
I don't know about the non-reproducibility part.
On 10/19/21 7:06 PM, Dipterix Wang wrote:
> Dear R-devel Team,
> I'm developing a neuroscience signal pipeline package in R (https://github.com/dipterix/ravetools) and I noticed a weird issue that failed my unit test.
> Basically I was trying to use `fftw3` library to implement fast multivariate fft function in C++. When I tried to compare my results with stats::fft, the test result showed the first element of **expected** (which was produced by stats::fft) was zero, which, I am pretty sure, is wrong, and I can confirm that my function produces correct results.
> However, somehow I couldn’t reproduce this issue on my personal computer (osx, M1, R4.1.1), the error simply went away.
> The catch is my function produced consistent and correct results but stats::fft was not. This does not mean `stats::fft` has bugs. Instead, I suspect there could be some weird interactions between my code and stats::fft at C/C++ level, but I couldn’t figure it out why.
> +++ Details:
> Here’s the code I used for the test:
> ————————Test code————————
> x <- rnorm(1000)
> dim(x) <- c(100,10)
> a <- ravetools:::mvfftw_r2c(x, 0)
> c <- apply(x, 2, stats::fft)[1:51,]
> expect_equal(a, c)
> Here are the tests that gave me the errors:
> The test logs on win-builder
> Test logs on GitHub
> —————————————— Failed tests ——————————————
> -- Failure (test-fftw.R:41:3): mvfftw_r2c --------------------------------------
> `a` (`actual`) not equal to `c` (`expected`).
> actual vs expected
> [,1] [,2] [,3] [,4] ...
> - actual[1, ] 10.8887367+ 0.0000000i -3.7808077+ 0.0000000i 2.967354+ 0.000000i 5.160186+ 0.000000i ...
> + expected[1, ] 0.0000000+ 0.0000000i -3.7808077+ 0.0000000i 2.967354+ 0.000000i 5.160186+ 0.000000i...
> The first columns are different, `actual` is the results I produced via `ravetools:::mvfftw_r2c`, and `expected` was produced by `stats::fft`
> Any help or attention is very much appreciated.
> - Zhengjia
> R-devel using r-project.org mailing list
Dr. Benjamin Bolker
Professor, Mathematics & Statistics and Biology, McMaster University
Director, School of Computational Science and Engineering
Graduate chair, Mathematics & Statistics
More information about the R-devel