runSD does not return 0 for TTR::runSD(rep(0.001,12))

[vaclavcepelak]

Dear Joshua,

the following code TTR::runSD(rep(0.001,12))

should in my opinion return vector of 9 NAs and 3 zeros but returns [1] NA NA NA NA NA NA NA NA NA [10] 0.000000e+00 2.285699e-19 2.285699e-19

I haven't found this issue reported earlier. Thank you very much for looking at it!

Vaclav

[vaclavcepelak]

Here is my sessionInfo if relevant:

Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=Czech_Czechia.1250  LC_CTYPE=Czech_Czechia.1250    LC_MONETARY=Czech_Czechia.1250
[4] LC_NUMERIC=C                   LC_TIME=Czech_Czechia.1250    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] eventdetection_0.0.0.9000 data.table_1.10.4-3      

loaded via a namespace (and not attached):
 [1] httr_1.3.1           tidyr_0.7.2          jsonlite_1.5         modelr_0.1.1         assertthat_0.2.0    
 [6] TTR_0.23-2           cellranger_1.1.0     backports_1.1.1      lattice_0.20-35      glue_1.2.0          
[11] digest_0.6.12        rvest_0.3.2          colorspace_1.3-2     plyr_1.8.4           psych_1.7.8         
[16] pkgconfig_2.0.1      devtools_1.13.4      broom_0.4.3          haven_1.1.0          purrr_0.2.4         
[21] scales_0.5.0         tibble_1.3.4         ggplot2_2.2.1        withr_2.1.0          tidyverse_1.2.1     
[26] lazyeval_0.2.1       mnormt_1.5-5         magrittr_1.5         crayon_1.3.4         readxl_1.0.0        
[31] memoise_1.1.0        nlme_3.1-131         forcats_0.2.0        xts_0.10-0           xml2_1.1.1          
[36] foreign_0.8-69       tools_3.4.3          hms_0.4.0            stringr_1.2.0        munsell_0.4.3       
[41] lambda.r_1.2         bindrcpp_0.2         compiler_3.4.3       rlang_0.1.4          futile.logger_1.4.3 
[46] grid_3.4.3           rstudioapi_0.7       testthat_2.0.0       gtable_0.2.0         roxygen2_6.0.1      
[51] curl_3.0             reshape2_1.4.2       R6_2.2.2             zoo_1.8-0            lubridate_1.7.1     
[56] dplyr_0.7.4          bindr_0.1            commonmark_1.4       rprojroot_1.2        futile.options_1.0.0
[61] readr_1.1.1          desc_1.1.1           stringi_1.1.6        parallel_3.4.3       Rcpp_0.12.14

[joshuaulrich]

That's a consequence of floating point arithmetic. David Goldberg provides a great overview in What Every Computer Scientist Should Know About Floating-Point Arithmetic.

I agree it would be nice to less rounding error, but there's a trade-off between speed and accuracy. It would also be non-trivial to implement a patch for greater accuracy. You should compare the absolute value of the difference of two numbers if you want to know whether they're significantly different from each other.

[vaclavcepelak]

Many thanks for an explanation.

[joshuaulrich]

Closing this, since I am very unlikely to implement a change to address this behavior.