Warnings when targets::tar_load() but I cannot make a reprex

[psychelzh]

When I call targets::tar_load(), the object is loaded correctly, but there is always a warning (following {reprex} is from my working directory):

Local .Rprofile detected at C:\Users\liang\OneDrive\Documents\Research\cogstruct\.Rprofile

targets::tar_load(config_game)
#> Warning in file_diff_dbl(chr): NAs introduced by coercion

^{Created on 2021-04-24 by the reprex package (v2.0.0)}

Session info

sessionInfo()
#> R version 4.0.5 (2021-03-31)
#> Platform: x86_64-w64-mingw32/x64 (64-bit)
#> Running under: Windows 10 x64 (build 19042)
#> 
#> Matrix products: default
#> 
#> locale:
#> [1] LC_COLLATE=Chinese (Simplified)_China.936 
#> [2] LC_CTYPE=Chinese (Simplified)_China.936   
#> [3] LC_MONETARY=Chinese (Simplified)_China.936
#> [4] LC_NUMERIC=C                              
#> [5] LC_TIME=Chinese (Simplified)_China.936    
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices datasets  utils     methods   base     
#> 
#> loaded via a namespace (and not attached):
#>  [1] igraph_1.2.6      knitr_1.32        magrittr_2.0.1    tidyselect_1.1.0 
#>  [5] R6_2.5.0          rlang_0.4.10      fansi_0.4.2       stringr_1.4.0    
#>  [9] styler_1.4.1      highr_0.9         targets_0.4.1     tools_4.0.5      
#> [13] data.table_1.14.0 xfun_0.22         utf8_1.2.1        cli_2.4.0        
#> [17] withr_2.4.2       htmltools_0.5.1.1 ellipsis_0.3.1    yaml_2.2.1       
#> [21] digest_0.6.27     tibble_3.1.1      lifecycle_1.0.0   crayon_1.4.1     
#> [25] processx_3.5.1    callr_3.7.0       purrr_0.3.4       codetools_0.2-18 
#> [29] ps_1.6.0          vctrs_0.3.7       fs_1.5.0          glue_1.4.2       
#> [33] evaluate_0.14     rmarkdown_2.7     reprex_2.0.0      stringi_1.5.3    
#> [37] compiler_4.0.5    pillar_1.6.0      backports_1.2.1   renv_0.13.2      
#> [41] pkgconfig_2.0.3

I have no idea how to reproduce it in a new project, but I am sure it is there after upgrading {targets} to the newer version of 0.4.1.

[wlandau]

It's probably the upgrade. It should go away for targets your run with 0.4.1. For now, you can ignore those warnings.

[jarauh]

I have the same problem with targets-0.10.0, and I did not upgrade recently. Using debug(targets:::file_diff_dbl), I can see what the problem is, at least for me. The input is of the following form:

[61] NA                   "t19074.4822615773s" "t19074.4817933748s" "t19074.48231149s"   "t19074.4827771608s"
[66] "t19074.4828071807s" NA                   "t19074,4829786199s" "t19074,483772159s"  "t19074,4839215354s"
[71] NA 

So, it appears that some numbers are printed with commas instead of points. It seems that my meta-information is corrupted in this way. @wlandau Does that information help?

[wlandau]

That's strange. Do you have a reprex? Was the project ever transferred between computers?

[jarauh]

No, I don't switch computers.

Maybe the following details can help: The last target runs rmarkdown::render on a .Rmd-file that sets options(OutDec = ","). All times of that last target are corrupted (but apparently none of the other times). With this option, as.numeric(as.character(1.2)) and the like fail. So my assumption is that target at some place writes the time (which internally is a numeric) as a character taking into account OutDec.

I will try to setup a reprex.

[wlandau]

Ah, knowing about OutDec does help. Thanks! I suspect targets will just need to temporarily set it to "." in one of the methods.

[jarauh]

Ok, I posted the reprex as a new answer. If it helps, I can also post a bug report.

[jarauh]

Here comes the reprex. I adjusted the file _targets from the minimal example targets-minimal:

library(targets)
library(tarchetypes)
source("R/functions.R")
options(tidyverse.quiet = TRUE)
tar_option_set(packages = c("dplyr", "ggplot2", "readr", "tidyr"))
list(
  tar_target(
    raw_data_file,
    "data/raw_data.csv",
    format = "file"
  ),
  tar_target(
    raw_data,
    read_csv(raw_data_file, col_types = cols())
  ),
  tar_target(
    data,
    {
      options(OutDec = ",")
      raw_data %>%
        filter(!is.na(Ozone)) %>% head()
    }
  ),
  tar_target(hist, create_plot(data)),
  tar_target(fit, lm(Ozone ~ Wind + Temp, data)),
  tar_render(report, "index.Rmd")
)

Actually, I think that targets should not change the option, but rather use the corresponding parameter of format or so. But I'm not familiar with the internals of targets.

[wlandau]

Thanks, should be fixed for new pipelines in fb3f9f6. withr::local_options() temporarily modifies OutDec so the option is set back to its original state when the calling function exits.

[jarauh]

Great, thanks. That was quick!

[tjmahr]

Just chiming here because I had this warning and solved it by clearing out a dead target. It seems like one of the hash values had landed in the time column for a target:

m_raw <- readr::read_delim("_targets/meta/meta") 
m_raw[77, ] |> glimpse()

Rows: 1
Columns: 18
$ name       <chr> "file_rate_means"
$ type       <chr> "stem"
$ data       <chr> "049b9b6f9ff3148c"
$ command    <chr> "443fc55068e50d6b"
$ depend     <chr> "ef46db3751d8e999"
$ seed       <dbl> -1279112270
$ path       <chr> "data/rate-means-for-export.csv"
$ time       <chr> "eb429828efa39012"
$ size       <chr> "f1f5b67ba969b5ce"
$ bytes      <dbl> 410479
$ format     <chr> "file"
$ repository <chr> "local"
$ iteration  <chr> "vector"
$ parent     <lgl> NA
$ children   <lgl> NA
$ seconds    <dbl> 0
$ warnings   <chr> NA
$ error      <lgl> NA

Which led to these warnings for me whenever tar_meta() or tar_read() were called. The target described by the metadata didn't actually exist anymore (I had renamed it in the pipeline), so using tar_prune() removed the bad metadata entry and fixed the warning for me.