expect_equal() 以一种顺序传递参数，如果交换参数则失败

Question

当尝试将testthat::expect_equal() 与两个数字和一个容差参数一起使用时，它会在参数按特定顺序排列时通过，但如果两个数字交换参数位置则失败。我注意到这个函数使用基本包中的all.equal() 引用，并且该函数在交换参数时也有通过/失败的区别。

无论两个函数的前两个参数的顺序如何，我都希望得到相同的答案。如果这不是正确的期望，请告诉我。

library(testthat)

# expect_equal does not throw error with one pair of numbers to compare
expect_equal(5, 1, tolerance=1)

# But does when the two numbers are reversed in their arguments
tryCatch(expect_equal(1, 5, tolerance=1), expectation_failure=conditionMessage)
#> [1] "1 not equal to 5.\n1/1 mismatches\n[1] 1 - 5 == -4\n"

# Since this seems to reference `all.equal()` I tried there too, and see an issue:
all.equal(1, 5, tolerance=1)
#> [1] "Mean absolute difference: 4"
all.equal(5, 1, tolerance=1)
#> [1] TRUE

# My session info:
sessionInfo()
#> R version 3.3.3 (2017-03-06)
#> Platform: x86_64-w64-mingw32/x64 (64-bit)
#> Running under: Windows 7 x64 (build 7601) Service Pack 1
#> 
#> locale:
#> [1] LC_COLLATE=English_United States.1252 
#> [2] LC_CTYPE=English_United States.1252   
#> [3] LC_MONETARY=English_United States.1252
#> [4] LC_NUMERIC=C                          
#> [5] LC_TIME=English_United States.1252    
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] testthat_1.0.2
#> 
#> loaded via a namespace (and not attached):
#>  [1] backports_1.0.5 R6_2.2.1        magrittr_1.5    rprojroot_1.2  
#>  [5] tools_3.3.3     htmltools_0.3.6 yaml_2.1.14     crayon_1.3.2   
#>  [9] Rcpp_0.12.10    stringi_1.1.5   rmarkdown_1.5   knitr_1.15.1   
#> [13] stringr_1.2.0   digest_0.6.12   evaluate_0.10

Answer 1

tl;博士

你的容忍度大于

mean(abs(target-current)) / abs(target)
mean(5 - 1) / 5
4 / 5
0.8  

所以函数在比较 0.8 < 1 之后返回，因为你允许它通过这个容差检查

---

完整答案

all.equal() 函数有参数target 和current

如果您深入研究代码，当target 为 5 而current 为 1 时，它会计算平均绝对差

mean(abs(target - current))
# 4

然后是目标和公差之间的比较，在这种情况下是

5 > 1
## TRUE

因为这是 TRUE，所以它会计算相对差异

4 / 5 
## 0.8

这里，0.8不大于公差，所以它以TRUE退出函数，即

all.equal(5, 1, tolerance = 1)
# [1] TRUE

所以tolerance 的值用于比较相对差异。这就是为什么容差默认为 1.5e-8 的小值的原因

这里我取了all.equal.numeric的相关代码，只去掉了感兴趣的部分，这样你就可以看得更清楚了

allEqual <- function (target, current, tolerance = sqrt(.Machine$double.eps)) 
{
    msg <- NULL
    target <- as.vector(target)
    current <- as.vector(current)
    out <- is.na(target)

    out <- out | target == current

    if (all(out)) 
        return(if (is.null(msg)) TRUE else msg)

    target <- target[!out]
    current <- current[!out]

    xy <- mean(abs(target - current))

    ## THIS BIT HERE vv
    what <- {
        xn <- mean(abs(target))
        # print(paste0("xn: ", xn))
        if (is.finite(xn) && xn > tolerance) {
            #print("xn (target) is GREATER than the tolerance")
            xy <- xy/xn
            #print(paste0("relative difference: ", xy))
            "relative"
        }
        else{
            # print("xn (target) is SMALLER than the tolerance")
            "absolute"
        } 
    }
    ## THIS BIT HERE ^^

    if (is.na(xy) || xy > tolerance) 
        #print("xy is GREATER than the tolerance")
        msg <- c(msg, paste("Mean", what, "difference:", format(xy)))
    if (is.null(msg)) 
        TRUE
    else msg
}

---

allEqual(5, 1, tolerance = 0.79)
# [1] "Mean relative difference: 0.8"
allEqual(5, 1, tolerance = 0.81)
# [1] TRUE
allEqual(5, 1, tolerance = 1)
# [1] TRUE
allEqual(1, 5, tolerance = 1)
# [1] "Mean absolute difference: 4"