ARROW-13626: [R] Bindings for log base b

[paleolimbot]

Implements log(x, base = (something other than 2, e, or 10)).

[github-actions]

https://issues.apache.org/jira/browse/ARROW-13626

[github-actions]

⚠️ Ticket has not been started in JIRA, please click 'Start Progress'.

[nealrichardson]

Thanks for doing this! If you don't have one already, can you sign up for an ASF Jira account so I can assign the issue to you?

[nealrichardson]

Looks like the old Windows build failed--perhaps the magrittr improvement that supports this is too new for 3.6 binary packages. So maybe the best option is to move the suppressWarnings and the accompanying comment to be around/outside of expect_dplyr_equal().

[paleolimbot]

I uncovered another corner case in testing this (log(10, base = 1) evaluates to Inf in both R and Arrow). I think the latest commit does this well without silencing any warnings that come out of the other (more common) cases!

[edponce]

log(x) with base 1 is not a reasonable calculation because this implies solving the following equation 1^y = x which is only defined for x = 1. Nevertheless, most math libraries will yield a value of infinity for base 1 logarithms when x != 1 and NaN for x = 1. Arrow C++ follows this scheme.

These cases are missing from the tests.

[paleolimbot]

I added a test in 812f941 that ensures NaN for log(1, base = 1) and Inf for log(10, base = 1) in both dplyr and arrow. Is that test reasonable or should it be rewritten/clarified?

[edponce]

I think it is reasonable. I also noticed that these corner cases are not tested in the C++ log tests, but that corresponds to another PR.

[paleolimbot]

Sure! I should be 'paleolimbot' in ASF Jira.

[paleolimbot]

Fixed the base with length != 1 issue, although I think my error message could be better since base with length != 1 is supported, just not as a non-expression. Reprex summary of new functionality:

library(arrow)
library(dplyr, warn.conflicts = FALSE)

# with base as a column
RecordBatch$create(a = 2:5, b = 2:5) %>% 
  mutate(log(a, base = b)) %>% 
  collect()
#> # A tibble: 4 × 3
#>       a     b `log(a, base = b)`
#>   <int> <int>              <dbl>
#> 1     2     2                  1
#> 2     3     3                  1
#> 3     4     4                  1
#> 4     5     5                  1
# with base as a scalar
RecordBatch$create(a = 2:5) %>% 
  mutate(log(a, base = 3)) %>% 
  collect()
#> # A tibble: 4 × 2
#>       a `log(a, base = 3)`
#>   <int>              <dbl>
#> 1     2              0.631
#> 2     3              1    
#> 3     4              1.26 
#> 4     5              1.46
# errors for base with length != 1
RecordBatch$create(a = 2:5) %>% 
  mutate(log(a, base = 3:4)) %>% 
  collect()
#> Warning: In log(a, base = 3:4), base with length != 1 not supported by Arrow;
#> pulling data into R
#> # A tibble: 4 × 2
#>       a `log(a, base = 3:4)`
#>   <int>                <dbl>
#> 1     2                0.631
#> 2     3                0.792
#> 3     4                1.26 
#> 4     5                1.16

Should x as a scalar be supported? It currently triggers bringing the data in to R but could theoretically not:

# errors
RecordBatch$create(a = 2:5) %>% 
  mutate(a * log(2)) %>% 
  collect()
#> Warning: Expression a * log(2) not supported in Arrow; pulling data into R
#> # A tibble: 4 × 2
#>       a `a * log(2)`
#>   <int>        <dbl>
#> 1     2         1.39
#> 2     3         2.08
#> 3     4         2.77
#> 4     5         3.47
# works!
RecordBatch$create(a = 2:5) %>% 
  mutate(a * log(Expression$scalar(2))) %>% 
  collect()
#> # A tibble: 4 × 2
#>       a `a * log(Expression$scalar(2))`
#>   <int>                           <dbl>
#> 1     2                            1.39
#> 2     3                            2.08
#> 3     4                            2.77
#> 4     5                            3.47

[paleolimbot]

With updated handling of x:

library(arrow, warn.conflicts = FALSE)
library(dplyr, warn.conflicts = FALSE)

# with base as a column
RecordBatch$create(a = 2:5, b = 2:5) %>% 
  mutate(log(a, base = b)) %>% 
  collect()
#> # A tibble: 4 × 3
#>       a     b `log(a, base = b)`
#>   <int> <int>              <dbl>
#> 1     2     2                  1
#> 2     3     3                  1
#> 3     4     4                  1
#> 4     5     5                  1
# with base as a scalar
RecordBatch$create(a = 2:5) %>% 
  mutate(log(a, base = 3)) %>% 
  collect()
#> # A tibble: 4 × 2
#>       a `log(a, base = 3)`
#>   <int>              <dbl>
#> 1     2              0.631
#> 2     3              1    
#> 3     4              1.26 
#> 4     5              1.46
# with x as a scalar
RecordBatch$create(a = 2:5) %>% 
  mutate(log(2, base = a)) %>% 
  collect()
#> # A tibble: 4 × 2
#>       a `log(2, base = a)`
#>   <int>              <dbl>
#> 1     2              1    
#> 2     3              0.631
#> 3     4              0.5  
#> 4     5              0.431
# errors for base with length != 1
RecordBatch$create(a = 2:5) %>% 
  mutate(log(a, base = 3:4)) %>% 
  collect()
#> Warning: In log(a, base = 3:4), base must be a column or a length-1 numeric;
#> other values not supported by Arrow; pulling data into R
#> # A tibble: 4 × 2
#>       a `log(a, base = 3:4)`
#>   <int>                <dbl>
#> 1     2                0.631
#> 2     3                0.792
#> 3     4                1.26 
#> 4     5                1.16
# errors for x with length != 1
RecordBatch$create(a = 2:5) %>% 
  mutate(log(a, base = 3:4)) %>% 
  collect()
#> Warning: In log(a, base = 3:4), base must be a column or a length-1 numeric;
#> other values not supported by Arrow; pulling data into R
#> # A tibble: 4 × 2
#>       a `log(a, base = 3:4)`
#>   <int>                <dbl>
#> 1     2                0.631
#> 2     3                0.792
#> 3     4                1.26 
#> 4     5                1.16

^{Created on 2021-09-14 by the reprex package (v2.0.1)}

[nealrichardson]

One note about a test I think we need to add but otherwise this is excellent, thank you!