Rewrite reshaping functions to improve performance

[etiennebacher]

data_to_wide() and data_to_long() are rewritten to use stack() and unstack() instead of reshape() (this was suggested in nathaneastwood/poorman#48). This significantly improves the performance of these two functions and doesn't modify any of the use cases and tests.

Benchmarks

I just show the results here, the code to remake the benchmarks is in WIP/_BENCHMARKS_RESHAPE.R:

### DATA_TO_LONG ==========================================

ex1_l
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old           56.8s    56.8s    0.0176    4.05GB   0.0528
#> 2 new            9.2s     9.2s    0.109      2.1GB   0.109 
#> 3 tidyr         774ms    774ms    1.29      1.14GB   1.29
ex2_l
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old          3.05ms   3.79ms      247.    87.4KB     0   
#> 2 new           1.8ms   2.07ms      462.    83.7KB     4.66
#> 3 tidyr        2.39ms   3.22ms      278.    98.9KB     0
ex3_l
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old        246.58ms 276.52ms      3.63   42.36MB   0.151 
#> 2 new        200.69ms 233.01ms      4.28    5.02MB   0.0873
#> 3 tidyr        3.31ms   4.02ms    233.      1.79MB   4.75
ex4_l
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old         656.3ms  713.5ms      1.32   325.5MB    0.147
#> 2 new         203.1ms  229.4ms      4.22   125.7MB    0.468
#> 3 tidyr        20.3ms   23.2ms     42.2     33.5MB    0
ex5_l
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old         863.8ms  879.7ms      1.12     353MB    0.280
#> 2 new         326.1ms  338.8ms      2.92   111.1MB    0    
#> 3 tidyr        18.3ms   18.6ms     53.5     32.6MB    0

### DATA_TO_WIDE ==========================================

ex1_w
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old          3.68ms   4.41ms      212.     468KB        0
#> 2 new          3.94ms   4.42ms      212.     943KB        0
#> 3 tidyr         4.3ms    4.9ms      193.     436KB        0
ex2_w
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old         522.3ms    1.37s     0.765  445.75MB    0.191
#> 2 new          56.2ms  57.81ms    17.2     11.59MB    0    
#> 3 tidyr        11.1ms  12.63ms    80.9      1.78MB    0
ex3_w
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old           653ms    2.02s     0.545  446.11MB    0.136
#> 2 new          64.9ms  77.16ms    13.2     11.67MB    0    
#> 3 tidyr        12.3ms   16.7ms    61.2      1.81MB    0
ex4_w
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old           1.23s    1.25s     0.784    82.6MB    0.196
#> 2 new        367.93ms 505.28ms     2.10     42.1MB    0.234
#> 3 tidyr       36.44ms  42.25ms    23.0      16.6MB    0
ex5_w
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old           9.22s    9.22s     0.108   557.8MB    0.108
#> 2 new        797.88ms 797.88ms     1.25    287.7MB    0    
#> 3 tidyr      135.69ms 135.69ms     7.37     97.1MB    0
ex6_w
#> # A tibble: 3 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old           2.15m    2.15m   0.00776    6.03GB    0.271
#> 2 new             11s      11s   0.0909     3.25GB    0.182
#> 3 tidyr          2.4s     2.4s   0.416    930.38MB    0.416

[codecov-commenter]

Codecov Report

Merging #285 (bf87604) into main (8698506) will increase coverage by 0.14%.
The diff coverage is 90.00%.

@@            Coverage Diff             @@
##             main     #285      +/-   ##
==========================================
+ Coverage   85.60%   85.74%   +0.14%     
==========================================
  Files          53       54       +1     
  Lines        3695     3753      +58     
==========================================
+ Hits         3163     3218      +55     
- Misses        532      535       +3     

Impacted Files	Coverage Δ
R/data_to_long.R	89.13% <89.13%> (ø)
R/data_to_wide.R	90.62% <90.62%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[IndrajeetPatil]

Thanks for the amazing work, @etiennebacher!

I didn't even know about stack() and unstack() 👀

Feel free to merge when you think this is ready. Also, all code snippets relevant for development, like benchmarking, can live in the /dev folder (https://github.com/easystats/datawizard/tree/main/dev).

[strengejacke]

Sorry, I'm working on a paper revision, and this PR included to much code for a timely review ;-)
But benchmarks are impressive!

[nathaneastwood]

These are really impressive performance improvements! Well done @etiennebacher!