Refactor and NA handling for 'decode()'

[ChrisMuir]

The first commit is in reference to issue #34, initial commit to allow decode() to handle NA input. Also added a test to test-Decode.R.

Previous code:

polylines <- c(
  "ohlbDnbmhN~suq@am{tAw`qsAeyhGvkz`@fge}A",
  NA_character_, 
  "ggmnDt}wmLgc`DesuQvvrLofdDorqGtzzV"
)

googlePolylines::decode(polylines)
#> [[1]]
#>      lat       lon
#> 1 26.774 -80.18999
#> 2 18.466 -66.11799
#> 3 32.321 -64.75700
#> 4 26.774 -80.18999

#> [[2]]
#>   lat lon
#> 1 -8e-05 1e-05

#> [[3]]
#>      lat       lon
#> 1 28.745 -70.57899
#> 2 29.570 -67.51400
#> 3 27.339 -66.66800
#> 4 28.745 -70.57899

Updated code:

polylines <- c(
  "ohlbDnbmhN~suq@am{tAw`qsAeyhGvkz`@fge}A",
  NA_character_, 
  "ggmnDt}wmLgc`DesuQvvrLofdDorqGtzzV"
)

googlePolylines::decode(polylines)
#> [[1]]
#>      lat       lon
#> 1 26.774 -80.18999
#> 2 18.466 -66.11799
#> 3 32.321 -64.75700
#> 4 26.774 -80.18999

#> [[2]]
#>   lat lon
#> 1  NA  NA

#> [[3]]
#>      lat       lon
#> 1 28.745 -70.57899
#> 2 29.570 -67.51400
#> 3 27.339 -66.66800
#> 4 28.745 -70.57899

For the encode() cases from issue #34, I'm happy to contribute PR's for them as well. I just really don't know enough about how mapdeck uses googlePolylines::encode() to suggest a best solution though.

[ChrisMuir]

The second commit contains edits to the Rcpp code related to decode(), refactoring for speed. All tests are passing locally. Changes include:

• Return an Rcpp::List from function decode_polyline(), as opposed to an Rcpp::DataFrame. If you assign values to specific attributes of an Rcpp::List, it will "become" a data.frame on the R side, similar to the code/details in this Rcpp Gallary post. This is the source of most the speed up in the benchmarks below.
• Create vectors pointsLat and pointsLon once, pass them by reference to func decode_polyline(), which calls .clear() on them to reuse them in each loop iteration. Originally, these vectors were being created everytime decode_polyline() was getting called.

Benchmarks

Test below is very simplistic and not robust, but the code edits yield a ~250x speed up.

library(googlePolylines)
library(microbenchmark)

# Create vector of 2000 polylines
polylines <- c(
  "ohlbDnbmhN~suq@am{tAw`qsAeyhGvkz`@fge}A",
  "ggmnDt}wmLgc`DesuQvvrLofdDorqGtzzV"
)
polylines <- rep(polylines, 1000)

Original code

microbenchmark::microbenchmark(
  decode(polylines), 
  times = 10
)
#> Unit: milliseconds
#>               expr      min       lq     mean   median       uq      max neval
#> decode(polylines) 473.4338 492.1889 497.8863 497.521 506.2926 517.1034    10

Updated code

microbenchmark::microbenchmark(
  decode(polylines), 
  times = 100
)
#> Unit: milliseconds
#>               expr      min       lq     mean   median       uq      max neval
#>  decode(polylines) 1.797718 1.868518 2.149002 1.921164 2.084554 5.080253   100

[SymbolixAU]

great work @ChrisMuir , I'll review in the coming days, but it all looks good.

[ChrisMuir]

👍 Cool sounds good. If you have any comments/questions, just let me know, I'm happy to make additional edits or tweaks to edits in my commits.

[SymbolixAU]

@ChrisMuir everything looks good - merged into master