add tidyr::unnest method/support

[jmlondon]

Thanks to 2f5914d, there is now a method for tidyr::nest(). That, obviously, brings up the question of how to support tidyr::unnest().

Here's an example of tidyr::nest() in action with sf objects

data(meuse, package = "sp")
meuse_sf <- st_as_sf(meuse,coords = c("x","y")) %>% group_by(landuse) %>% 
  nest()

> meuse_sf %>% head()
# A tibble: 6 × 2
  landuse               data
   <fctr>             <list>
1      Ah <tibble [39 × 12]>
2      Ga  <tibble [3 × 12]>
3      Ab  <tibble [8 × 12]>
4       W <tibble [50 × 12]>
5      Fh  <tibble [1 × 12]>
6      Ag  <tibble [5 × 12]>

> meuse_sf$data[[1]]
Simple feature collection with 39 features and 11 fields
geometry type:  POINT
dimension:      XY
bbox:           xmin: 178786 ymin: 329822 xmax: 181307 ymax: 333611
epsg (SRID):    NA
proj4string:    NA
# A tibble: 39 × 12
   cadmium copper  lead  zinc  elev       dist    om  ffreq
     <dbl>  <dbl> <dbl> <dbl> <dbl>      <dbl> <dbl> <fctr>
1     11.7     85   299  1022 7.909 0.00135803  13.6      1
2      8.6     81   277  1141 6.983 0.01222430  14.0      1
3      6.5     68   199   640 7.800 0.10302900  13.0      1
4      2.8     48   117   269 7.480 0.27709000   8.7      1
5      3.2     31   132   346 8.217 0.19009400   9.2      1
6      2.5     31   183   504 8.815 0.11393200   8.4      1
7      2.0     27   130   326 8.937 0.16833600   9.1      1
8      1.7     24    85   218 9.020 0.34232100   7.0      1
9      1.8     22    87   240 8.973 0.24986300   5.3      1
10     2.1     27    77   250 8.328 0.27162200   2.4      1
# ... with 29 more rows, and 4 more variables: soil <fctr>,
#   lime <fctr>, dist.m <dbl>, geometry <simple_feature>

In theory, we should be able to meuse_sf %>% tidyr::unnest() and get back our original ungrouped/unnested sf object. But,

> st_as_sf(meuse,coords = c("x","y"))  %>% class()
[1] "sf"         "data.frame"
> meuse_sf %>% unnest() %>% class()
[1] "tbl_df"     "tbl"        "data.frame"

In order to get back to an sf object, we need to

meuse_sf %>% unnest() %>%
  mutate(geometry = st_sfc(geometry)) %>% 
  st_as_sf()

Providing a generalized solution initially strikes me as non-trivial ... And, I have admittedly not spent much time thinking through how to implement such a solution. But, I'd like to, eventually, see a solution that mirrors the tidyr::nest/tidyr::unnest cycle.

[edzer]

I think that on the side of sf we can do little here:

• meuse_sf is of class tbl_df, and I don't think that nest.sf should return a nested simple features object
• when calling unnest on this, the nested sf objects become a tbl_df, and the geometry column becomes a list
• essentially, tidyr drops all the attributes, hence we have to rebuild it using st_sfc
• you lost crs and precision along the way in the example above

I believe the place to improve this would be in tidyr.

[jmlondon]

That all makes sense to me. I'll think about this a bit more and maybe file an issue with tidyr.

Could you expand more on where/how precision was lost in the example? Is there a better approach that would preserve precision?

[edzer]

In the call to st_sfc you don't specify crs or precision; if these were set before, they're lost (for this particular case, only crs gets lost, precision is default, zero)