Sf: vars argument for plot_sf

[hughjonesd]

Related to #457. I appreciate one should do plot(sf_object$geometry, ...), but I think the geometry column is an implementation detail, and many users will just hope to do plot(sf_object) and select the variable(s) they want to plot. So, could I suggest an argument to the plot function that selects one or more variables to plot? If you're agreeable to this, I could try to provide a pull request.

[Nowosad]

Correct me if I'm wrong, but you can do that right now by plot(sf_object["var_name"]).

[hughjonesd]

I guess that is true. It is perhaps not very easy to discover, though.
Intuitively, that looks like plotting a single vector.

David

On 14 October 2017 at 19:08, Jakub Nowosad [e-mail] wrote:

Correct me if I'm wrong, but you can do that right now by
plot(sf_object["var_name"]).

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/r-spatial/sf/issues/516#issuecomment-336653080, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AFjJ98akMLwYfPp8CPnFUbDg5sj43sGWks5ssPi2gaJpZM4P5d3l
.

[lbusett]

On a similar note I find that I often end up just writing plot(sf_object) and then, if my sf object has many columns, I start waiting for all the first 10 columns to be plotted one alongside the other (or have to kill the process because it takes too long depending on the input).

I would too vouch for a "default" behaviour which just plots the geometry, or maybe just the first "column" of the object. If I read the documentation correctly, implementing the second option would just mean to change the default value for max.plot (currently, set to 9).

[edzer]

The "logic", if you wish, of this behaviour is that plot(x) gives you a plot of x, i.e. its entirety. I believe that many people make different assumptions because in sp, plot(x) was synonymous to plot(geometry(x)), and this has been so for over a decade. The disadvantage of that was that for every quick plot, you'd have to start messing with classifying a column and mapping it to colors. Think of what you see with you plot(mtcars): it's not good, but does plot all variables, and pairs.

I agree that in case of many attributes the default of plotting the first 10 is rather arbitrary (but similar to what tibbles do). We could manipulate this by creating a global option (e.g., options("sf.plot_ncol")). I'm not a fan of global options, but would consider a PR doing (and documenting) this in a clean way (see e.g. file AAA.R in package sp).

I don't agree that for and sf object with attributes, plot(x) and plot(st_geometry(x)) should do the same thing.

[hughjonesd]

Maybe just reducing the default would be reasonable. I had the same problem
as Lorenzo of waiting for very large objects to plot.

David

On 15 October 2017 at 14:26, Edzer Pebesma [e-mail] wrote:

The "logic", if you wish, of this behaviour is that plot(x) gives you a
plot of x, i.e. its entirety. I believe that many people make different
assumptions because in sp, plot(x) was synonymous to plot(geometry(x)),
and this has been so for over a decade. The disadvantage of that was that
for every quick plot, you'd have to start messing with classifying a column
and mapping it to colors. Think of what you see with you plot(mtcars):
it's not good, but does plot all variables, and pairs.

I agree that in case of many attributes the default of plotting the first
10 is rather arbitrary (but similar to what tibbles do). We could
manipulate this by creating a global option (e.g., options("sf.plot_ncol")).
I'm not a fan of global options, but would consider a PR doing (and
documenting) this in a clean way (see e.g. file AAA.R in package sp).

I don't agree that for and sf object with attributes, plot(x) and
plot(st_geometry(x)) should do the same thing.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/r-spatial/sf/issues/516#issuecomment-336711493, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AFjJ99bBzZSDioNtWfJEO0g6VTEy3_myks5ssggPgaJpZM4P5d3l
.

[edzer]

On most platforms, plot can be easily interrupted without disturbing your R session. A single map can also be so large that you have to wait for it to plot. And when you're ready to use geom_sf in ggplot2, you'll have to redefine what you meant by "waiting". I think it'll be tough to make everyone happy, here.

[lbusett]

I think the "global option" idea could be a good "compromise" (and could maybe useful also for customizing other options users may wish to customize in the future). I'll see if I can manage to do a PR in the next days.

[etiennebr]

I find using select() very effective:

library(sf)
library(tidyverse)

nc = st_read(system.file("shape/nc.shp", package="sf"), quiet = TRUE)

nc %>% 
    select(AREA, NAME) %>% 
    plot()