mkborregaard
commented
5 years ago Then @yeesian: Yeesian Ng [5:18 AM] I think the best way forward is to create the package you want to be using for your daily needs, and to just start using it. And feel free to evangelize it to people if it’s something you feel like maintaining and supporting. @juliohm has been developing and maintaining GeoStats.jl successfully for instance. It could be that parts of the JuliaGeo ecosystem may mature and make things easier in the future, but there’s no timeline for it (since there’s no consistent developmental effort that we can count on). So IMO people shouldn’t refrain from working on their own packages for processing rasters/etc in the meantime. I am interested in the questions you’re bringing up too, e.g. when you write: “Here are a number of issues. First of all, VerySimpleRasters doesn’t know about Shapefiles. It just knows about polygons, and that they are a vector of Tuples. Maybe it would be useful for Shapefiles to define a polygon in accordance with GeometryTypes? Maybe GeometryTypes could satisfy the Geointerface, or the other way around? So, first I have to extract the points as a Tuple. Then, when I plot it, I see there is a problem: There are actually two polygons here, but they are treated as one. Since this breaks my masking code (but not R’s), I have to define a function to split the Polygons whenever a line returns to the starting point.” Ideally GeoInterface should have a sufficiently granular interface for others to use. The most common comprehensive one that I know of is https://www.opengeospatial.org/standards/sfa. But it will be hard to provide and maintain implementations for the full spec upfront, and it’ll be more reasonable for us to build up to it as we go (according to demonstrated need).
yeesian
rafaqz
juliohm
evetion
tpoisot
visr
Ported from Slack. Me:
Over the last week or two I’ve been working on very simple raster implementation in pure julia. The reason for doing that is that when I had a workshop earlier this month, with 30 very interested biogeographers at a conference, I asked them what it would take for them to move to Julia. And they almost all said “convenient raster support”. And - it is my impression that that support is not quite there yet in Julia. Of course you can do amazing things with ArchGDAL, but in truth almost all of my work with rasters involve getting data out of them and using that for analyses, and some plotting.
So, I designed an analysis in R: https://www.dropbox.com/s/n8790r0s63d56hm/R%2Bworked%2Braster%2Bexample.html?dl=0 and set out to create a package that could replicate it in Julia. The result is https://github.com/mkborregaard/VerySimpleRasters.jl and the replicated analysis is here: https://www.dropbox.com/s/d5mc52corz5hm7s/VerySimpleRasters%2Bexample.html?dl=0
All operations happen on disk, out of memory, and the script is very fast even for the relatively big raster in the example (except the predictable waiting time on plots).
There’s still lots to do - so far it uses R’s native raster format as it’s own native format, and supports importing ESRI Ascii grids to that format. Other formats could possibly be provided by GDAL in some way like the old RasterIO package?
I’d really love inputs - how to make this consistent with the Geointerface, is it worth continuing in this direction or have I overlooked some obvious existing capabilities etc.
For anyone wanting to play with the package, notebooks and data, I’ve collected everything in a nice folder here: https://www.dropbox.com/sh/4shiomz5gj4ggx5/AAASQv_K5p2OadhwL-CdRuEja?dl=0
cc @yeesian @visr @c42f