Closed mfitz closed 1 year ago
I have updated to work with py3.9-3.11 in #188. However, it comes with a slight change in the result of reprojections. This is caused by updating the PROJ package from v7.x to >v7.x (ideally v9.x). The cause is a change in the underlying steps taken by PROJ to reproject (expand below to see the steps).
I would be inclined to say that reprojection is more correct now than it was before. If you agree @KasiaKoz then I would update the expected coordinate values in the failing tests to the result of the newer PROJ reprojections.
I ran a test: I generated our small Londinium test network in epsg:4326
, then projected to epsg:27700
and then back to epsg:4326
, using two set-ups:
proj --version Rel. 4.9.2, 08 September 2015
on an old macproj --version Rel. 9.1.0, September 1st, 2022
on an M1 macmain (636a8b7
) branch of genet but M1 packages are closer to this because the older versions of packages from main branch don't work
TLDR:
epsg:4326->epsg:27700->epsg:4326
produces the same network on both set-ups - at least looking at pictures, I didn't check numericallyepsg:27700
output on proj v4
is visibly worse proving your inclination @brynpickering; on proj 9
is almost identical to epsg:4326
(reprojected to epsg:27700
by qgis to be viewed in that projection)epsg:4326->epsg:27700->epsg:4326
:
epsg:27700
networks:
GeNet does not install in a Python 3.11.0 environment. On an EC2 instance running ubuntu:
I can make the pip installation work by upgrading a couple of dependencies in
requirements.txt
:However, the unit test suite then fails with a lot of errors:
An example stack trace:
There are also a fair number of warnings:
Most likely this is a question of upgrading the right dependencies to the correct versions, but this may leave us in a situation where GeNet can no longer be used in a Python 3.7 environment. If we can be backwards compatible, we should.