Contribute¶
This page is also meant for internal documentation.
Editor setup¶
There is only setup files provided for VSCode and no other editor (yet).
A list of extensions and some settings can be found in the .vscode
.
At the first start, VSCode should ask you if you want to install the recommended extension.
The settings should be automaticly used by VSCode.
Both should provide the developers with a better experience and enforce code-style.
Environment setup¶
Prereq:
- Rye:
curl -sSf https://rye.astral.sh/get | bash
- GDAL:
sudo apt update && sudo apt install libpq-dev gdal-bin libgdal-dev
or for HPCconda install conda-forge::gdal
- Clang:
sudo apt update && sudo apt install clang
or for HPCconda install conda-forge::clang_linux-64
If you install GDAL via apt for linux you can view the supported versions here: https://pkgs.org/search/?q=libgdal-dev. For a finer controll over the versions please use conda.
Now first check your gdal-version:
And your CUDA version (if you want to use CUDA):
The GDAL version is relevant, since the version of the python bindings needs to match the installed GDAL version
Now, to sync with a specific gdal
version, add gdalXX
to the --features
flag.
To sync with a specific cuda
version, add cuda1X
or without cuda cpu
.
E.g.:
As of right now, the supported gdal
versions are: 3.9.2 (gdal39
), 3.8.5 (gdal38
), 3.8.4 (gdal384
), 3.7.3 (gdal37
) and 3.6.4 (gdal36
).
If your GDAL version is not supported (yet) please sync without GDAL and then install GDAL to an new optional group. For example, if your GDAL version is 3.8.4:
IMPORTANT! If you installed any of clang or gdal with conda, please ensure that while installing dependencies and working on the project to have the conda environment activated in which you installed clang and or gdal.
Another option is to install the windows GDAL binary wheels compiled by cgoehlke from https://github.com/cgohlke/geospatial-wheels:
Troubleshoot: Rye can't find the right versions¶
Because the pyproject.toml
specifies additional sources, e.g. https://download.pytorch.org/whl/cpu
, it can happen that the a package with an older version is found in these package-indexes.
If such a version is found, uv
(the installer behind Rye
) currently stops searching other sources for the right version and stops with an Version not found
error.
This can look something like this:
No solution found when resolving dependencies:
╰─▶ Because only torchmetrics==1.0.3 is available and you require torchmetrics>=1.4.1, we can conclude that your requirements are unsatisfiable.
To fix this you can set an environment variable to tell uv
to search all package-indicies:
I recommend adding the following to your .zshrc
or .bashrc
:
# Change the behaviour of uv package resolution to enable additional sources without breaking existing version-requirements
export UV_INDEX_STRATEGY="unsafe-best-match"
Please see these issues:
- Rye: Can't specify per-dependency package index / can't specify uv behavior in config file
- UV: Add support for pinning a package to a specific index
Recommended Notebook header¶
The following code snipped can be put in the very first cell of a notebook to already to add logging and initialize earth engine.
import logging
from rich.logging import RichHandler
from rich.traceback import install
from darts.utils.earthengine import init_ee
from darts.utils.logging import setup_logging
setup_logging()
logging.basicConfig(
level=logging.INFO,
format="%(message)s",
datefmt="[%X]",
handlers=[RichHandler(rich_tracebacks=True)],
)
install(show_locals=True) # Change to False if you encounter too large tracebacks
init_ee("ee-project") # Replace with your project