inst/python_wrapper/README.md

In climextRemes: Tools for Analyzing Climate Extremes

Python wrapping of R climextRemes package

Overview on how the Python package works

1. The climextremes package is just the python_wrapper/climextremes/__init__.py file. This gets run when package is loaded.
2. Next is an attempt to install the climextRemes R package, which should pull in the R package dependencies. Note that we try to get the version of climextRemes (R) that matches version of climextremes (python), but there is a fall-back.
3. assuming climextRemes installs, Python calls out to R to figure out where climextRemes_wrapper is and then imports everything from climextRemes_wrapper
4. climextRemes_wrapper goes through and creates Python fxns to wrap the R functions.

Installation

See user-facing installation instructions in top-level README file of climextremes repository.

Documentation

Probably best to do this in a Conda environment.

To make documentation:

1. Make sure climextremes (Python package, presumably using setup.py method) is installed first.
2. Put new version number into python_wrapper/docs/build_sphinx_docs.sh (ideally I'd figure out how to leverage version.py).
3. One needs the Ubuntu latexmk package which is available for 18.04 but not for 16.04.
4. If Roxygen documentation in R package has been updated, make sure you've locally installed the up-to-date R package so don't get old help info from CRAN version of R package.

pip install sphinx
pip install sphinx-autodoc-annotation
cd python_wrapper/docs
python make_docs.py

The make_docs.py script should create a documentation template and call build_sphinx_docs.sh. This script should then create the makefile and call make html and make latexpdf.

Note there are a couple warnings about formatting in fit_gev and fit_pot that I haven't been able to figure out.

/accounts/gen/vis/paciorek/.local/lib/python3.7/site-packages/climextremes/__init__.py:docstring of climextremes.fit_gev:84: WARNING: Block quote ends without a blank line; unexpected unindent.
/accounts/gen/vis/paciorek/.local/lib/python3.7/site-packages/climextremes/__init__.py:docstring of climextremes.fit_pot:100: WARNING: Block quote ends without a blank line; unexpected unindent.

Building for installation via pip and conda.

See the README.md files in the pip and conda directories.

Other notes

To update version number one should be able to change this in only python_wrapper/version.py.

The Python package should be OS- and Python3-version- independent. See more details here: https://www.anaconda.com/condas-new-noarch-packages/

To build the Python package see inst/pip/README.md or inst/conda/README.md.

When debugging, if needed one can directly modify contents of .py files in the installed R climextRemes package (e.g., ~/R/x86.../3.5/climextRemes/*)

numpy,pandas,rpy2,tzlocal apparently required in conda meta.yaml build: stanza. Not clear why but if try to only have python and setuptools in build: you get this:

Processing dependencies for climextremes==0.2.1rc8
Searching for tzlocal
Traceback (most recent call last):
  File "setup.py", line 75, in <module>
    install_requires=['numpy', 'pandas', 'rpy2', 'tzlocal'],
  File "/accounts/gen/vis/paciorek/.conda/envs/climextremes_build_min/conda-bld/climextremes_1556215226148/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold/lib/python3.7/site-packages/setuptools/__init__.py", line 145, in setup
    return distutils.core.setup(**attrs)

Don't need compilers in dependencies as conda r package pulls them in, specific for an OS.

noarch version seems to be Python version independent (at least for Python 3).

conda R packages always installed from source.

r-rcpparmadillo avoids an issue with installing RcppArmadillo at run-time. RcppArmadillo is pulled in by climextRemes dependencies.

rpy2 pulls in Rcpp, which is needed for a climextRemes dependency.

Could try to set up build.sh and bld.bat to install R packages (including climextRemes) at install time, but might be harder to debug user issues.

When import climextremes, the code in __init__.py tries to install climextRemes. It can fail (e.g. I've seen download from cran.r-project.us fail) and in that case in line 91 of init.py it can be unable to find climextRemes_wrapper because __climextRemes_python_path__ not set so it can't find it in the python dir in the installed R package.

Note that the code in climextRemes_wrapper.py could be in the Python package but as it is now the same Python wrapper can point to different R versions and we bundle the Python code with the R version.

The only thing permanent in inst/python_wrapper is climextremes/__init__.py (the actual Python package code), docs, setup.py and metadata files (COPYRIGHTS.txt, LICENSE.txt, version.py, README.md) . All the rest has been created during experiments with building the package.

Building package uses setup.py, which is the build script for setuptools.

conda-forge approach should be able to install from pypi tarball via staged recipe. conda-forge recipe should be able to avoid usage of build.sh/bld.bat

In some cases if one doesn't force rpy2 >=2.9.4, Conda installation will pull in rpy2=2.8.5, r=3.3.1, and there will be problems. Not clear why.

As of August 2020, default Conda channel has rpy2 2.9.4 but pulls in pandas >= 1.0.0, which causes a run-time error because of an apparent rpy2-pandas incompatibility. conda-forge has new rpy2 but it doesn't seem possible to force use of a channel in the conda meta.yaml, nor does asking for rpy2 >=3.0.0 work when building the Conda package.

As of July 2022, I've modified the Conda deploy.sh to use -c conda-forge which allows me to build based on newer rpy2 and pandas, so removing the version specifications in meta.yaml.

climextRemes documentation built on May 29, 2024, 4:39 a.m.