DIVAnd.py

DIVAnd performs an n-dimensional variational analysis of arbitrarily located observations (python interface).
This package is a python interface to DIVAnd, written in Julia.

The primary use case of DIVAnd.py is to build web service and therefore the only OS supported by DIVAnd.py is Linux.

Installation

Besides Python and NumPy you need to:

• Install Julia
• Python's julia package by following closely the instructions at https://github.com/JuliaPy/pyjulia

Test the integration of Julia and Python by running the following code in python's shell (python or python-jl in Ubuntu/Debian):

from julia.Base import banner; banner()

If this does not work, have a look again at https://github.com/JuliaPy/pyjulia and file an issue at pyjulia if you belief you found an issue in pyjulia.

• Clone this package

git clone https://github.com/gher-ulg/DIVAnd.py

Testing

You are advised to try one of these examples in the directory examples:

• DIVAnd_argo.py
• DIVAnd_small.py

The environment variable PYTHONPATH should contain the directory with the file DIVAnd.py.

You can set this variable for the current shell and then run the example:

export PYTHONPATH="$HOME/src/DIVAnd.py/DIVAnd:$PYTHONPATH"
python DIVAnd_argo.py

or just for a single call of the python script (the following command should be on one line):

PYTHONPATH="$HOME/src/DIVAnd.py/DIVAnd:$PYTHONPATH" python DIVAnd_argo.py

Note that you should adapt the path in the previous example ($HOME/src/DIVAnd.py/DIVAnd) to match the installation location of DIVAnd.py.

Troubleshooting

Segmentation fault at initialization of pyjulia

On Ubuntu, the default python interpreter (/usr/bin/python) produces a segmentation fault because it is linked statically to libpython2.7.so. As a work-around, use the python interpreter from e.g. Conda after installing the julia package Conda.jl. The python interpreter is located at $HOME/.julia/v0.6/Conda/deps/usr/bin/python2.7.

C++ runtime library error

If you see the following error:

OSError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.20' not found (required by /home/abarth/opt/julia-3c9d75391c/bin/../lib/libjulia.so.0.5)

Define the variable LD_PRELOAD before starting python or (ipython)

export LD_PRELOAD="$HOME/opt/julia-3c9d75391c/lib/julia/libstdc++.so.6"

You should adapt the path in the previous command to match the installation location of Julia.

Python crash

On some systems, python crashes on j = julia.Julia() (JuliaPy/pyjulia#66). Rebuilding PyCall after installing juliapy seems to solve this issue. In Julia, run the following command:

Pkg.build("PyCall")

Exception 'UVError'

could not spawn `/home/abarth/.local/lib/python2.7/site-packages/julia-0.1.1-py2.7.egg/julia/../fake-julia/julia -Cx86-64 -J/home/abarth/opt/julia-0.6.0/lib/julia/sys.so --compile=yes --depwarn=yes -O0 --output-ji /home/abarth/Test/julia-v0.6-dev/lib/pyjulia2-v0.6/PyCall.ji --output-incremental=yes --startup-file=no --history-file=no --color=no --eval 'while !eof(STDIN)
    eval(Main, deserialize(STDIN))
end
'`: no such file or directory (ENOENT)Traceback (most recent call last):
  File "test_simple_module.py", line 2, in <module>
    j = julia.Julia()
  File "/home/abarth/.local/lib/python2.7/site-packages/julia-0.1.1-py2.7.egg/julia/core.py", line 361, in __init__
    self._call(u"using PyCall")
  File "/home/abarth/.local/lib/python2.7/site-packages/julia-0.1.1-py2.7.egg/julia/core.py", line 403, in _call
    self.check_exception(src)
  File "/home/abarth/.local/lib/python2.7/site-packages/julia-0.1.1-py2.7.egg/julia/core.py", line 423, in check_exception
    .format(exception_type, src))
julia.core.JuliaError: Exception 'UVError' occurred while calling julia code:
using PyCall

Pkg.build("PyCall")
using PyCall

Julia exception: MethodError

If you optain the following error by running one of the examples, consider to upgrade PyCall (to a version later than 1.14 or using the git version)

WARNING: redefining constant JULIA_HOME
Traceback (most recent call last):
  File "DIVAnd_small.py", line 25, in <module>
    va = DIVAnd(mask, (pm, pn), (xi, yi), (x, y), v, (lenx, leny), epsilon2)
  File "/home/abarth/src/DIVAnd.py/DIVAnd/DIVAnd.py", line 14, in DIVAnd
    x, f, corlen, epsilon2)
RuntimeError: Julia exception: MethodError(DIVAnd.DIVAndrunfi, (Bool[true true true true; true true true true; true true true true], ([2.0 2.0 2.0 2.0; 2.0 2.0 2.0 2.0; 2.0 2.0 2.0 2.0], [3.0 3.0 3.0 3.0; 3.0 3.0 3.0 3.0; 3.0 3.0 3.0 3.0]), ([0.0 0.0 0.0 0.0; 0.5 0.5 0.5 0.5; 1.0 1.0 1.0 1.0], [0.0 0.333333 0.666667 1.0; 0.0 0.333333 0.666667 1.0; 0.0 0.333333 0.666667 1.0]), ([1.0e-10 0.5 1.0; 1.0e-10 0.5 1.0; 1.0e-10 0.5 1.0], [1.0e-10 1.0e-10 1.0e-10; 0.5 0.5 0.5; 1.0 1.0 1.0]), [6.0e-10 0.14112 -0.279415; -5.93995e-10 -0.139708 0.276619; 5.76102e-10 0.135499 -0.268286], (0.15, 0.15), 0.05), 0x000000000000553e)

Installation of python modules

You may want to use modules such as netCDF4 or matplotlib. Two possibilies are offered, starting in the $HOME/.julia/v0.6/Conda/deps/usr/bin/python2.7 directory:

1. Using the Conda installation:

conda install numpy

2. Using pip command:

pip install numpy

The installed packages can be listed using pip list or conda list.

Note that the modules available through Conda are those in one of the following repositories:

• https://repo.continuum.io/pkgs/free/linux-64
• https://repo.continuum.io/pkgs/free/noarch
• https://repo.continuum.io/pkgs/r/linux-64
• https://repo.continuum.io/pkgs/r/noarch
• https://repo.continuum.io/pkgs/pro/linux-64
• https://repo.continuum.io/pkgs/pro/noarch while pip provides access to more resources through https://pypi.python.org/pypi.

Order of the dimensions

The default order of dimension in python is reversed relative to Julia (Row- vs column-major order). For example a 3D analysis mask, po, pn, pm, zi, yi, xi are 3D arrays with the dimensions corresponding to depth, latitude and longitude:

va2 = DIVAnd.DIVAnd(mask,(po,pn,pm),(zi,yi,xi),(z,y,x),f,lenx,epsilon2)

This is also the order of dimensions produced by e.g. python's netCDF4 for CF compliant files.

Python 3

See python3.