Run a shape optimization
==============================

To run a shape optimization, you have to write a configurations file with the optimization (number of iterations, costs to use...)


The launcher script
---------------------
Stellacode contain a python script called **launcher** in the root of Stellacode. It takes a config name as argument and run a simulation.

::

    python launcher.py path_to_config_file

The script will save the following

1. The config file given (under the name *config.ini*)
2. A log file (*log.txt*)
3. The intermediate results at the rate given in the config file (a new file each time)
4. The final result (*result*)

.. warning:: the repository *tmp/* has to be created before launching the script


Plot and compares the results
------------------------------
You can use the notebook *show_simu.ipynb* in the root folder of Stellacode to extract all the simulation folder generated by the execution of *launcher.py*. It provides plot examples and a latex array comparison.

Customize on your cluster
-------------------------------
In order to ensure a reasonnable Parallelization, Stellacode uses `Dask <https://dask.org/>`_ (only for the most costly part, it would be great to use it for everything).
We refer to the `Dask documentation <https://docs.dask.org/en/latest/setup.html>`_ for further details.

.. warning:: 
    Several layers of parallelization takes place in Stellacode:
       - **Dask** will parallelize the *Task graph* generated (which increase for smaller chunks)
       - **Numpy** parallelize all linear algebra computations.

    As a consequence, it is important to limit numpy cores usage (for a single thread). For example if numpy uses MKL, you can use the following line **before** importing numpy.
    ::
    
        os.environ["MKL_NUM_THREADS"] = "4" # export MKL_NUM_THREADS=6