=======================
PuLP How-to
=======================

(Draft)

Date: 2-15-2023

This page includes the following items and is based on `PuLP documentations <https://coin-or.github.io/pulp/>`_. 

- How to install PuLP on `csusb-hpc <https://csusb-hpc.nrp-nautilus.io>`_. 
- How to run PuLP. 
- How to add GUROBI. 
  
.. note:: 

    ``GUROBI`` is a licensed program. They offer education licenses for free, but one needs to create an [account](https://www.gurobi.com/) and get the license on a campus network.

Installation
-------------

If it is not already installed, use the following command to install PuLP via [PyPI](https://pypi.org/project/PuLP/) and run a check.

.. code-block::

    pip install https://pypi.org/project/PuLP/
    pulptest 

Instead, one can install it in JupyterLab as follows.

.. code-block::

    %pip install pulp
    import pulp as pl
    pl.pulpTestAll()


Multi-threads
--------------

PuLP utilized multiple cores via the `solvers' thread option <https://coin-or.github.io/pulp/technical/solvers.html#pulp.apis.PULP_CBC_CMD.name>`_. 

.. code-block::

    cbc_solver = PULP_CBC_CMD(threads=4)

Change the number of threads to the number of cores requested. 

.. note:: 

    In the test, 32 cores seemed to be the most reasonable as the number of trees becomes smaller. Also, it did not use much RAM. So, 16 GB should be a safe choice.  

.. note:: 

    When this page was written, according to the web search, there was no practical way to use GPUs for PuLP. See this article https://stackoverflow.com/questions/73096443/using-colab-gpu-for-pulp-python. 

GUROBI solver
--------------

In our testing, GUROBI solver caused an error with ``pulptest``. As a default, only package paths are added into the ``stack-pulp-test``. To install GUROBI, simply run

.. code-block::

    conda install gurobi=10.0.1

Or in JupyterLab, 

.. code-block::
    
    %conda install gurobi=10.0.1
    # %ensures that it is installed in the right conda environment.

To use GUROBI on the server, one needs ``WSL`` license. It supports up to 8 threads. (It is free for educational use.) Once you have the `gurobi.lic` file, please it into your home folder (likely ``~\jobyan\``).

.. code-block::

    gurobi_solver = GUROBI_CMD(threads=8)