OCaT // Oral Comps Assigning Tool

Assigning oral comprehensive exam chairs for Wabash seniors.

Introduction:

The problem is wellknown to Wabash students. The actual prompt is in the file prompt.pdf.

This solver is basically a proof-of-concept - that is why CPLEX/Gurobi is used for a considerable boost in speed; CBC is recommended, since it is open-source, and thus denotes that a free solution is available to create. However, if the school decides to use AMPL everytime it needs, be my guest.

How it works:

init.py takes in the two csv files and parse the data out - this serves as a connector to any data container and form. Then, ampl_gen.py takes in 2 pandas DataFrame and yield a AMPL script. Then, use AMPL/AMPLIDE to solve the file. Finally, run parse_output.py. Each of the python script takes an optional parameter that is the location of the Excel spreadsheet to be supplied by the school - by default it is set to '2016.xlsx.'

This is a completely linear BIP problem, after turning quadratic/cubic constraints into weighted linear ones. The objective function is to minimize sum of squares, but since it is too slow for this project, I decided to minimize sum of absolute values (norm-1 instead of norm-2).

Howto:

Step 0: Manually remove bad data: for example, Allen M. Betts and his unavailability, Jensen A. Kirch his EMU minor (corrected to EDU). Manually add data: list of departments division-wise, and list of faculties' IDs division-wise, in the format of academic_divisions.xlsx.

_{Note: The department names are truncated to the last 3 characters to deal with MLL faculties - unforeseen side effects might occur if departments are named weirdly. Also, faculties in departments that no current senior is in will be ignored.}

Step 0.5: Install AMPL, with CBC solver if you somehow don't like super-fast but propriety solvers.

Step 1: Open Terminal and run git clone https://github.com/ngoctnq/Oral-Comps-Assigning-Tool.git. The file is made so that you can run on any Mac at the school and be good to go.

_{Note: you have to install Xcode (it will happen automatically), which takes ~2min.}

Step 2: Run the following code

cd Oral-Comps-Assigning-Tool
bash ensure_python.sh

_{Note: This script, by default, will not use superuser privilege, and thus only install python modules locally for the current user. If this is were to install globally, sudo it. If you know how to use virtualenv, this project uses Python 2, and all the modules this needs are pandas and xlrd.}

Step 3: Run init.py, ampl_gen.py, then ampl with mock.mod and mock.dat, and finally parse_output.py. Run AMPL with CPLEX (or Gurobi, or CBC), or just use cplex.run or gurobi.run. All the parameters are optional, and the python calls are:

python2 init.py [/path/to/ppl/data='2016.xlsx']
python2 ampl_gen.py [/path/to/ppl/data='2016.xlsx'] [/path/to/div/data='divisions.xlsx']
python2 parse_output.py [/path/to/ppl/data='2016.xlsx']

_{Consult main.sh for a short example - it is a retired script that does not work because I don't know how to use bash.}

Step 4: Voila - It's easy as 1-2-3(-4?)! Do what you want with the new generated schedule.csv.

Notes

• Anything other than CPLEX or Gurobi will be too slow for this.
• CPLEX linearizes quadratic objective, triggering a bug or such that says there are no feasible solutions - necessary flags are needed to prevent this. See this for more info: https://groups.google.com/forum/#!topic/ampl/YO65Sr0rwL4