AI project for CS440

note: most runnable Python files use the argument parser library, and so help messages can often be displayed with python [file.py] -h.

Data sets for the algorithms must be generated before the algorithms can be run:

$ cd data; python gen_all.py

This generates twelve sub-directories in data/, whose name is of the form: (c|i|s)-(l|h)-(l|h), where C stands for consistent matrix, I stands for inconsistent, and S stands for semi-consistent. The next two groups are for task- and machine- heterogeneity (Low, or High), respectively.

There are a few supporting files that provide model-abstraction APIs to the algorithms:

• chromosome.py - used by GA, GSA, and Tabu
• gautil.py - used by GA and GSA
• model.py - used by the rest of the algorithms

The rest of the files that are not the algorithm implementations (see below) are utility scripts for running, gathering, and plotting the results.

There are seven algorithms in this project: Opportunistic Load Balancing (OLB), Minimum Completion Time (MCT), Min-Min, Genetic Algorithm (GA), Genetic Simulated Annealing (GSA), Tabu, and A*. They reside in the following files:

• OLB: olb.py
• MCT: mct.py
• Min-Min: minmin.py
• GA: ga.py
• GSA: gsa.py
• Tabu: tabu.py
• A*: astar.py

To run MCT (for example) against the first data file for a consistent-low-task-heterogeneity-low-machine-heterogeneity, run:

python mct.py -f data/c-l-l/1.csv

Help messages for each of the algorithms can be displayed on any of the algorithm files with the -h switch. For example:

python mct.py -h

Use this script to run a set of algorithms on the Computer Science (CS) machines:

Login to a CS machine and run:

python runner.py -a astar -v c-l-l -n 50

This will start fifty runs of the A* algorithm, running the variant 'c-l-l' (loads the datasets that are consistent/low/low [as described in the paper])

Use this script to kill runs started on the CS machines.

python kill.py -a astar -v c-l-l just lists out processes that are still running. To force the script to kill the processes it finds, introduce the switch --kill.

Plot a variant as a PNG file:

python plot.py -v c-l-l generates the file c-l-l.png.