Checklist when pushing:

• When adding features/improvements, has the weightless ZDD codebase also been updated?

This project is in active development.

The main file is zdd.jl in ZDD/jl_zdd/zdd.jl. An example of a script that runs the script is in ZDD/jl_zdd/run_algo.jl.

There is a sister weightless_zdd.jl that is very similar to zdd.jl but without weighting the components. This implementation usually runs faster than zdd.jl, but can only produce all possible ways to partition a graph, whereas the zdd.jl one produces all possible population balanced partitions of a graph.

Testing is done by running julia test_zdd.jl, which tests for correctness for small rook and queen graphs.

The ZDDs can be visualized and plotted by calling construct_zdd(..... , viz=true). Then, you can plot the ZDD by calling draw(zdd, graph, graph_edges). Constructing a ZDD with viz=true is more memory+time intensive because we will be storing things specifically needed for visualization, and visualization gets very complicated for large graphs anyway. It is recommended that this option be turned off when building ZDDs larger than 4x4 grids, and that the user only attempt visualizing graphs less than the size of 4x4 for best performance.

• Identify "dead" nodes (nodes that do not lead to a valid partition) early on in the tree.
• Identify best edge orderings
• Delete the previous layers of the ZDD from memory when we are done with them
• Use parallelization for speedups
• Make the size of the Node smaller
• Merge every i layers?