In Pull Request #27, we changed the config file to include a directed label under each reconstruction algorithm. This is either true or false, and is used to determine whether to write summary statistics and graphspace JSON files for a directed graph or an undirected graph. This may not be the ideal way to handle algorithm-level attributes. A couple thoughts:

1. We may want to have a way to designate the input network files as directed or undirected as well. For example, the shortest-paths reconstruction algorithm could probably take either a directed or undirected graph. The algorithm's prepare-inputs would handle that according to the dataset.
2. There are cases where the output should be considered directed (e.g, visualizing a graph) and it should be considered undirected (e.g., we often compute precision-recall in an undirected manner, so if an algorithm returns (u,v) or (v,u) the two edges are counted the same way). This motivates some directed tag later under analysis -> precrec when we get that implemented.
3. There may be other algorithm-level attributes that we would want to consider. Output may be ranked or unranked (e.g., analysis should consider the set of edges, even for pathlinker, as a single subnetwork rather than a ranked list of edges). I do not know whether they should be algorithm-level or analysis-level labels, or some combination of both.