all-features

Stage 1: all features on a subset of observations

Here's a description of this stage quoting from the project report:

The all-features stage assesses feature performance and does not require computing features for all negatives. Here we selected a random subset of 3,020 (4 × 755) negatives. Little error was introduced by this optimization, since the predominant limitation to performance assessment was the small number of positives (755) rather than negatives.

Datasets

Here are some of the notable datasets:

• data/metapaths.json contains information on and queries for each metapath used to generate features.
• data/dwpc.tsv.bz2 contains is a tidy (long) TSV of the output from each DWPC query performed including path count (PC), degree-weighted path count (DWPC), and query runtime.

For documentation requests, open a GitHub Issue. Documentation pull requests also welcome.