node2vec -- reproducing Grover & Leskovec

As I work through understanding "node2vec: Scalable Feature Learning for Networks"1, I'm experimenting with reproducing some of Grover & Leskovec's findings.

Les Mis' Case Study

G&L use the Les Mis' data (see my json version) to demonstrate both homophily and structural equivalence. The dataset has 77 Character nodes and 254 APPEARED_WITH relationships connecting them.

Tooling

See my python code in kmeans.py. To use it:

1. Load the sample data using APOC's apoc.import.json proc. It should just load.
2. Create a Python virtualenv and install deps: pip install -r requirements.txt
3. Run the code: $ python kmeans.py -A bolt://localhost:7687 -U neo4j -P password -d 16 -p 1 -q 0.6

Les Mis Homophily

G&L claim they set d = 16 and p = 1, q = 0.5.

I'm struggling to reproduce this...

...more details coming.