Planned new primitives using current operator set

[yzhwang]

• [ ] Graph Construction: taking k-dimension unstructured data, use similarity metric (such as dot product) to generate weighted/unweighted graphs.
• [ ] Local Structure: K-hop neighbor primitive k in [1,3]
• [ ] General centrality computing primitive: reference paper: http://arxiv.org/pdf/1308.2140v2.pdf
• [ ] Belief propagation

[neoblizz]

Belief propagation maybe label propagation? (it was implemented in v1.0)

Removing the todo tag, these will be good primitives to implement.

[yzhwang]

Awesome! But I believe BP is different than LP, which is a semi-supervised algorithm proposed here: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.13.8280. While BP is a family of algorithms in the context of probabilistic graphical models (which I barely know but I think I wanted to say belief propagation when I filed this issue). But good thing is LP is in Gunrock now! Also, k-hop neighbor can be very useful for some graph embedding algorithms where they will need such kind of information.