Skip to content

Latest commit

 

History

History
31 lines (25 loc) · 1.65 KB

data_structures.md

File metadata and controls

31 lines (25 loc) · 1.65 KB
Raw

Data structures

The main data structures of this project are Data and NAG.

Data

A Data object stores a single-level graph. It inherits from torch_geometric's Data and has a similar behavior (see the official documentation for more on this).

Important specificities of our Data object are:

  • Data.super_index stores the parent's index for each node in Data
  • Data.sub holds a Cluster object indicating the children of each node in Data
  • Data.to_trimmed() works like torch_geometric's Data.coalesce() with the additional constraint that (i,j) and (j,i) edges are considered duplicates
  • Data.save() and Data.load() allow optimized, memory-friendly I/O operations
  • Data.select() indexes the nodes à la numpy

NAG

NAG (Nested Acyclic Graph) stores the hierarchical partition in the form of a list of Data objects.

Important specificities of our Data object are:

  • NAG[i] returns a Data object holding the partition level ì
  • NAG.get_super_index() returns the index mapping nodes from any level i to j with i<j
  • NAG.get_sampling() produces indices for sampling the superpoints with certain constraints
  • NAG.save() and NAG.load() allow optimized, memory-friedly I/O operations
  • NAG.select() indexes the nodes of a specified partition level à la numpy and updates the rest of the NAG structure accordingly

🚀  Getting familiar with the data structures

See the notebooks/demo_nag.ipyng notebook to play with and visualize a provided NAG, without needing a whole dataset.