https://github.com/colltoaction/nx_yaml
A higher-order syntax
https://github.com/colltoaction/nx_yaml
category-theory geometric-deep-learning higher-order hypergraph intermediate-representation language network-analysis networkx programming-language yaml
Last synced: 23 days ago
JSON representation
A higher-order syntax
- Host: GitHub
- URL: https://github.com/colltoaction/nx_yaml
- Owner: colltoaction
- Created: 2023-07-24T14:32:09.000Z (almost 2 years ago)
- Default Branch: main
- Last Pushed: 2025-02-16T22:50:43.000Z (3 months ago)
- Last Synced: 2025-04-11T22:03:12.660Z (23 days ago)
- Topics: category-theory, geometric-deep-learning, higher-order, hypergraph, intermediate-representation, language, network-analysis, networkx, programming-language, yaml
- Language: Jupyter Notebook
- Homepage:
- Size: 1.59 MB
- Stars: 3
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# nx_yaml
`nx_yaml` is a general purpose Hypergraph Intermediate Representation for [YAML](https://yaml.org/spec/1.2.2) using [NetworkX](https://github.com/networkx/networkx). It provides native semantics for higher-order applications such as Hypergraph Analysis, Quantum Computing, Geometric Deep Learning and Monoidal Computing.
There are two notebook tutorials: [a basic one](tutorial.ipynb) and an advanced one using [XGI](xgi.ipynb) for higher-order interpretation.
## Design
Graph Theory is well established in compiler infrastructure, popularized by the use of Abstract Syntax Trees, and recently projects such as [MLIR](https://mlir.llvm.org/docs/Rationale/RationaleGenericDAGRewriter/) have further explored Direct Acyclic Graphs. This project formalizes a [Hypergraph](https://en.wikipedia.org/wiki/Hypergraph) IR with native semantics for higher-order applications such as:
* Hypergraph Analysis: https://github.com/pnnl/HyperNetX
* Quantum Computing: https://github.com/zxcalc/pyzx
* Geometric Deep Learning: https://github.com/pyg-team/pytorch_geometric
* Monoidal Computing: https://github.com/discopy/discopy
YAML has the advantage of being a widely known syntax that is implemented in all programming languages. NetworkX enables seamless collaboration between graph libraries. We can implement all sorts of document transformations, leveraging a large ecosystem of software and well-known algorithms for complex networks.
## Implementation
This project implements the PyYAML `yaml.compose` and `yaml.serialize` APIs using the NetworkX graph library.
`compose` takes text input and converts it to a graph, while `serialize` does the inverse operation.
Understanding this data structure is key to create portable and correct tooling for YAML, moving freely between text-based and code representations.
Check out the [notebook tutorial](tutorial.ipynb).
We use `NxSafeLoader` and `NxSafeDumper` with the PyYAML `yaml.compose` and `yaml.serialize` APIs.
### Milestone 1
1. Copy and paste from PyYAML
1. Replace uses of `nodes` with NetworkX
1. Publish a drop-in pip replacement
### Milestone 2
1. Compare performance
1. Compare alignment with spec
1. Share findings
## Development environment
This is work in progress.
`pypi` is updated on demand.
The development environment is self-contained using the `pipenv` tool.
### Testing
* Just `pytest`
* Store graphs in `resources/tests`
[Representation Graph]: https://yaml.org/spec/1.2.2/#321-representation-graph
[pyyaml.nodes]: https://github.com/yaml/pyyaml/blob/main/lib/yaml/nodes.py