https://github.com/andrew-johnson-4/perplexity

A notational semantic for documenting neural networks through diagrams
https://github.com/andrew-johnson-4/perplexity

category-theory diagramming documentation domain-specific-language lambda-calculus modelling neural-network visual-programming

Last synced: 7 months ago
JSON representation

A notational semantic for documenting neural networks through diagrams

• Host: GitHub
• URL: https://github.com/andrew-johnson-4/perplexity
• Owner: andrew-johnson-4
• License: mit
• Created: 2022-04-25T21:15:01.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2023-08-05T02:58:58.000Z (about 2 years ago)
• Last Synced: 2025-01-11T01:57:18.103Z (9 months ago)
• Topics: category-theory, diagramming, documentation, domain-specific-language, lambda-calculus, modelling, neural-network, visual-programming
• Language: HTML
• Homepage: https://andrew-johnson-4.github.io/lsts-tutorial/
• Size: 210 KB
• Stars: 14
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Funding: .github/FUNDING.yml
  • License: LICENSE
  • Code of conduct: CODE_OF_CONDUCT.md

Awesome Lists containing this project

README

          
# Perplexity 😵

The Perplexity 😵 language is a notational semantic for documenting neural networks through diagrams.

![MNIST Keras Python Perplexity](/img/mnist_keras_python_perplexity.png)

# Scope

This visual language was created to help document neural networks.

The language may not be suitable to formally describe all characteristics of all possible neural network configurations.

# Format

The Perplexity 😵 language consists of two languages: one [textual](/syntax.md), the other [visual](/syntax.md).

The visual language consists of two-dimensional images that can be created in Paint or other visual editors.

The textual language consists of [Typed Lambda Calculus expressions](https://andrew-johnson-4.github.io/lsts-tutorial/).

Any file, textual or visual, defines a substitution rule according to its filename.

For example, a diagram "helpful.png" may be referred to as "helpful" in other textual language documentation.

# Why?

UML is not information dense enough and often leaves out important details. Mathematical Notation is too information dense and often repeats itself. Perplexity 😵 is created specifically to model Neural Networks and cuts a lot of corners by specializing itself for this use-case.

# Checklist for Documenting a Model

1. How many languages/algebras are you going to use in documentation?

> This determines the rank of your model. It is recommended to use different colored lines and circles

> when swapping between languages.

> For example, if you just want to document a TensorFlow model, then you only need Python and your model's rank will be 1.

2. What prelude do you want to use?

> A prelude will include some terms that you then don't need to define again yourself. Each prelude defines its own ranks, kinds, types, values etc. Please consider using a prelude.

# List of Preludes

Preludes are optional-ish.

- [Categorical Prelude](/categorical_prelude.md)

- [TensorFlow Prelude](/tensorflow_prelude.md)