g-lago8/RNN-JAX

Recurrent neural networks (RNNs) in JAX and Equinox

What's novel

Recurrent neural networks (RNNs) in JAX and Equinox

Code Analysis

17 files read · 4 rounds

A JAX/Equinox library implementing 10+ RNN cell types (Elman, LSTM, GRU, Unitary, Antisymmetric, Lipschitz, Clockwork, Oscillator, IndRNN), state space models (S5, LRU), reservoir computing (ESN), and bidirectional/deep RNN layers, all built on jax.lax.scan and associative_scan for efficient sequenc

Strengths

Genuine algorithmic implementations of non-trivial architectures (Clockwork periodic masking, Unitary RNN with Householder reflectors, S5 with NPLR decomposition, Lipschitz RNN with RK2 discretization) rather than thin wrappers, combined with a clean cell→encoder→layer hierarchy and comprehensive test coverage of every cell type.

Weaknesses

Several copy-paste bugs in core logic (GRU uses wrong weight matrix, coRNN heterogeneous eps initialized with gamma bounds, DeepBidirectionalRNN duplicate assert), plus code hygiene issues like sys.path hacks, global JAX config side effects, and placeholder docstrings that undermine the otherwise solid architecture.

Score Breakdown

Innovation

5 (25%)

Craft

77 (35%)

Traction

6 (15%)

Scope

73 (25%)

Signal breakdown

Innovation

Not Fork+1

Code Novelty+1

Concept Novelty+1

Craft

Ci+5

Tests+8

Polish+1

Releases+0

Has License+5

Code Quality+21

Readme Quality+15

Recent Activity+7

Structure Quality+5

Commit Consistency+5

Has Dependency Mgmt+5

Traction

Forks+0

Stars+6

Hn Points+0

Watchers+0

Early Traction+0

Devto Reactions+0

Community Contribs+0

Scope

Commits+8

Languages+5

Subsystems+5

Bloat Penalty+0

Completeness+7

Contributors+5

Authored Files+12

Readme Code Match+3

Architecture Depth+7

Implementation Depth+8

Evidence

Commits

101

Contributors

Files

Active weeks

TestsCI/CDREADMELicenseContributing

Repository

Language

Python

Stars

Forks

License

MIT