cossio/RestrictedBoltzmannMachines.jl

Train and sample Restricted Boltzmann machines in Julia

What's novel

Train and sample Restricted Boltzmann machines in Julia

Code Analysis

15 files read · 4 rounds

A Julia package for training and sampling Restricted Boltzmann Machines with multiple unit types (binary, spin, Potts, Gaussian, ReLU variants), supporting PCD training, Gibbs/Metropolis sampling, annealed importance sampling for partition function estimation, and centered/standardized RBM variants

Strengths

Exceptionally clean architecture with a macro-based layer system that reduces boilerplate while maintaining type stability, mathematically rigorous implementations verified against analytical solutions and automatic differentiation, and genuinely novel layer types (dReLU as two-sided truncated Gaussian mixture) with proper gauge handling for Potts layers.

Weaknesses

The codebase is Julia-specific and the @declare_layer macro, while elegant, creates a learning curve for contributors; some advanced features like the standardized RBM PCD loop have complex interdependencies between offset/scale updates and gauge constraints that could benefit from more documentation of the invariants.

Score Breakdown

Innovation

5 (25%)

Craft

83 (35%)

Traction

37 (15%)

Scope

89 (25%)

Signal breakdown

Innovation

Not Fork+1

Code Novelty+1

Concept Novelty+1

Craft

Ci+5

Tests+8

Polish+2

Releases+5

Has License+5

Code Quality+27

Readme Quality+15

Recent Activity+7

Structure Quality+4

Commit Consistency+5

Has Dependency Mgmt+0

Traction

Forks+12

Stars+20

Hn Points+0

Watchers+3

Early Traction+0

Devto Reactions+0

Community Contribs+2

Scope

Commits+8

Languages+5

Subsystems+13

Bloat Penalty+0

Completeness+6

Contributors+7

Authored Files+15

Readme Code Match+3

Architecture Depth+7

Implementation Depth+8

Evidence

Commits

Contributors

Files

101

Active weeks

TestsCI/CDREADMELicenseContributing

Repository

Language

Jupyter Notebook

Stars

Forks

License

MIT