---

name: rbm description: "Routes legacy binary RBM examples, CD versus PCD decisions, binarized MNIST assumptions, and NumPy compatibility troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: Unlicense

RBM

Use this sub-skill for the repository's binary Restricted Boltzmann Machine examples: Contrastive Divergence and Persistent Contrastive Divergence.

Route here when

• The user asks for a binary RBM, Bernoulli visible/hidden variables, CD, or PCD.
• The user asks how the RBM examples binarize MNIST or generate preview images.
• The user needs to diagnose np.float, TensorFlow MNIST-loader, or working-directory failures in the RBM examples.

Route elsewhere when

• GAN variants: ../gan/SKILL.md
• VAE variants: ../vae/SKILL.md
• Helmholtz Machine wake-sleep: ../helmholtz-machine/SKILL.md
• Shared catalog, provenance, and compatibility: ../../SKILL.md

Fast decisions

• Use CD when the user wants the canonical binary RBM baseline or does not mention persistent chains.
• Use PCD when the user asks for persistent sampling, persistent contrastive divergence, or model samples carried between updates.
• Treat both scripts as legacy NumPy training loops that still import the TensorFlow 1.x MNIST helper.
• Expect binarized MNIST (> 0.5) and preview image grids under a working-directory-local out/ folder.

Read next

• references/workflows.md for CD-vs-PCD behavior and data assumptions.
• references/troubleshooting.md for RBM-specific failures.
• ../../references/model-catalog.md for the repo-wide source artifact catalog.
• ../../references/compatibility.md for the shared legacy stack notes.
• ../../scripts/check_legacy_stack.py before claiming an unchanged RBM example can execute.