machine-learning

Machine Learning is a concise concept and best-practices guide for developing machine learning code with JAX using functional programming patterns and high-performance computing techniques. It solves the problem of applying JAX idiomatically: it frames the assistant as an expert in JAX and functional programming and lays out the principles that keep JAX code fast and correct, such as immutability, pure functions, and structuring code for JIT compilation.

The single-file skill covers JAX fundamentals including jax.numpy array operations, automatic differentiation with jax.grad, JIT compilation with jax.jit, and vectorization with jax.vmap. It gives control-flow guidance (jax.lax.scan, cond, and fori_loop instead of Python control flow inside jitted functions), the functional random-number API with proper key splitting, and performance and memory-management advice like choosing appropriate dtypes, batching, checkpointing, and profiling. It also outlines common patterns (pytrees, custom vjp/jvp, multi-device sharding) and model-development practices using Flax or Haiku with functional training loops.

This is a thin, guidance-only reference rather than an executable tool or library, so it is best understood as a curated set of conventions. It targets machine learning engineers and researchers adopting JAX who want a compact checklist of idiomatic patterns and pitfalls. Use cases include establishing coding conventions for a JAX project, onboarding to functional ML patterns, and reminders on correct random-key handling and JIT-friendly control flow.