Skip to content

Examples

Four sketches that cover the three training shapes from the master plan plus ecosystem composition. Each is a code-first walkthrough of how the API in api/ flows together; they are intentionally short (under one screen each) and use the v0.1 Equinox adapter.

Index

File Shape Master plan ref
direct.md Direct supervised — cloud mask §2.1
emulator.md Emulator — chemistry surrogate §2.2
amortized.md Amortized inference — plume SBI §2.3
integration.md Cross-package composition §4

Reading order

If you're new to the framework: direct → emulator → amortized → integration. The complexity grows linearly:

  • direct.md introduces CatalogDataset, Loss, TrainingLoop, Checkpoint, LogToExperiment.
  • emulator.md adds SimulationDataset and the pipekit-cycle.NeuralForward handoff.
  • amortized.md adds ConditionalNormalizingFlow from pyrox, obs_op, and the SBI loss path.
  • integration.md composes pipekit-train with pipekit-cycle, pipekit-experiment, geocatalog, geopatcher, somax, vardax, pyrox-nn — the full vertical slice.