model

model

ML model utilities: helpers for building, inspecting, and training models.

Two kinds of submodules live here:

• Framework-agnostic helpers – no import of a specific ML framework, work on plain arrays/indices/callables. Flat siblings at this level, e.g.:

  q.model.selection – leakage-aware CV splitters (walk-forward, purged K-fold, embargo), named after scikit-learn’s model_selection

• Framework-specific adapters – thin wrappers around one library’s actual model objects/training API. Each framework gets its own submodule, which may grow into its own subpackage as it picks up multiple concerns (e.g. training loops, checkpointing, dataloaders alongside model summaries):

  q.model.torch – PyTorch helpers (model summaries via torchinfo today; training/checkpointing/dataloaders as they land, per the roadmap)

Don’t add a generic dispatching helper (e.g. a framework-agnostic train() that inspects the model type and picks a backend) – signatures don’t unify across frameworks, and it’s the same registry/ABC ceremony already rejected elsewhere in this codebase (see qrt.data.sources). Add a new framework submodule only once it’s actually used, not speculatively.

To add your own framework family or model-lifecycle helper, create qrt/model/<name>.py (or a qrt/model/<name>/ subpackage), then register it in the imports and __all__ below. It becomes available as q.model.<name>.