Stochastic optimization

Tools for fitting models on batched data via stochastic_fit.

Data loaders

Helper classes loading data for stochastic optimization.

DataLoader	Protocol for data loaders that stream batches.
ArrayDataLoader	DataLoader for in-memory arrays.
LazyArrayDataLoader	DataLoader for lazy/out-of-core arrays (e.g. dask, zarr, HDF5).

Callbacks

Callback interfaces for stochastic optimization and post-fit run summaries.

Callback	Base class for training callbacks.
CallbackList	Composite callback that dispatches each hook to all registered callbacks.
SolverConvergenceCallback	Delegate convergence checking to the solver's built-in criterion.
TestLossLogger	Log the loss evaluated on a fixed test set at the requested events.
TrainingContext	Mutable context object passed to callbacks during training.
StochasticFitSummary	Post-fit summary of a stochastic training run.