Trainer Datasets

Trainer.Dataset input target is the public supervised-data interface. It delays both sample construction and conversion into the selected arithmetic representation until training begins. Consequently the same dataset can be used with every executable arithmetic implementation supported by the trainer.

The materialized value is a finite Data.SampleStream: samples are requested by index and need not be stored eagerly. Dataset constructors live in NN.API.Data.Training.

structure TorchLean.Trainer.Dataset (input target : Shape) : Type 1

Supervised data with statically known input and target shapes.

• materialize {α : Type} [Storage α] [Context α] [Runtime.FromFloat α] : IO (Data.SampleStream (Sample.Supervised α input target))

  Materialize the finite sample stream after the trainer selects its runtime arithmetic.