Dataset and Sample Transforms (Torchvision-Style)

This module provides a small transform library inspired by torchvision.transforms:

• composable pure transforms (Compose, Lambda)
• dataset mapping helpers
• common tensor/sample normalization utilities

PyTorch Mapping

• torchvision.transforms: https://pytorch.org/vision/stable/transforms.html
• torch.utils.data.Dataset map-style datasets: https://pytorch.org/docs/stable/data.html

TorchLean difference: transforms are pure functions over typed tensors/samples, so shape mistakes are caught by the typechecker rather than at runtime.

def NN.API.Data.Transforms.Compose {a : Type} (ts : List (a → a)) : a → a

Torchvision-style transform composition.

Applies transforms left-to-right:

Compose [f, g, h] x = h (g (f x)).

def NN.API.Data.Transforms.Lambda {a : Type} (f : a → a) : a → a

Torchvision-style "Lambda" transform wrapper.

def NN.API.Data.Transforms.compose {a b c : Type} (g : b → c) (f : a → b) : a → c

Compose two pure transforms.

def NN.API.Data.Transforms.onDataset {a b : Type} (f : a → b) (ds : Dataset a) : Dataset b

Apply a pure transform to every element of a dataset.

def NN.API.Data.Transforms.mapTensor {α : Type} {s : Spec.Shape} (f : α → α) (x : Spec.Tensor α s) : Spec.Tensor α s

Map a tensor elementwise.

def NN.API.Data.Transforms.normalizeTensor {α : Type} [Sub α] [Div α] {s : Spec.Shape} (mean std : α) (x : Spec.Tensor α s) : Spec.Tensor α s

Normalize any tensor elementwise: (x - mean) / std.

def NN.API.Data.Transforms.normalizeTensorF {α : Type} [Runtime.Scalar α] [Sub α] [Div α] {s : Spec.Shape} (mean std : Float) (x : Spec.Tensor α s) : Spec.Tensor α s

Float-literal normalization helper for runtime scalar backends.

def NN.API.Data.Transforms.mapLabels {a : Type} (f : ℕ → ℕ) (xs : List (a × ℕ)) : List (a × ℕ)

Transform labels in (sample, label) datasets.

def NN.API.Data.Transforms.mapSamples {a b : Type} (f : a → b) (xs : List (a × ℕ)) : List (b × ℕ)

Transform samples in (sample, label) datasets.

def NN.API.Data.Transforms.onSamples {a b : Type} (f : a → b) (ds : Dataset (a × ℕ)) : Dataset (b × ℕ)

Apply a sample transform to a labeled dataset.

def NN.API.Data.Transforms.onLabels {a : Type} (f : ℕ → ℕ) (ds : Dataset (a × ℕ)) : Dataset (a × ℕ)

Apply a label transform to a labeled dataset.

def NN.API.Data.Transforms.onSupervisedInput {α : Type} {σ τ : Spec.Shape} (f : Spec.Tensor α σ → Spec.Tensor α σ) : sample.Supervised α σ τ → sample.Supervised α σ τ

Transform the input component of a supervised TorchLean sample TList α [σ, τ].

def NN.API.Data.Transforms.onSupervisedTarget {α : Type} {σ τ : Spec.Shape} (f : Spec.Tensor α τ → Spec.Tensor α τ) : sample.Supervised α σ τ → sample.Supervised α σ τ

Transform the target component of a supervised TorchLean sample TList α [σ, τ].

def NN.API.Data.Transforms.onSupervisedDatasetInput {α : Type} {σ τ : Spec.Shape} (f : Spec.Tensor α σ → Spec.Tensor α σ) (ds : Dataset (sample.Supervised α σ τ)) : Dataset (sample.Supervised α σ τ)

Apply an input transform over a supervised TorchLean dataset.