Transform

A type-safe transformation operation that converts input of type I to output of type O.

Transforms are the building blocks of data preprocessing pipelines in SKaiNET. They can be composed together using the then infix function to create complex preprocessing chains while maintaining full type safety.

Design Principles

1. Type Safety: Generic types I and O ensure that transforms can only be composed when their types are compatible.

2. Composability: Transforms can be chained using then to create pipelines:

    val pipeline = loadImage then resize(224, 224) then toTensor then normalize

   Content copied to clipboard

3. Shape Awareness: Each transform can compute its output shape from the input shape, enabling shape inference without executing the full pipeline.

4. Immutability: Transforms should be stateless and immutable. Any configuration should be provided at construction time.

Example Usage

// Define a simple scaling transform
class Scale(private val factor: Float) : Transform<Float, Float> {
    override fun apply(input: Float): Float = input * factor
    override fun getOutputShape(inputShape: Shape): Shape = inputShape
}

// Compose transforms
val pipeline = Scale(2.0f) then Scale(0.5f)
val result = pipeline.apply(10.0f)  // Returns 10.0f

Parameters

Inheritors