MLPWeights

Creates weight tensors for MLP layers with common initialization strategies

Constructors

constructor(builder: TensorBuilder<T, V>)

fun he(inputSize: Int, outputSize: Int, random: Random = Random.Default): TensorInitializer<T, V>

Creates a weight matrix for a linear layer with He initialization (good for ReLU)

fun normal(inputSize: Int, outputSize: Int, mean: Float = 0.0f, std: Float = 0.01f, random: Random = Random.Default): TensorInitializer<T, V>

Creates a weight matrix initialized with normal distribution

fun ones(inputSize: Int, outputSize: Int): TensorInitializer<T, V>

Creates a weight matrix initialized with ones

fun xavier(inputSize: Int, outputSize: Int, random: Random = Random.Default): TensorInitializer<T, V>

Creates a weight matrix for a linear layer with Xavier/Glorot initialization

fun zeros(inputSize: Int, outputSize: Int): TensorInitializer<T, V>

Creates a weight matrix initialized with zeros