Module weight_initializer 

Weight initialization strategies for tensor operations.

Provides common initialization methods used in deep learning:

• Xavier/Glorot (uniform and normal)
• He/Kaiming (uniform and normal)
• LeCun (uniform and normal)
• Orthogonal initialization
• Sparse initialization
• Truncated normal
• Constant/zeros/ones

Uses a custom xorshift64 PRNG to avoid external random number dependencies.

Structs

InitStats

Statistics about initialised weights.

TensorShape

Shape information for computing fan-in/fan-out.

WeightInitConfig

Configuration for weight initialization.

WeightInitializer

Weight initializer that generates tensors according to a chosen strategy.

Enums

FanMode

Fan mode for calculating fan-in/fan-out

InitDistribution

Distribution type for initialization

InitStrategy

Weight initialization strategy