Enum InitializationStrategyType

Namespace

AiDotNet.Initialization

Assembly

AiDotNet.dll

Specifies the type of initialization strategy to use for layer weights.

public enum InitializationStrategyType

Fields

Eager = 0

Eager initialization - initialize weights immediately during layer construction. This is the traditional approach that ensures all weights are ready before training.

FromFile = 3

Load weights from an external file. Use for transfer learning or resuming training from a checkpoint. Requires setting the WeightsFilePath property.

Lazy = 1

Lazy initialization - defer weight initialization until the first forward pass. This makes network construction faster, especially for large models or when just inspecting network architecture.

Zero = 2

Zero initialization - set all weights to zero. Use only for testing or specific architectures. Not recommended for training.

Remarks

Use this enum with TrainingMemoryConfig to configure the default initialization strategy for all layers.