Class FromFileInitializationStrategy<T>

Namespace

AiDotNet.Initialization

Assembly

AiDotNet.dll

Initialization strategy that loads weights from an external file.

public class FromFileInitializationStrategy<T> : InitializationStrategyBase<T>, IInitializationStrategy<T>

Type Parameters

T

The numeric type used for calculations.

Inheritance

object

InitializationStrategyBase<T>

FromFileInitializationStrategy<T>

Implements

IInitializationStrategy<T>

Inherited Members

InitializationStrategyBase<T>.NumOps

InitializationStrategyBase<T>.Random

InitializationStrategyBase<T>.XavierUniformInitialize(Tensor<T>, int, int)

InitializationStrategyBase<T>.XavierNormalInitialize(Tensor<T>, int, int)

InitializationStrategyBase<T>.HeUniformInitialize(Tensor<T>, int)

InitializationStrategyBase<T>.HeNormalInitialize(Tensor<T>, int)

InitializationStrategyBase<T>.ZeroInitializeBiases(Tensor<T>)

InitializationStrategyBase<T>.SampleGaussian(double, double)

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

This strategy loads pre-trained weights from a file, enabling transfer learning and model checkpointing. Weights are loaded during the first initialization call and cached for subsequent layers.

For Beginners: Transfer learning is like giving your network a head start by using weights that were already trained on a similar task. Instead of starting from random values, you start with values that already know useful patterns.

Supported formats: - JSON: Human-readable format with weight arrays - Binary: Compact binary format for faster loading

Constructors

FromFileInitializationStrategy(string, WeightFileFormat)

Initializes a new instance of the FromFileInitializationStrategy<T> class.

public FromFileInitializationStrategy(string filePath, WeightFileFormat format = WeightFileFormat.Auto)

Parameters

filePath string

The path to the weights file.

format WeightFileFormat

The format of the weights file. Default is Auto-detect.

Properties

IsLazy

Gets a value indicating whether this strategy defers initialization until first use.

public override bool IsLazy { get; }

Property Value

bool

true if initialization is deferred until first Forward() call; false if initialization happens immediately.

LoadFromExternal

Gets a value indicating whether weights should be loaded from an external source.

public override bool LoadFromExternal { get; }

Property Value

bool

true if weights should be loaded from file or other external source; false if weights should be randomly initialized.

Methods

ClearCache()

Clears the cached weights, forcing a reload on next initialization.

public void ClearCache()

Remarks

This method is thread-safe. All cached data is cleared atomically.

InitializeBiases(Tensor<T>)

Initializes the biases tensor with appropriate values.

public override void InitializeBiases(Tensor<T> biases)

Parameters

biases Tensor<T>

The biases tensor to initialize.

InitializeWeights(Tensor<T>, int, int)

Initializes the weights tensor with appropriate values.

public override void InitializeWeights(Tensor<T> weights, int inputSize, int outputSize)

Parameters

weights Tensor<T>

The weights tensor to initialize.

inputSize int

The number of input features.

outputSize int

The number of output features.

Reset()

Resets the layer indices for a fresh initialization pass.

public void Reset()

Remarks

Call this method if you need to re-initialize a network with the same weights. This method is thread-safe.