Class Model<TInputData, TPrediction>

Namespace

NeuralNetworks.Models

Assembly

NeuralNetworks.dll

Represents an abstract neural network model that processes input data and produces predictions. Provides core functionality for forward and backward passes, training, parameter updates, and checkpointing.

public abstract class Model<TInputData, TPrediction> where TInputData : notnull where TPrediction : notnull

Type Parameters

TInputData

The type of input data provided to the model. Must not be null.

TPrediction

The type of prediction output produced by the model. Must not be null.

Inheritance

object

Model<TInputData, TPrediction>

Derived

BaseModel<TInputData, TPrediction>

GenericModel<TInputData, TPrediction>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

This class serves as a base for implementing neural network models with customizable layers and loss functions. It supports training workflows, including batch training and parameter optimization, and provides mechanisms for saving and restoring model checkpoints. Derived classes must implement the method for constructing the layer list. Thread safety is not guaranteed; concurrent access should be managed externally.

Constructors

Model(LayerListBuilder<TInputData, TPrediction>?, Loss<TPrediction>?, SeededRandom?, string?)

protected Model(LayerListBuilder<TInputData, TPrediction>? layerListBuilder, Loss<TPrediction>? defaultLossFunction = null, SeededRandom? random = null, string? modelFilePath = null)

Parameters

layerListBuilder LayerListBuilder<TInputData, TPrediction>

defaultLossFunction Loss<TPrediction>

random SeededRandom

modelFilePath string

Properties

Random

protected SeededRandom? Random { get; }

Property Value

SeededRandom

Methods

Backward(TPrediction)

public void Backward(TPrediction lossGrad)

Parameters

lossGrad TPrediction

CalculateLoss(TPrediction, TPrediction, Loss<TPrediction>?)

public float CalculateLoss(TPrediction predictions, TPrediction targets, Loss<TPrediction>? lossFunction = null)

Parameters

predictions TPrediction

targets TPrediction

lossFunction Loss<TPrediction>

Returns

float

Describe(int)

public virtual List<string> Describe(int indentation = 0)

Parameters

indentation int

Returns

List<string>

Forward(TInputData, bool)

public TPrediction Forward(TInputData input, bool inference)

Parameters

input TInputData

inference bool

Returns

TPrediction

GetParamCount()

public int GetParamCount()

Returns

int

InferFromLayer<TLayerInputData, TLayerOutputData>(Layer<TLayerInputData, TLayerOutputData>, TLayerInputData)

protected TPrediction InferFromLayer<TLayerInputData, TLayerOutputData>(Layer<TLayerInputData, TLayerOutputData> fromLayer, TLayerInputData input) where TLayerInputData : notnull where TLayerOutputData : notnull

Parameters

fromLayer Layer<TLayerInputData, TLayerOutputData>

input TLayerInputData

Returns

TPrediction

Type Parameters

TLayerInputData

TLayerOutputData

LoadParams(string, TInputData?)

public void LoadParams(string filePath, TInputData? initializationSample = default)

Parameters

filePath string

initializationSample TInputData

SaveParams(string, string?)

public void SaveParams(string filePath, string? comment = null)

Parameters

filePath string

comment string

TrainBatch(TInputData, TPrediction, Loss<TPrediction>?)

public float TrainBatch(TInputData xBatch, TPrediction yBatch, Loss<TPrediction>? lossFunction = null)

Parameters

xBatch TInputData

yBatch TPrediction

lossFunction Loss<TPrediction>

Returns

float

UpdateParams(Optimizer)

public void UpdateParams(Optimizer optimizer)

Parameters

optimizer Optimizer