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> lossFunction, SeededRandom? random = null, string? modelFilePath = null)

Parameters

layerListBuilder LayerListBuilder<TInputData, TPrediction>

lossFunction Loss<TPrediction>

random SeededRandom

modelFilePath string

Properties

LossFunction

public Loss<TPrediction> LossFunction { get; }

Property Value

Loss<TPrediction>

Random

protected SeededRandom? Random { get; }

Property Value

SeededRandom

Methods

Backward(TPrediction)

public void Backward(TPrediction lossGrad)

Parameters

lossGrad TPrediction

Clone()

Makes a deep copy of this neural network.

public Model<TInputData, TPrediction> Clone()

Returns

Model<TInputData, TPrediction>

Describe(int)

public 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

HasCheckpoint()

public bool HasCheckpoint()

Returns

bool

LoadParams(string, TInputData?)

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

Parameters

filePath string

initializationSample TInputData

RestoreCheckpoint()

public void RestoreCheckpoint()

SaveCheckpoint()

public void SaveCheckpoint()

SaveParams(string, string?)

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

Parameters

filePath string

comment string

TrainBatch(TInputData, TPrediction)

public float TrainBatch(TInputData xBatch, TPrediction yBatch)

Parameters

xBatch TInputData

yBatch TPrediction

Returns

float

UpdateParams(Optimizer)

public void UpdateParams(Optimizer optimizer)

Parameters

optimizer Optimizer