Class AutoMLModelBase<T, TInput, TOutput>

Namespace

AiDotNet.AutoML

Assembly

AiDotNet.dll

Base class for AutoML models that automatically search for optimal model configurations

public abstract class AutoMLModelBase<T, TInput, TOutput> : IAutoMLModel<T, TInput, TOutput>, IFullModel<T, TInput, TOutput>, IModel<TInput, TOutput, ModelMetadata<T>>, IModelSerializer, ICheckpointableModel, IParameterizable<T, TInput, TOutput>, IFeatureAware, IFeatureImportance<T>, ICloneable<IFullModel<T, TInput, TOutput>>, IGradientComputable<T, TInput, TOutput>, IJitCompilable<T>

Type Parameters

T

The numeric type used for calculations

TInput

The input data type

TOutput

The output data type

Inheritance

object

AutoMLModelBase<T, TInput, TOutput>

Implements

IAutoMLModel<T, TInput, TOutput>

IFullModel<T, TInput, TOutput>

IModel<TInput, TOutput, ModelMetadata<T>>

IModelSerializer

ICheckpointableModel

IParameterizable<T, TInput, TOutput>

IFeatureAware

IFeatureImportance<T>

ICloneable<IFullModel<T, TInput, TOutput>>

IGradientComputable<T, TInput, TOutput>

IJitCompilable<T>

Derived

DiffusionAutoML<T>

NasAutoMLModelBase<T>

SupervisedAutoMLModelBase<T, TInput, TOutput>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Extension Methods

DistributedExtensions.AsDistributedForHighBandwidth<T, TInput, TOutput>(IFullModel<T, TInput, TOutput>, ICommunicationBackend<T>)

DistributedExtensions.AsDistributedForLowBandwidth<T, TInput, TOutput>(IFullModel<T, TInput, TOutput>, ICommunicationBackend<T>)

DistributedExtensions.AsDistributed<T, TInput, TOutput>(IFullModel<T, TInput, TOutput>, ICommunicationBackend<T>)

DistributedExtensions.AsDistributed<T, TInput, TOutput>(IFullModel<T, TInput, TOutput>, IShardingConfiguration<T>)

Fields

_candidateModels

protected readonly List<ModelType> _candidateModels

Field Value

List<ModelType>

_constraints

protected readonly List<SearchConstraint> _constraints

Field Value

List<SearchConstraint>

_earlyStoppingMinDelta

protected double _earlyStoppingMinDelta

Field Value

double

_earlyStoppingPatience

protected int? _earlyStoppingPatience

Field Value

int?

_lock

protected readonly object _lock

Field Value

object

_maximize

protected bool _maximize

Field Value

bool

_modelEvaluator

protected IModelEvaluator<T, TInput, TOutput>? _modelEvaluator

Field Value

IModelEvaluator<T, TInput, TOutput>

_optimizationMetric

protected MetricType _optimizationMetric

Field Value

MetricType

_optimizationMetricExplicitlySet

protected bool _optimizationMetricExplicitlySet

Field Value

bool

_searchSpace

protected readonly Dictionary<string, ParameterRange> _searchSpace

Field Value

Dictionary<string, ParameterRange>

_trialHistory

protected readonly List<TrialResult> _trialHistory

Field Value

List<TrialResult>

_trialsSinceImprovement

protected int _trialsSinceImprovement

Field Value

int

Properties

BestModel

Gets the best model found so far

public IFullModel<T, TInput, TOutput>? BestModel { get; protected set; }

Property Value

IFullModel<T, TInput, TOutput>

BestScore

Gets the best score achieved

public double BestScore { get; protected set; }

Property Value

double

DefaultLossFunction

Gets the default loss function for gradient computation.

public virtual ILossFunction<T> DefaultLossFunction { get; }

Property Value

ILossFunction<T>

Remarks

AutoML delegates to the best model found during search. If no best model exists yet, returns Mean Squared Error as a sensible default.

FeatureNames

Gets the feature names

public virtual string[] FeatureNames { get; set; }

Property Value

string[]

MaximizeOptimizationMetric

Gets a value indicating whether higher metric values are better.

public bool MaximizeOptimizationMetric { get; }

Property Value

bool

Remarks

For Beginners: For some metrics higher is better (Accuracy). For error metrics lower is better (RMSE).

OptimizationMetric

Gets the optimization metric used to rank trials.

public MetricType OptimizationMetric { get; }

Property Value

MetricType

Remarks

This is safe to expose and does not reveal proprietary hyperparameter values or model internals.

For Beginners: This is the score AutoML tries to make better (like Accuracy or RMSE).

ParameterCount

Gets the number of parameters

public virtual int ParameterCount { get; }

Property Value

int

Status

Gets the current optimization status

public AutoMLStatus Status { get; protected set; }

Property Value

AutoMLStatus

SupportsJitCompilation

Gets whether this model currently supports JIT compilation.

public virtual bool SupportsJitCompilation { get; }

Property Value

bool

True if the best model found supports JIT compilation, false otherwise.

Remarks

AutoML models delegate JIT compilation support to their best model. If no best model has been found yet, JIT compilation is not supported.

For Beginners: AutoML models can only be JIT compiled if the best model they found supports it.

Since AutoML searches across multiple model types, JIT support depends on:

• Whether a best model has been selected
• Whether that specific model supports JIT compilation

Before running SearchAsync, this will return false. After finding a best model, it delegates to that model's JIT support.

TimeLimit

Gets or sets the time limit for the AutoML search

public TimeSpan TimeLimit { get; set; }

Property Value

TimeSpan

TrialLimit

Gets or sets the maximum number of trials to run

public int TrialLimit { get; set; }

Property Value

int

Type

Gets the model type

public virtual ModelType Type { get; }

Property Value

ModelType

Methods

ApplyGradients(Vector<T>, T)

Applies gradients by delegating to the best model.

public virtual void ApplyGradients(Vector<T> gradients, T learningRate)

Parameters

gradients Vector<T>

learningRate T

Clone()

Creates a memberwise clone of the AutoML model using MemberwiseClone(). This performs a shallow copy where reference types are shared between the original and clone.

public virtual IFullModel<T, TInput, TOutput> Clone()

Returns

IFullModel<T, TInput, TOutput>

A memberwise clone of the current AutoML model

Remarks

For a deep copy with independent collections and state, use DeepCopy() instead.

ComputeGradients(TInput, TOutput, ILossFunction<T>?)

Computes gradients by delegating to the best model.

public virtual Vector<T> ComputeGradients(TInput input, TOutput target, ILossFunction<T>? lossFunction = null)

Parameters

input TInput

target TOutput

lossFunction ILossFunction<T>

Returns

Vector<T>

ConfigureSearchSpace(Dictionary<string, ParameterRange>)

Configures the search space for hyperparameter optimization

public virtual void ConfigureSearchSpace(Dictionary<string, ParameterRange> searchSpace)

Parameters

searchSpace Dictionary<string, ParameterRange>

Dictionary defining parameter ranges to search

CreateInstanceForCopy()

Factory method for creating a new instance for deep copy. Derived classes must implement this to return a new instance of themselves. This ensures each copy has its own collections and lock object.

protected abstract AutoMLModelBase<T, TInput, TOutput> CreateInstanceForCopy()

Returns

AutoMLModelBase<T, TInput, TOutput>

A fresh instance of the derived class with default parameters

Remarks

When implementing this method, derived classes should create a fresh instance with default parameters, and should not attempt to preserve runtime or initialization state from the original instance. The deep copy logic will transfer relevant state (trial history, search space, etc.) after construction.

CreateModelAsync(ModelType, Dictionary<string, object>)

Creates a model instance for the given type and parameters

protected abstract Task<IFullModel<T, TInput, TOutput>> CreateModelAsync(ModelType modelType, Dictionary<string, object> parameters)

Parameters

modelType ModelType

parameters Dictionary<string, object>

Returns

Task<IFullModel<T, TInput, TOutput>>

DeepCopy()

Creates a deep copy of the AutoML model

public virtual IFullModel<T, TInput, TOutput> DeepCopy()

Returns

IFullModel<T, TInput, TOutput>

Deserialize(byte[])

Deserializes the model from bytes

public virtual void Deserialize(byte[] data)

Parameters

data byte[]

EnableEarlyStopping(int, double)

Enables early stopping

public virtual void EnableEarlyStopping(int patience, double minDelta = 0.001)

Parameters

patience int

minDelta double

EnableNAS(bool)

Enables Neural Architecture Search (NAS) for automatic network design

public virtual void EnableNAS(bool enabled = true)

Parameters

enabled bool

Whether to enable NAS

EvaluateModelAsync(IFullModel<T, TInput, TOutput>, TInput, TOutput)

Evaluates a model on the validation set

protected virtual Task<double> EvaluateModelAsync(IFullModel<T, TInput, TOutput> model, TInput validationInputs, TOutput validationTargets)

Parameters

model IFullModel<T, TInput, TOutput>

validationInputs TInput

validationTargets TOutput

Returns

Task<double>

ExportComputationGraph(List<ComputationNode<T>>)

Exports the computation graph for JIT compilation by delegating to the best model.

public virtual ComputationNode<T> ExportComputationGraph(List<ComputationNode<T>> inputNodes)

Parameters

inputNodes List<ComputationNode<T>>

List to populate with input computation nodes.

Returns

ComputationNode<T>

The output computation node representing the model's prediction.

Remarks

AutoML models delegate graph export to their best model found during search. The graph structure and complexity depends on the specific best model type.

For Beginners: This creates a computation graph from the best model found.

AutoML itself doesn't have a fixed computation structure since it tries multiple model types. Instead, it delegates to the best model it found:

• If the best model is a neural network, you get a neural network graph
• If it's a regression model, you get a regression graph
• And so on

This only works after SearchAsync has found and selected a best model.

Exceptions

InvalidOperationException

Thrown when no best model exists (SearchAsync not called yet).

NotSupportedException

Thrown when the best model does not support JIT compilation.

ExtractMetricFromEvaluation(ModelEvaluationData<T, TInput, TOutput>)

Extracts the appropriate metric value from the evaluation results

protected virtual double ExtractMetricFromEvaluation(ModelEvaluationData<T, TInput, TOutput> evaluationData)

Parameters

evaluationData ModelEvaluationData<T, TInput, TOutput>

Returns

double

GetActiveFeatureIndices()

Gets the indices of active features

public virtual IEnumerable<int> GetActiveFeatureIndices()

Returns

IEnumerable<int>

GetDefaultSearchSpace(ModelType)

Gets the default search space for a model type

protected abstract Dictionary<string, ParameterRange> GetDefaultSearchSpace(ModelType modelType)

Parameters

modelType ModelType

Returns

Dictionary<string, ParameterRange>

GetFeatureImportance()

Gets the feature importance scores

public virtual Dictionary<string, T> GetFeatureImportance()

Returns

Dictionary<string, T>

GetFeatureImportanceAsync()

Gets feature importance from the best model

public virtual Task<Dictionary<int, double>> GetFeatureImportanceAsync()

Returns

Task<Dictionary<int, double>>

GetModelMetadata()

Gets model metadata

public virtual ModelMetadata<T> GetModelMetadata()

Returns

ModelMetadata<T>

GetParameters()

Gets the model parameters

public virtual Vector<T> GetParameters()

Returns

Vector<T>

GetResults()

Gets the results of all trials performed during search

public virtual List<TrialResult> GetResults()

Returns

List<TrialResult>

List of trial results with scores and parameters

GetTrialHistory()

Gets the history of all trials

public virtual List<TrialResult> GetTrialHistory()

Returns

List<TrialResult>

IsFeatureUsed(int)

Checks if a feature is used

public virtual bool IsFeatureUsed(int featureIndex)

Parameters

featureIndex int

Returns

bool

LoadModel(string)

Loads the model from a file

public virtual void LoadModel(string filePath)

Parameters

filePath string

LoadState(Stream)

Loads the AutoML model's state from a stream.

public virtual void LoadState(Stream stream)

Parameters

stream Stream

The stream to read the model state from.

Remarks

This method deserializes a best model that was previously saved with SaveState. It uses the existing Deserialize method after reading data from the stream.

For Beginners: This is like loading a saved snapshot of your best AutoML model.

When you call LoadState:

• The best model is read from the stream
• All parameters and configuration are restored

After loading, the model can:

• Make predictions using the restored best model
• Be further optimized if needed
• Be deployed to production

This is essential for:

• Loading the best model after AutoML search
• Deploying optimized models to production
• Knowledge distillation workflows

Exceptions

ArgumentNullException

Thrown when stream is null.

IOException

Thrown when there's an error reading from the stream.

InvalidOperationException

Thrown when the stream contains invalid or incompatible data, or when BestModel is not initialized.

Predict(double[][])

Makes predictions using the best model (legacy method)

public virtual double[] Predict(double[][] inputs)

Parameters

inputs double[][]

Returns

double[]

Predict(TInput)

Makes predictions using the best model found

public virtual TOutput Predict(TInput input)

Parameters

input TInput

Returns

TOutput

ReportTrialFailureAsync(Dictionary<string, object>, Exception, TimeSpan)

Reports a failed trial result without terminating the full AutoML run.

protected virtual Task ReportTrialFailureAsync(Dictionary<string, object> parameters, Exception error, TimeSpan duration)

Parameters

parameters Dictionary<string, object>

The parameters used in the trial.

error Exception

The exception that caused the trial to fail.

duration TimeSpan

The duration of the failed trial.

Returns

Task

ReportTrialResultAsync(Dictionary<string, object>, double, TimeSpan)

Reports the result of a trial

public virtual Task ReportTrialResultAsync(Dictionary<string, object> parameters, double score, TimeSpan duration)

Parameters

parameters Dictionary<string, object>

score double

duration TimeSpan

Returns

Task

Run(TInput, TOutput, TInput, TOutput)

Runs the AutoML optimization process (alternative name for Search)

public virtual void Run(TInput inputs, TOutput targets, TInput validationInputs, TOutput validationTargets)

Parameters

inputs TInput

Training inputs

targets TOutput

Training targets

validationInputs TInput

Validation inputs

validationTargets TOutput

Validation targets

SaveModel(string)

Saves the model to a file

public virtual void SaveModel(string filePath)

Parameters

filePath string

SaveState(Stream)

Saves the AutoML model's current state to a stream.

public virtual void SaveState(Stream stream)

Parameters

stream Stream

The stream to write the model state to.

Remarks

This method serializes the best model found during the AutoML search. It uses the existing Serialize method and writes the data to the provided stream.

For Beginners: This is like creating a snapshot of your best AutoML model.

When you call SaveState:

• The best model found during search is written to the stream
• All model parameters and configuration are preserved

This is particularly useful for:

• Saving the best model after AutoML search
• Checkpointing during long-running searches
• Knowledge distillation from AutoML-optimized models
• Deploying optimized models to production

You can later use LoadState to restore the model.

Exceptions

ArgumentNullException

Thrown when stream is null.

InvalidOperationException

Thrown when no best model exists.

IOException

Thrown when there's an error writing to the stream.

Search(TInput, TOutput, TInput, TOutput)

Performs the AutoML search process (synchronous version)

public virtual void Search(TInput inputs, TOutput targets, TInput validationInputs, TOutput validationTargets)

Parameters

inputs TInput

Training inputs

targets TOutput

Training targets

validationInputs TInput

Validation inputs

validationTargets TOutput

Validation targets

SearchAsync(TInput, TOutput, TInput, TOutput, TimeSpan, CancellationToken)

Searches for the best model configuration

public abstract Task<IFullModel<T, TInput, TOutput>> SearchAsync(TInput inputs, TOutput targets, TInput validationInputs, TOutput validationTargets, TimeSpan timeLimit, CancellationToken cancellationToken = default)

Parameters

inputs TInput

targets TOutput

validationInputs TInput

validationTargets TOutput

timeLimit TimeSpan

cancellationToken CancellationToken

Returns

Task<IFullModel<T, TInput, TOutput>>

SearchBestModel(TInput, TOutput, TInput, TOutput)

Searches for the best model configuration (synchronous version)

public virtual IFullModel<T, TInput, TOutput> SearchBestModel(TInput inputs, TOutput targets, TInput validationInputs, TOutput validationTargets)

Parameters

inputs TInput

Training inputs

targets TOutput

Training targets

validationInputs TInput

Validation inputs

validationTargets TOutput

Validation targets

Returns

IFullModel<T, TInput, TOutput>

Best model found

Serialize()

Serializes the model to bytes

public virtual byte[] Serialize()

Returns

byte[]

SetActiveFeatureIndices(IEnumerable<int>)

Sets the active feature indices

public virtual void SetActiveFeatureIndices(IEnumerable<int> featureIndices)

Parameters

featureIndices IEnumerable<int>

SetCandidateModels(List<ModelType>)

Sets the models to consider in the search

public virtual void SetCandidateModels(List<ModelType> modelTypes)

Parameters

modelTypes List<ModelType>

SetConstraints(List<SearchConstraint>)

Sets constraints for the search

public virtual void SetConstraints(List<SearchConstraint> constraints)

Parameters

constraints List<SearchConstraint>

SetModelEvaluator(IModelEvaluator<T, TInput, TOutput>)

Sets the model evaluator to use for evaluating candidate models

public virtual void SetModelEvaluator(IModelEvaluator<T, TInput, TOutput> evaluator)

Parameters

evaluator IModelEvaluator<T, TInput, TOutput>

SetModelsToTry(List<ModelType>)

Sets which model types should be considered during the search

public virtual void SetModelsToTry(List<ModelType> modelTypes)

Parameters

modelTypes List<ModelType>

List of model types to evaluate

SetOptimizationMetric(MetricType, bool)

Sets the optimization metric

public virtual void SetOptimizationMetric(MetricType metric, bool maximize = true)

Parameters

metric MetricType

maximize bool

SetParameters(Vector<T>)

Sets the model parameters

public virtual void SetParameters(Vector<T> parameters)

Parameters

parameters Vector<T>

SetSearchSpace(Dictionary<string, ParameterRange>)

Sets the search space for hyperparameters

public virtual void SetSearchSpace(Dictionary<string, ParameterRange> searchSpace)

Parameters

searchSpace Dictionary<string, ParameterRange>

SetTimeLimit(TimeSpan)

Sets the time limit for the AutoML search process

public virtual void SetTimeLimit(TimeSpan timeLimit)

Parameters

timeLimit TimeSpan

Maximum time to spend searching for optimal models

SetTrialLimit(int)

Sets the maximum number of trials to execute during search

public virtual void SetTrialLimit(int maxTrials)

Parameters

maxTrials int

Maximum number of model configurations to try

ShouldStop()

Checks if early stopping criteria is met

protected bool ShouldStop()

Returns

bool

SuggestNextTrialAsync()

Suggests the next hyperparameters to try

public abstract Task<Dictionary<string, object>> SuggestNextTrialAsync()

Returns

Task<Dictionary<string, object>>

Train(double[][], double[])

Trains the model (legacy method - use SearchAsync instead)

public virtual void Train(double[][] inputs, double[] outputs)

Parameters

inputs double[][]

outputs double[]

Train(TInput, TOutput)

Trains the AutoML model by searching for the best configuration

public virtual void Train(TInput input, TOutput expectedOutput)

Parameters

input TInput

expectedOutput TOutput

ValidateConstraints(Dictionary<string, object>, IFullModel<T, TInput, TOutput>?)

Validates constraints for a given configuration

protected bool ValidateConstraints(Dictionary<string, object> parameters, IFullModel<T, TInput, TOutput>? model = null)

Parameters

parameters Dictionary<string, object>

model IFullModel<T, TInput, TOutput>

Returns

bool

WithParameters(Vector<T>)

Creates a new instance with the given parameters

public virtual IFullModel<T, TInput, TOutput> WithParameters(Vector<T> parameters)

Parameters

parameters Vector<T>

Returns

IFullModel<T, TInput, TOutput>