Class IncrementalPredictor

Incremental regression block that learns to map input feature vectors to multi-dimensional output predictions in real time.

JSON configuration

Use these block entries inside a pipeline configuration. Input names must match other blocks in that pipeline; device addresses and file paths must match your setup.

JSON configuration:

{
  "Predictor": {
    "Type": "IncrementalPredictor",
    "DesiredRate": 200,
    "Inputs": [ "LowpassFilter", "Trigger" ],
    "Params": [ "RidgeRFF", 8, 12, 0.5, 1.0, 300 ],
    "Path": "C:/recordings"
  }
}

How Type, Inputs and Params work · Block catalogue and parameter reference

Inheritance
IncrementalPredictor
Namespace: MOSAIC.Models.MachineLearning
Assembly: MOSAIC.dll
Syntax
public sealed class IncrementalPredictor : BaseBlock, INotifyPropertyChanged, INotifyPropertyChanging, ISubscriber, IPublisher, IBaseBlockProps, IJsonExportable, IDisposable, IAsyncDisposable
Remarks

Overview: The IncrementalPredictor sits between a feature-extraction pipeline (e.g. EMG → Filter → Features) and a control algorithm. It learns an online mapping from input features to output activations using incremental ridge regression (optionally with Random Fourier Features for non-linear kernels). Training is driven by a Trigger block that supplies labeled target vectors.

Inputs: Expects exactly two upstream connections:

  • Feature source — any block publishing Vector<double> of length InputDim, or a Matrix<double> of shape [timesteps × InputDim] where each row is processed as an individual feature vector.
  • Trigger — publishes a target Vector<double> of length OutputDim to start training, or null to stop.

Output: After every feature input, publishes a prediction Vector<double> of length OutputDim. For matrix input, only the last row's prediction is published (all rows are trained on if training is active).

Constructors

IncrementalPredictor(string, double, int, int)

Declaration
public IncrementalPredictor(string name, double desiredRate, int inputDim, int outputDim)
Parameters
Type Name Description
string name
double desiredRate
int inputDim
int outputDim

IncrementalPredictor(string, double, int, int, RegressorType, RegressorConfig)

Declaration
public IncrementalPredictor(string name, double desiredRate, int inputDim, int outputDim, RegressorType modelType, RegressorConfig config)
Parameters
Type Name Description
string name
double desiredRate
int inputDim
int outputDim
RegressorType modelType
RegressorConfig config

IncrementalPredictor(string, double, int, int, RegressorType, RegressorConfig, string?)

Declaration
public IncrementalPredictor(string name, double desiredRate, int inputDim, int outputDim, RegressorType modelType, RegressorConfig config, string? savePath)
Parameters
Type Name Description
string name
double desiredRate
int inputDim
int outputDim
RegressorType modelType
RegressorConfig config
string savePath

Fields

_confidence

Raw confidence value from the regressor.

Declaration
[ObservableProperty]
private double _confidence
Field Value
Type Description
double

_currentTarget

Current training target vector, or null when not training.

Declaration
private Vector<double>? _currentTarget
Field Value
Type Description
Vector<double>

_currentTargetName

Name of the current training target.

Declaration
[ObservableProperty]
private string _currentTargetName
Field Value
Type Description
string

_isTraining

Whether the block is currently training.

Declaration
[ObservableProperty]
[NotifyPropertyChangedFor("TrainingStatusText")]
private bool _isTraining
Field Value
Type Description
bool

_lastPrediction

Most recent prediction vector.

Declaration
private Vector<double>? _lastPrediction
Field Value
Type Description
Vector<double>

_lastVizFeedTicks

Timestamp of last viz feed for stall detection.

Declaration
private long _lastVizFeedTicks
Field Value
Type Description
long

_model

The active regression model.

Declaration
private IRegressor _model
Field Value
Type Description
IRegressor

_sampleCount

Total training samples processed.

Declaration
[ObservableProperty]
private int _sampleCount
Field Value
Type Description
int

_savePath

Directory for model auto-save.

Declaration
private readonly string? _savePath
Field Value
Type Description
string

_viz

Visualization bundle set by the ViewModel.

Declaration
private BlockVisualization? _viz
Field Value
Type Description
BlockVisualization

Properties

Confidence

Raw confidence value from the regressor.

Declaration
[ExcludeFromCodeCoverage]
public double Confidence { get; set; }
Property Value
Type Description
double

Config

Hyperparameters for the active regressor.

Declaration
public RegressorConfig Config { get; private set; }
Property Value
Type Description
RegressorConfig

CurrentTargetName

Name of the current training target.

Declaration
[ExcludeFromCodeCoverage]
public string CurrentTargetName { get; set; }
Property Value
Type Description
string

DumpFilePrefix

Recordings from this block are named <block>_predictions.csv.

Declaration
protected override string DumpFilePrefix { get; }
Property Value
Type Description
string
Overrides
BaseBlock.DumpFilePrefix

InputDim

Expected length of incoming feature vectors.

Declaration
public int InputDim { get; set; }
Property Value
Type Description
int

IsTraining

Whether the block is currently training.

Declaration
[ExcludeFromCodeCoverage]
public bool IsTraining { get; set; }
Property Value
Type Description
bool

JsonTypeName

Block type identifier for JSON serialization. Override for a custom name.

Declaration
protected override string JsonTypeName { get; }
Property Value
Type Description
string
Overrides
BaseBlock.JsonTypeName

LastPrediction

Most recent prediction vector.

Declaration
public Vector<double>? LastPrediction { get; }
Property Value
Type Description
Vector<double>

MaxInputs

Needs exactly two inputs (e.g. a data stream + a Trigger/TriggerBuffer).

Declaration
public override int MaxInputs { get; }
Property Value
Type Description
int
Overrides
BaseBlock.MaxInputs

MinInputs

Needs exactly two inputs (e.g. a data stream + a Trigger/TriggerBuffer).

Declaration
public override int MinInputs { get; }
Property Value
Type Description
int
Overrides
BaseBlock.MinInputs

ModelType

Active regression algorithm type.

Declaration
public RegressorType ModelType { get; private set; }
Property Value
Type Description
RegressorType

OutputDim

Number of prediction output dimensions.

Declaration
public int OutputDim { get; set; }
Property Value
Type Description
int

OutputLabels

Per-output-column labels from the Trigger's action dictionary.

Declaration
public List<string>? OutputLabels { get; private set; }
Property Value
Type Description
List<string>

SampleCount

Total training samples processed.

Declaration
[ExcludeFromCodeCoverage]
public int SampleCount { get; set; }
Property Value
Type Description
int

TrainingStatusText

Human-readable training status for UI.

Declaration
public string TrainingStatusText { get; }
Property Value
Type Description
string

Visualization

This block's visualization, or null when it has none.

Declaration
protected override BlockVisualization? Visualization { get; }
Property Value
Type Description
BlockVisualization
Overrides
BaseBlock.Visualization
Remarks

Blocks own their scope as a public Viz property rather than inheriting one, so the base class has no way to reach it. Overriding this is what lets the rate below be wired up automatically. With correct publication metadata, the bundle derives the time axis for one vector per publication or fixed-size contiguous sample packets from the feed shape.

When a plot needs a rate that publication frequency and feed shape cannot express (for example, variable-sized packets or overlapping windows), declare it with Viz.UpdateSignalRate. An explicit declaration wins over the derived rate. A block with several bundles must configure each one; this hook exposes only one bundle.

Viz

Live plot for this block. Assigned by the ViewModel, which owns the scope.

Declaration
public BlockVisualization? Viz { get; set; }
Property Value
Type Description
BlockVisualization
Remarks

Setting this re-pushes the publish rate: the rate is normally pushed when it changes, which for these blocks happens during construction — before the ViewModel has handed over the scope — so without this the scope would never learn its time base.

Methods

AutoSave()

Declaration
public void AutoSave()

ChangeModel(RegressorType, RegressorConfig?)

Declaration
public void ChangeModel(RegressorType type, RegressorConfig? config = null)
Parameters
Type Name Description
RegressorType type
RegressorConfig config

ConfigureInput(IServiceProvider, JsonModel)

Declaration
public static IncrementalPredictor ConfigureInput(IServiceProvider sp, JsonModel m)
Parameters
Type Name Description
IServiceProvider sp
JsonModel m
Returns
Type Description
IncrementalPredictor

Dispose()

Performs application-defined tasks associated with freeing, releasing, or resetting unmanaged resources.

Declaration
public override void Dispose()
Overrides
BaseBlock.Dispose()

GetDefaultSavePath()

Declaration
private string? GetDefaultSavePath()
Returns
Type Description
string

GetJsonParams()

Block-specific parameters for JSON. Override to provide custom parameters.

Declaration
protected override IReadOnlyList<object>? GetJsonParams()
Returns
Type Description
IReadOnlyList<object>

The values for Params, in the order this block's ConfigureInput reads them, or null when the block takes no parameters.

Overrides
BaseBlock.GetJsonParams()

GetJsonPath()

Overrides what this block writes to Path.

Declaration
protected override string? GetJsonPath()
Returns
Type Description
string

The path to export, or null to fall back to the block's own recording folder.

Overrides
BaseBlock.GetJsonPath()
Remarks

Only for blocks that read Path as something other than a recording folder — a predictor's model directory, say. Such a block also overrides PathIsDumpFolder; without this override it would export null and saving the pipeline would silently drop the directory.

HandleFeatureInput(Vector<double>)

Processes a feature vector: trains (if active), predicts, publishes, and visualizes.

Declaration
private void HandleFeatureInput(Vector<double> x)
Parameters
Type Name Description
Vector<double> x

HandleFeatureInputSilent(Vector<double>)

Processes a feature vector for training and prediction only — no Publish, no Viz feed. Used for intermediate rows of a matrix input so only the last row triggers output.

Declaration
private void HandleFeatureInputSilent(Vector<double> x)
Parameters
Type Name Description
Vector<double> x

HandleTriggerInput(object?, object?)

Processes Trigger messages: target vectors start training, null stops it.

Declaration
private void HandleTriggerInput(object? sender, object? data)
Parameters
Type Name Description
object sender
object data

LoadModel(string)

Declaration
public void LoadModel(string path)
Parameters
Type Name Description
string path

OnConfidenceChanged(double)

Executes the logic for when Confidence just changed.

Declaration
private void OnConfidenceChanged(double value)
Parameters
Type Name Description
double value

The new property value that was set.

Remarks

This method is invoked right after the value of Confidence is changed.

OnConfidenceChanged(double, double)

Executes the logic for when Confidence just changed.

Declaration
private void OnConfidenceChanged(double oldValue, double newValue)
Parameters
Type Name Description
double oldValue

The previous property value that was replaced.

double newValue

The new property value that was set.

Remarks

This method is invoked right after the value of Confidence is changed.

OnConfidenceChanging(double)

Executes the logic for when Confidence is changing.

Declaration
private void OnConfidenceChanging(double value)
Parameters
Type Name Description
double value

The new property value being set.

Remarks

This method is invoked right before the value of Confidence is changed.

OnConfidenceChanging(double, double)

Executes the logic for when Confidence is changing.

Declaration
private void OnConfidenceChanging(double oldValue, double newValue)
Parameters
Type Name Description
double oldValue

The previous property value that is being replaced.

double newValue

The new property value being set.

Remarks

This method is invoked right before the value of Confidence is changed.

OnCurrentTargetNameChanged(string)

Executes the logic for when CurrentTargetName just changed.

Declaration
private void OnCurrentTargetNameChanged(string value)
Parameters
Type Name Description
string value

The new property value that was set.

Remarks

This method is invoked right after the value of CurrentTargetName is changed.

OnCurrentTargetNameChanged(string?, string)

Executes the logic for when CurrentTargetName just changed.

Declaration
private void OnCurrentTargetNameChanged(string? oldValue, string newValue)
Parameters
Type Name Description
string oldValue

The previous property value that was replaced.

string newValue

The new property value that was set.

Remarks

This method is invoked right after the value of CurrentTargetName is changed.

OnCurrentTargetNameChanging(string)

Executes the logic for when CurrentTargetName is changing.

Declaration
private void OnCurrentTargetNameChanging(string value)
Parameters
Type Name Description
string value

The new property value being set.

Remarks

This method is invoked right before the value of CurrentTargetName is changed.

OnCurrentTargetNameChanging(string?, string)

Executes the logic for when CurrentTargetName is changing.

Declaration
private void OnCurrentTargetNameChanging(string? oldValue, string newValue)
Parameters
Type Name Description
string oldValue

The previous property value that is being replaced.

string newValue

The new property value being set.

Remarks

This method is invoked right before the value of CurrentTargetName is changed.

OnIsTrainingChanged(bool)

Executes the logic for when IsTraining just changed.

Declaration
private void OnIsTrainingChanged(bool value)
Parameters
Type Name Description
bool value

The new property value that was set.

Remarks

This method is invoked right after the value of IsTraining is changed.

OnIsTrainingChanged(bool, bool)

Executes the logic for when IsTraining just changed.

Declaration
private void OnIsTrainingChanged(bool oldValue, bool newValue)
Parameters
Type Name Description
bool oldValue

The previous property value that was replaced.

bool newValue

The new property value that was set.

Remarks

This method is invoked right after the value of IsTraining is changed.

OnIsTrainingChanging(bool)

Executes the logic for when IsTraining is changing.

Declaration
private void OnIsTrainingChanging(bool value)
Parameters
Type Name Description
bool value

The new property value being set.

Remarks

This method is invoked right before the value of IsTraining is changed.

OnIsTrainingChanging(bool, bool)

Executes the logic for when IsTraining is changing.

Declaration
private void OnIsTrainingChanging(bool oldValue, bool newValue)
Parameters
Type Name Description
bool oldValue

The previous property value that is being replaced.

bool newValue

The new property value being set.

Remarks

This method is invoked right before the value of IsTraining is changed.

OnReceive(object, object)

Dispatches incoming data. Trigger inputs control training state; feature inputs (Vector or Matrix) drive prediction.

Declaration
protected override void OnReceive(object sender, object data)
Parameters
Type Name Description
object sender
object data
Overrides
BaseBlock.OnReceive(object, object)

OnSampleCountChanged(int)

Executes the logic for when SampleCount just changed.

Declaration
private void OnSampleCountChanged(int value)
Parameters
Type Name Description
int value

The new property value that was set.

Remarks

This method is invoked right after the value of SampleCount is changed.

OnSampleCountChanged(int, int)

Executes the logic for when SampleCount just changed.

Declaration
private void OnSampleCountChanged(int oldValue, int newValue)
Parameters
Type Name Description
int oldValue

The previous property value that was replaced.

int newValue

The new property value that was set.

Remarks

This method is invoked right after the value of SampleCount is changed.

OnSampleCountChanging(int)

Executes the logic for when SampleCount is changing.

Declaration
private void OnSampleCountChanging(int value)
Parameters
Type Name Description
int value

The new property value being set.

Remarks

This method is invoked right before the value of SampleCount is changed.

OnSampleCountChanging(int, int)

Executes the logic for when SampleCount is changing.

Declaration
private void OnSampleCountChanging(int oldValue, int newValue)
Parameters
Type Name Description
int oldValue

The previous property value that is being replaced.

int newValue

The new property value being set.

Remarks

This method is invoked right before the value of SampleCount is changed.

ResetModel()

Declaration
public void ResetModel()

SaveModel(string?)

Declaration
public void SaveModel(string? path = null)
Parameters
Type Name Description
string path

SetTarget(Vector<double>?, string?)

Declaration
public void SetTarget(Vector<double>? target, string? targetName = null)
Parameters
Type Name Description
Vector<double> target
string targetName

TryExtractLabels(object?)

Declaration
private void TryExtractLabels(object? sender)
Parameters
Type Name Description
object sender

Events

OnLabelsUpdated

Fired once when output labels are derived.

Declaration
public event Action<List<string>>? OnLabelsUpdated
Event Type
Type Description
Action<List<string>>

OnPrediction

Fired after every prediction.

Declaration
public event Action<Vector<double>, Vector<double>?, double, int>? OnPrediction
Event Type
Type Description
Action<Vector<double>, Vector<double>, double, int>

OnTrainingStateChanged

Fired when training starts/stops.

Declaration
public event Action<bool, string?>? OnTrainingStateChanged
Event Type
Type Description
Action<bool, string>

Implements

INotifyPropertyChanged
INotifyPropertyChanging
ISubscriber
IPublisher
IBaseBlockProps
IJsonExportable
IDisposable
IAsyncDisposable

Search documentation

Start typing to search the documentation.