TimeSeriesTransformerPipeline<T>

Models & Types · AiDotNet.Preprocessing.TimeSeries

A pipeline that chains multiple time series transformers together.

Implements: ITimeSeriesFeatureExtractor<T>

For Beginners

Think of this like an assembly line in a factory:

Raw Material -> [Machine 1] -> [Machine 2] -> [Machine 3] -> Final Product

Similarly, your data flows through each transformer in order:

Raw Data -> [Lag Features] -> [Rolling Stats] -> [Technical Indicators] -> Enhanced Features

This makes it easy to build complex feature engineering workflows by combining simple, focused transformers.

How It Works

This class allows you to compose multiple transformers into a single pipeline, applying them in sequence. Each transformer’s output becomes the next transformer’s input.

Example

var data = Tensor<double>.CreateRandom(2, 4);
// One window configuration, shared by every transformer in the pipeline.
var featureOptions = new TimeSeriesFeatureOptions { WindowSizes = [7, 30] };
            
var lagTransformer = new LagLeadTransformer<double>(featureOptions);
var statsTransformer = new RollingStatsTransformer<double>(featureOptions);
var indicatorTransformer = new TechnicalIndicatorsTransformer<double>(featureOptions);
            
// Build a pipeline
var pipeline = new TimeSeriesTransformerPipeline<double>()
    .AddTransformer(lagTransformer)
    .AddTransformer(statsTransformer)
    .AddTransformer(indicatorTransformer);
            
// Fit and transform
var enrichedData = pipeline.FitTransform(data);

Constructors

Constructor Summary
TimeSeriesTransformerPipeline() Creates a new pipeline with default settings.
TimeSeriesTransformerPipeline(bool includeOriginalFeatures, PipelineMode pipelineMode) Creates a new pipeline with the specified settings.

Properties

Property Summary
ITimeSeriesFeatureExtractor<T> Item { get; }
bool AutoDetectEnabled { get; }
bool IsFitted { get; set; }
bool SupportsIncrementalTransform { get; }
bool SupportsInverseTransform { get; }
int InputFeatureCount { get; }
int OutputFeatureCount { get; }
int TransformerCount { get; } Gets the number of transformers in the pipeline.
int[] ColumnIndices { get; }
int[] WindowSizes { get; }
string[] FeatureNames { get; }

Methods

Method Summary
IncrementalState<T> GetIncrementalState()
List<string> GetValidationErrors(Tensor<T> data)
T[] TransformIncremental(T[] newDataPoint)
Tensor<T> FitTransform(Tensor<T> data)
Tensor<T> InverseTransform(Tensor<T> data)
Tensor<T> Transform(Tensor<T> data)
TimeSeriesTransformerPipeline<T> AddTransformer(ITimeSeriesFeatureExtractor<T> transformer) Adds a transformer to the end of the pipeline.
TimeSeriesTransformerPipeline<T> AddTransformers(IEnumerable<ITimeSeriesFeatureExtractor<T>> transformers) Adds multiple transformers to the pipeline.
TimeSeriesTransformerPipeline<T> AddTransformers(ITimeSeriesFeatureExtractor<T>[] transformers) Adds multiple transformers to the pipeline.
TimeSeriesTransformerPipeline<T> Clone() Creates a deep copy of this pipeline (transformers are not copied, only the pipeline structure).
bool ValidateInput(Tensor<T> data)
int[] DetectOptimalWindowSizes(Tensor<T> data)
string GetSummary() Gets a summary of the pipeline configuration.
string[] GetFeatureNamesOut(string[] inputFeatureNames)
void Fit(Tensor<T> data)
void InitializeIncremental(Tensor<T> historicalData)
void SetIncrementalState(IncrementalState<T> state)