Feature API

The Feature is an abstract class that encapsulates both numerical and semantic attributes of time series data through its subclasses, NumericalFeature and CategoricalFeature. These base classes serve as the foundation for concrete classes such as Peak, Min, Max, etc., providing a structured approach to define features of time series data. Developers can extend this base class to implement new features.

classDiagram
    direction TB
    class FeatureFactory {
        -data: TimeSeriesData
        +searchNumericalFeature(FeatureName, ...)
    }
    class Feature {
        <<abstract>>
        -date: Date
        -name: FeatureName
        +setHeight(number) this
        +getHeight() number
    }
    class NumericalFeature {
        <<abstract>>
    }
    class CategoricalFeature

    FeatureFactory ..> Feature : creates
    Feature <|-- NumericalFeature
    Feature <|-- CategoricalFeature
    NumericalFeature <|-- First
    NumericalFeature <|-- Current
    NumericalFeature <|-- Last
    NumericalFeature <|-- Max
    NumericalFeature <|-- Min
    NumericalFeature <|-- Peak
    NumericalFeature <|-- Valley
    NumericalFeature <|-- Fall
    NumericalFeature <|-- Raise
    NumericalFeature <|-- Slope

The class diagram of the MSB feature classes. Please see the source code for all classes, their methods and attributes.

Feature names

Features are defined as enumerators, e.g.,

export enum NumericalFeatureName {
  FIRST = "FIRST",
  CURRENT = "CURRENT",
  LAST = "LAST",
  MAX = "MAX",
  MIN = "MIN",
  PEAK = "PEAK",
  VALLEY = "VALLEY",
  FALL = "FALL",
  RAISE = "RAISE",
  SLOPE = "SLOPE",
  ...
}

Creating an MSB feature

To instantiate a feature, use the constructor of a concrete feature class. Set properties using method chaining. For example, when a feature PEAK is detected it creates a Peak object as follows. The default feature properties are used unless defined in the feature-action table. Use the getter functions to retrieve the feature properties.

new Peak()
    .setDate(...)
    .setHeight(...)
    .setNormWidth(...)
    .setNormHeight(...)
    .setRank(...)
    .setMetric(...)
    .setStart(...)
    .setEnd(...)
    .setDataIndex(...);

FeatureFactory

The FeatureFactory class implements a factory design pattern for streamlined feature creation, utilizing search functions to dynamically generate feature instances based on input feature-action table and time series data.

new FeatureFactory()
    .setProps()
    .setData()
    .searchNumericalFeatures(<feature name>, <properties>, ...);