raptor-regression

Search for linear regressions in a dataset using sliding T-tests.

Install

npm install --save raptor-regression

Quick start

var regression = require('raptor-regression');
 
var offender = { value: 202, offender: true }; 
var data = [
  { value: 101 },
  ...
  offender,
  ...
  { value: 200 }
];
 
var regressions = regression(data);
//=>
[{
  regressor: {
    source: { ... },
    backAnalysis: { ... },
    foreAnalysis: { ... },
    tValue: 468.0...
  },
  previous: { ... },
  next: { ... }
}]
 
assert( regressions[0].regressor.source === offender );

API

@type Function
@param Array<{ value: Number }>
@returns Array<{ regressor: Object, previous: Object, next: Object }>

The API for this module is a single function which accepts a single array as input. This array should contain objects with a value property. For example:

var param = [
  { value: 100 }
]

The API will return a new array containing regression offenders along with their adjacent array members, and all relevant regression metadata. The source object is accessible in each regression entry in the source property.

Evaluating the response

Using the following example response as a guide:

[ { regressor:
   { source: { value: 202, offender: true },
     backAnalysis: { length: 12, average: 101.5, variance: 0.2727272727272727 },
     foreAnalysis: { length: 12, average: 201.5, variance: 0.2727272727272727 },
     tValue: 468.01974568701814 },
  previous:
   { source: { value: 101, offender: false },
     backAnalysis: { length: 12, average: 101.5, variance: 0.2727272727272727 },
     foreAnalysis:
      { length: 12,
        average: 193.16666666666666,
        variance: 842.6969696969695 },
     tValue: 9.773584847914274 },
  next:
   { source: { value: 201 },
     backAnalysis:
      { length: 12,
        average: 109.83333333333333,
        variance: 842.6969696969696 },
     foreAnalysis: { length: 12, average: 201.5, variance: 0.2727272727272727 },
     tValue: 9.773584847914275 } } ]

• The response is always an array of regressions. If no regressions are detected, the array will be empty.
• Every regression has three properties: regressor, previous, and next.
  • The regressor property contains the analysis information for the array entry which introduced a detectable change point.
  • The previous property contains the analysis information for the array entry prior to the regressor.
  • The next property contains the analysis information for the array entry after the regressor.
• Each of the regression sub-objects contains the following properties: source, backAnalysis, foreAnalysis, and tValue.
  • The source property is a reference to the original object for which an analysis was performed on. This object is not mutated and any extraneous properties on the object have no effect.
  • The backAnalysis contains an object which represents the statistical information for the window leading up to the source.
  • The foreAnalysis contains an object which represents the statistical information for the window appearing after the source.
  • The tValue is the normalized T-test score for the backward and forward windows. This number is the determinant for a detectable change point.