@vapurrmaid/markov-chain

A lightweight TS library for computations with markov chains and probability matrices.

Installation

# yarn
yarn add @vapurrmaid/markov-chain

# npm
npm install --save @vapurrmaid/markov-chain

Modules

Markov Chain

Represents a finite, discrete-time Markov Chain.

The capabilities of this module are:

• Probabilistic state transition (see next)
  • A state transition function can be used to dynamically update probabilities
• Reporting if the current state is terminal (see isTerminal)
  • A terminal state will always transition back to itself

MarkovChain Import

import { MarkovChain } from '@vapurrmaid/markov-chain'

MarkovChain Constructor

• Must supply a N x N array of probabilities as number[][]
• Must supply an array of values as T[] of size N
• Optionally supply an initialState in [0, N)
  • If none is supplied, the default initialState = 0

Each row in the matrix corresponds to an index in the values array.

const values = ["a", "b", "c"]
const m = [
  [0, 1, 0], // always selects row 1 = index 1 = "b"
  [0, 0, 1], // always selects row 2 = index 2 = "c"
  [1, 0, 0]  // always selects row 0 = index 0 = "a"
]
const mc = new MarkovChain(values, m)

current Property

• Returns the value associated to the current state (row) as T

hasTransitionFn Property

• Returns true if a transition function is defined, otherwise false
• see setTransitionFn

isTerminal Property

• Returns true if the current row is terminal
• Returns false if the current row is not terminal

length Property

• Returns the size of the matrix, N as number

probabilityMatrix Property

• Returns the probability matrix as: number[][]

next() Method

• Computes and returns the next value as T using the probability matrix
• If a transition function is set, runs the transition function

setTransitionFn(prev, next) Method

• Sets a transition function that is used to alter the probability matrix
  • Prev is the index (row) of the state before next is called
  • Next is the next index (row) computed after next is called

Probability Matrix

Represents a probability matrix (aka transition matrix, Markov matrix or stochastic matrix). In typical mathematical representation, a probability matrix is formed as:

Which represents the probability of transitioning to the ith column from the jth column. However, column vectors are less intuitive in programming, as they require methods that span multiple arrays.

Instead, in this implementation each row vector entry represents transitioning from the ith row to the jth row. Therefore this representation is a transpose of the mathematical definition: ProbabilityMatrix = PT

Example

[
  [0, 1, 0], // row 0
  [0.5, 0, 0.5], // row 1
  [1, 0, 0], // row 2
];

In the above example, row 1 ([0.5, 0, 0.5]) reads:

• P=0.5 to transition from row 1 to row 0
• P=0 to transition from row 1 to row 1
• P=0.5 to transition from row 1 to row 2

ProbabilityMatrix Import

import { ProbabilityMatrix } from '@vapurrmaid/markov-chain'

ProbabilityMatrix Constructor

• Must be N x N
• Each row must add to 1.0
  • Each value must be in the interval [0, 1]

const m = [
  [0, 1, 0], // P = 1.0 to transition to row 1
  [0, 0, 1], // P = 1.0 to transition to row 2
  [1, 0, 0]  // P = 1.0 to transition to row 0
]
const matrix = new ProbabilityMatrix(m)

Properties

• value - returns the supplied probabilities as number[][]
• length - returns the size of the matrix, N as number

getRowVector(aRow) Method

• Returns the probability vector for the specified row as number[]
• aRow must be a number in the interval [0, N)

selectFrom(aRow) Method

• Using the probabilities defined in the given row, selects the next row as number
• aRow must be a number in the interval [0, N)

From the matrix defined above:

let nextRow = matrix.selectFrom(0)   // 1
nextRow = matrix.selectFrom(nextRow) // 2
nextRow = matrix.selectFrom(nextRow) // 0