Analytic Hierarchy Process Solver

The analytic hierarchy process (AHP) is a structured technique for organizing and analyzing complex decisions, based on mathematics and psychology. It was developed by Thomas L. Saaty in the 1970s and has been extensively studied and refined since then.[1]

The Wikipedia [2] page on AHP references two full examples of AHP and many more can be found on the internet.

pyAHP provides a flexible interface to build AHP models and solve them using a plethora of methods. Checkout the documentation here.

license CircleCI

Installation

To install pyAHP, simply:

pip install pyahp

Getting Started

Using as a python module

import json
from pyahp import parse

with open('model.json') as json_model:
    # model can also be a python dictionary
    model = json.load(json_model)

ahp_model = parse(model)
priorities = ahp_model.get_priorities()

Using on the command line

$> python -m pyahp -f examples/television.json
    [+] Television Model
        Method: eigenvalue
        Results:
            Samsung: 0.243
            Sony: 0.106
            Panasonic: 0.27
            Toshiba: 0.38
        Recommended is Toshiba

Model Schema

The models supplied to the library are in JSON format. The model has to follow a specific schema and a number of errors are raised in case the schema validation fails. A very simple model with three criteria and one criteria with two subcriteria and three alternatives is as follows:

{
  "name": "Sample Model",
  "method": "approximate",
  "criteria": ["critA", "critB", "critC"],
  "subCriteria": {
    "critA": ["subCritA", "subCritB"]
  },
  "alternatives": ["altA", "altB", "altC"],
  "preferenceMatrices": {
    "criteria": [
      [1, 1, 1],
      [1, 1, 1],
      [1, 1, 1]
    ],
    "subCriteria:critA": [
      [1, 1],
      [1, 1]
    ],
    "alternatives:subCritA": [
      [1, 1, 1],
      [1, 1, 1],
      [1, 1, 1]
    ],
    "alternatives:subCritB": [
      [1, 1, 1],
      [1, 1, 1],
      [1, 1, 1]
    ],
    "alternatives:critB": [
      [1, 1, 1],
      [1, 1, 1],
      [1, 1, 1]
    ],
    "alternatives:critC": [
      [1, 1, 1],
      [1, 1, 1],
      [1, 1, 1]
    ]
  }
}

Supported Methods

There are a wide variety of methods available for calculating the priorities from preference matrices. This library currently supports the following methods:

  • Approximate (approximate)

  • Geometric (geometric)

  • Eigenvalue (eigenvalue)

Fields in the model

Field

Type

Description

name

string

Name of the model. It is used when the library is called from the command line and ignored when used as a python library. Defaults to the filename in the command line mode.

method

string

The method/algorithm used to calculate the priority vectors from the preference matrices. It should be one of the supported methods. required

criteria

array

An array of strings containing the names of all the top level criteria. All the names should be unique. required

subCriteria

object

It contains the sub-criteria definitions with criterion as the key and an array of strings as the sub-criteria.

alternatives

array

An array of strings containing the names of all the alternatives. All the strings should be unique. required

preferenceMatrices

object

An object with key of the form criteria or subCriteria:<criteriaName> or alternative:<criteriaName> and the value is a 2D square matrix with integer elements. required

In the sample model above, due to the design of the model and hierarchy, critA has two sub-criteria. Hence, we need to provide a preference matrix for the sub-criteria of critA, named subCriteria:critA, and two alternative preferences matrices with the name alternatives:subCritA and alternatives:subCritB. All the other criteria have corresponding preference matrices.

Maintainer

Last updated