mcda-promethee 0.4.1

PROMETHEE I/II/Gamma/Sort MCDA methods

promethee

Python implementation of the PROMETHEE family of MCDA (Multi-Criteria Decision Analysis) methods, maintained by the SMG research group at ULB. Built on mcda-core.

Implements PROMETHEE I, II, III, Gamma, GAIA, FlowSort, and P2Clust, with all six standard preference functions.

About SMG

The Decision Support Systems (SMG) group is a research lab at Université libre de Bruxelles (ULB) specialising in multi-criteria decision analysis. The group develops theoretical foundations and practical tools for MCDA, with a focus on the PROMETHEE methods.

If you are using this library for research, feel free to reach out at [e-mail] — we are happy to help you apply the methods and to start scientific collaborations.

Installation

pip install mcda-promethee

Python ≥ 3.9.

Quick start

For a hands-on walkthrough, see the demo notebook.

import numpy as np
from promethee import PrometheeII, PerformanceTable, ParametersPromethee, build_preference_functions

# 4 countries evaluated on 2 indicators
table = PerformanceTable(
    values=np.array([
        [77.2, 0.887],
        [76.5, 0.832],
        [80.1, 0.919],
        [73.4, 0.798],
    ]),
    alternatives=["Country A", "Country B", "Country C", "Country D"],
    criteria=["life_expectancy", "education_index"],
    directions=[1, 1],  # maximise both
)

weights = np.array([0.5, 0.5])
prefs = build_preference_functions([
    ("v_shape_with_indifference", {"p": 5.0, "q": 1.0}),
    ("v_shape_with_indifference", {"p": 0.08, "q": 0.02}),
])

params = ParametersPromethee(weights=weights, preference_functions=prefs)
model = PrometheeII(performance_table=table, parameters=params)
model.fit()

print(model.net_flows)   # φ score per alternative
print(model.ranking)     # indices sorted best → worst

Parameters can also be loaded from a JSON file (see tests/data/epi_parameters.json for an example):

import json
from promethee import ParametersPromethee

with open("parameters.json") as f:
    params = ParametersPromethee.from_dict(json.load(f))

Preference functions

All six standard PROMETHEE preference functions are available:

kind string	parameters	description
"usual"	—	Full preference for any positive difference
"u_shape"	q	Step at indifference threshold
"v_shape"	p	Linear up to preference threshold
"v_shape_with_indifference"	q, p	Linear between indifference and preference thresholds
"level"	q, p	Half preference in the intermediate zone
"gaussian"	s	Smooth Gaussian curve (inflection point s)

from promethee import build_preference_functions

prefs = build_preference_functions([
    ("usual", {}),
    ("v_shape_with_indifference", {"q": 2.0, "p": 8.0}),
    ("gaussian", {"s": 3.0}),
])

API reference

PrometheeII

Computes a complete ranking via net flows φ.

property	description
net_flows	φ score per alternative, shape (n,)
pairwise_preferences	aggregated preference matrix π, shape (n, n)
monocriterion_net_flows	mono-criterion net flows φ_k, shape (n, m)
ranking	indices sorted best → worst
result()	returns net_flows

All properties trigger fit() automatically if not already called.

ParametersPromethee

field	type	description
weights	ndarray (m,)	criterion weights (need not sum to 1)
preference_functions	Sequence[PreferenceFunction]	one per criterion

Constructors:

• ParametersPromethee(weights, preference_functions) — direct
• ParametersPromethee.from_dict(data) — from a parameter dictionary
• ParametersPromethee.from_weights_and_criteria(weights, criteria_types, p, q) — shorthand using V-shape with indifference

Quality indicators

Four indicators measure how well the net flows capture the pairwise preference structure:

from promethee import r_squared, additive_transitivity_index, pairwise_agreement_ratio, spectral_rank2_energy_ratio

pi  = model.pairwise_preferences
phi = model.net_flows

print(r_squared(pi, phi))                   # how well φ explains π (0–1, higher is better)
print(additive_transitivity_index(pi))      # transitivity violations (0 = perfect)
print(pairwise_agreement_ratio(pi, phi))    # pairs where pairwise order agrees with φ
print(spectral_rank2_energy_ratio(pi))      # spectral consistency of π (0–1, higher is better)

Other algorithms

class	description
PrometheeI	Partial ranking via positive/negative flows
PrometheeIII	Interval-based preorder: each alternative gets [x(a), y(a)] centred on its net flow; overlapping intervals → indifference
PrometheeGamma	Extends II with γ-based incomparability detection (Dejaegere & De Smet, 2023)
PrometheeGAIA	SVD biplot of monocriterion net flows — visualises alternatives, criteria, and the decision stick δ
FlowSort	Assigns each alternative to an ordered category using limiting reference profiles (pessimistic rule)
P2Clust	Clusters alternatives by splitting the sorted net-flow sequence at the k−1 largest gaps

Demos of PrometheeIII, PrometheeGAIA, FlowSort, and P2Clust are included in the demo notebook.

All public classes and parameter types are importable directly from promethee.

PrometheeIII — ParametersPrometheeIII

Extends ParametersPromethee with one optional field:

field	type	default	description
alpha	float	1.0	interval half-width multiplier; larger → more indifferences

Key properties: net_flows, intervals (shape (n, 2)), relation (binary (n, n) outranking matrix).

PrometheeGAIA — ParametersPromethee

Key properties: alternatives (2D coordinates, shape (n, 2)), criteria (shape (m, 2)), delta (decision stick, shape (2,)), quality (fraction of variance explained by the plane).

FlowSort — ParametersFlowSort

Extends ParametersPromethee with:

field	type	description
reference_profiles	ndarray (p, m)	limiting profiles ordered best → worst
profile_names	Sequence[str], optional	labels for the p profiles
category_names	Sequence[str], optional	labels for the p+1 categories

Key properties: assignments (category index per alternative), net_flows.

P2Clust — ParametersP2Clust

Extends ParametersPromethee with:

field	type	description
k	int	number of clusters

Key properties: assignments (cluster index 0 = best), net_flows, ranking.

Ecosystem

Part of the modular MCDA Python ecosystem. See mcda-core for the full picture.

License

MIT — see LICENSE.