smaa

mcda.smaa

Stochastic multi-criteria acceptability analysis on an MCDA run.

Classes

Name	Description
SMAAReport	Outcome of an SMAA weight-sampling sensitivity analysis.

SMAAReport

mcda.smaa.SMAAReport(self, base, sampled_weights, sampled_ranks, rank_acceptability_index, central_weight_vector, confidence_factor, method, n_samples, seed)

Outcome of an SMAA weight-sampling sensitivity analysis.

Holds a reference run with equal weights, the sampled weight matrix, the per-sample rank matrix, and three summary fields: the rank acceptability index, the central weight vector per tool, and the confidence factor per tool (Lahdelma and Salminen 2001).

Shapes: sampled_weights: (n_samples, n_metrics) sampled_ranks: (n_samples, n_tools), each row a rank permutation rank_acceptability_index: (n_tools, n_tools), entry [a, k - 1] is the empirical probability that tool a obtains rank k. central_weight_vector: (n_tools, n_metrics), the mean of the sampled weight vectors restricted to samples where tool a is top-ranked; row a is the zero vector if tool a is never top-ranked. confidence_factor: (n_tools,), the share of samples in which tool a is top-ranked; equal to rank_acceptability_index[:, 0].

Functions

Name	Description
smaa	Run an SMAA-style weight-sampling sensitivity analysis on scores.

smaa

mcda.smaa.smaa(scores, polarity, n_samples=1000, method='saw', alpha=None, seed=None, normalization=None, bounds=None, baselines=None, targets=None, missing='error')

Run an SMAA-style weight-sampling sensitivity analysis on scores.

Draw n_samples weight vectors from a Dirichlet over the metrics simplex (so every sampled vector is non-negative and sums to 1). For each draw, run the full MCDA pipeline with run and record the rank vector. Tabulate three summaries:

1. The rank acceptability index, the empirical probability per tool and per rank.
2. The central weight vector per tool, the mean of the sampled weights restricted to samples where that tool is top-ranked. A tool that is never top-ranked gets the zero vector.
3. The confidence factor per tool, the share of samples in which the tool is top-ranked.

Parameters

Name	Type	Description	Default
scores		Array-like of shape (n_tools, n_metrics).	required
polarity	Sequence[str]	Length n_metrics sequence of "higher_is_better" or "lower_is_better". Use beam.cards.polarities_for to source this from the registry.	required
n_samples	int	Number of weight vectors to draw. Defaults to 1000.	1000
method	str	Aggregation method forwarded to run. Any of the five methods supported by run: "saw", "topsis", "vikor", "promethee_ii" or "comet".	'saw'
alpha	Sequence[float] | None	Optional length n_metrics concentration vector for the Dirichlet draw. Defaults to ones, which gives a uniform distribution over the simplex.	None
seed	int | None	Optional integer seed for the random generator. Recorded in the report so the run can be reproduced.	None
normalization		Optional per-metric normalization context forwarded to run. Default None keeps the run defaults (min-max with empirical extrema). Pass the values resolved by beam.mcda.registry_context so the SMAA analysis normalizes the scores the same way as the headline ranking.	None
bounds		Optional per-metric normalization context forwarded to run. Default None keeps the run defaults (min-max with empirical extrema). Pass the values resolved by beam.mcda.registry_context so the SMAA analysis normalizes the scores the same way as the headline ranking.	None
baselines		Optional per-metric normalization context forwarded to run. Default None keeps the run defaults (min-max with empirical extrema). Pass the values resolved by beam.mcda.registry_context so the SMAA analysis normalizes the scores the same way as the headline ranking.	None
targets		Optional per-metric normalization context forwarded to run. Default None keeps the run defaults (min-max with empirical extrema). Pass the values resolved by beam.mcda.registry_context so the SMAA analysis normalizes the scores the same way as the headline ranking.	None
missing	str	Missing-data policy forwarded to every run call; see beam.mcda.run. Defaults to "error".	'error'

Returns

Type	Description
SMAAReport