`skcriteria.pipelines.combinatorial` module

The Module implements utilities to build a combinatorial pipeline.

class skcriteria.pipelines.combinatorial.SKCCombinatorialPipeline(steps)[source]

Bases: SKCMethodABC

Model that encapsulates a pipeline of MCDA methods with alternatives.

This class allows you to define a sequential pipeline of data transformation and aggregation steps, where some steps may have multiple alternative implementations. The CombinatorialPipeline will generate all possible pipelines by combining these alternatives and evaluate them.

Parameters:

steps (list of (str, method or list of methods)) –

List of (name, transform) tuples (implementing fit/transform) that are chained, in the order in which they are chained. Steps can be a single method or a list of alternative methods.

# simple pipeline
steps = [
    ("inverter", invert_objectives.InvertObjectives()),
    ("scaler", scalers.SumScaler(target="matrix")),
    ("agg", simple.WeightedSum())
]

# pipeline with alternatives in the scaler step
steps = [
    ("inverter", invert_objectives.InvertObjectives()),
    (
        "scaler",
        [
            scalers.SumScaler(target="matrix"),
            scalers.VectorScaler(target="matrix"),
        ],
    ),
    ("agg", simple.WeightedSum()),
]

property steps: The raw steps provided during initialization.

property named_steps: The raw steps provided during initialization as a dict-like.

property pipelines: List of all generated pipelines.

property named_pipelines: A dict-like of all generated pipelines.

transform(dm)[source]

Transform the data, without applying the final evaluator.

This method applies the transformation steps (such as inverters and scalers) of each individual pipeline to the input decision matrix. It does not execute the final aggregation step (the evaluator), behaving analogously to the transform method of a standard SKCPipeline.

Since multiple pipelines are generated, this method does not return a single transformed decision matrix, but rather a dictionary-like containing all transformed matrices, each associated with the unique name of the pipeline that generated it.

Parameters:: dm (skcriteria.core.DecisionMatrix) – The decision matrix to transform.
Returns:: A dictionary mapping the name of each pipeline (str) to its corresponding transformed skcriteria.core.DecisionMatrix.
Return type:: dict-like

skcriteria.pipelines.combinatorial module

`skcriteria.pipelines.combinatorial` module