Skip to main content

MECO: Multi-objective Evolutionary Compression.

Project description

Travis (.org) Codecov PyPI license PyPI

The MECO (Multi-objective Evolutionary COmpression) algorithm is a tool to perform:

  • dataset compression,

  • feature selection, and

  • coreset discovery.

This python package provides a sklearn-like transformer implementation of the MECO algorithm.

Quick start

You can install the meco package along with all its dependencies from PyPI:

$ pip install meco


For this simple experiment, let’s use the digits dataset from sklearn. We first need to import the dataset, a simple sklearn classifier (e.g. Ridge) and the MECO transformer. We can then load the dataset, create a MECO model, and fit the model on the digits dataset:

from sklearn.datasets import load_digits
from sklearn.linear_model import RidgeClassifier

from meco import MECO

X, y = load_digits(return_X_y=True)

model = MECO(RidgeClassifier(random_state=42)), y)

Once training is over, we get a view of the compressed input data X containing the most relevant samples (i.e. a subset of the rows in X, a.k.a. the coreset), and the most relevant features (i.e. a subset of the columns in X):

x_reduced = model.transform(X)

Once trained, the model.best_set_ dictionary contains:

  • the indices of the most relevant samples,

  • the indices of the most relevant features, and

  • the validation accuracy of the compressed dataset x_reduced, e.g.:

>>> model.best_set_
    'samples': [0, 2, 4, ...],
    'features': [3, 7, 8, ...],
    'accuracy': 0.9219,

The compressed dataset (x_reduced, y_reduced) can be used instead of the original dataset (X, y) to train machine learning models more efficiently:

from sklearn.ensemble import RandomForestClassifier

y_reduced = y[model.best_set_['samples']]

classifier = RandomForestClassifier(random_state=42), y_reduced)


Dataset compression

Should you need to compress the whole dataset X (i.e. for dataset compression), you can set the parameter compression to 'both' (this is the default behaviour anyway):

model = MECO(RidgeClassifier(), compression='both')

Coreset discovery

Should you need to compress the rows of X only (i.e. for coreset discovery), you can set the parameter compression to 'samples':

model = MECO(RidgeClassifier(), compression='samples')

Feature selection

Should you need to compress the columns of X only (i.e. for feature selection), you can set the parameter compression to 'features':

model = MECO(RidgeClassifier(), compression='features')


If you find MECO useful in your research, please consider citing the following papers:

  title={A Novel Outlook on Feature Selection as a Multi-objective Problem},
  author={Barbiero, Pietro and Lutton, Evelyne and Squillero, Giovanni and Tonda, Alberto},
  booktitle={International Conference on Artificial Evolution (Evolution Artificielle)},

  title={Uncovering Coresets for Classification With Multi-Objective Evolutionary Algorithms},
  author={Barbiero, Pietro and Squillero, Giovanni and Tonda, Alberto},
  journal={arXiv preprint arXiv:2002.08645},


The source code and minimal working examples can be found on GitHub.


Pietro Barbiero, Giovanni Squillero, and Alberto Tonda.


Copyright 2020 Pietro Barbiero, Giovanni Squillero, and Alberto Tonda.

Licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at:

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

See the License for the specific language governing permissions and limitations under the License.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

meco-1.0.1.tar.gz (16.5 kB view hashes)

Uploaded source

Built Distribution

meco-1.0.1-py3-none-any.whl (14.1 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page