meco 1.0.1

Quick start

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

$ pip install meco

Example

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))
model.fit(X, 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)
classifier.fit(x_reduced, y_reduced)

Tasks

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

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

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

Citing

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

@inproceedings{barbiero2019novel,
  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)},
  pages={68--81},
  year={2019},
  organization={Springer}
}

@article{barbiero2020uncovering,
  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},
  year={2020}
}

Source

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

Authors

Pietro Barbiero, Giovanni Squillero, and Alberto Tonda.

Licence

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: http://www.apache.org/licenses/LICENSE-2.0.

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.