SKMoefs

SKMoefs is a Python module for machine learning specifically built for Multi-Objective Evolutionary Fuzzy Systems. It takes inspiration by the Scikit-Learn principles and their classes are all ScikitLearn's estimators. In addition, it builds upon Platypus, a library that deals with multi-objective algorithms.

Porting to Julia

This fork focuses on porting the original implementation to Julia.

Installation

Python Dependencies

skmoefs requires:

- Python (>= 3.6)
- NumPy
- SciPy
- Numba
- Matplotlib
- Scikit-Learn
- Platypus

In order to use the library, it is necessary to have the above dependencies installed:

pip install pandas numpy scipy numba matplotlib scikit-learn platypus-opt

Julia Dependencies

The Julia port requires the following packages to be installed:

- Julia (>= 1.7.2)
- ScikitLearn (0.6.4
- BisectPy (1.0.0
- JLD (0.13.1
- PyCall (1.93.1)

In order to use the library, it is necessary to have the above dependencies installed:

julia> import Pkg; Pkg.add("ScikitLearn"); Pkg.add("BisectPy"); Pkg.add("JLD"); Pkg.add("PyCall"); Pkg.add("Plots");

Source code

You can check the latest sources with the command::

git clone https://github.com/GionatanG/skmoefs.git

Testing

You can run some tests by executing::

cd python
python example.py

Examples

The simplest example is shown below::

from platypus.algorithms import *

from skmoefs.toolbox import MPAES_RCS, load_dataset, normalize
from skmoefs.rcs import RCSInitializer, RCSVariator
from skmoefs.discretization.discretizer_base import fuzzyDiscretization
from sklearn.model_selection import train_test_split

X, y, attributes, inputs, outputs = load_dataset('iris')
X_n, y_n = normalize(X, y, attributes)
Xtr, Xte, ytr, yte = train_test_split(X_n, y_n, test_size=0.3)
my_moefs = MPAES_RCS(variator=RCSVariator(), initializer=RCSInitializer())
my_moefs.fit(Xtr, ytr, max_evals=10000)

my_moefs.show_pareto()
my_moefs.show_pareto(Xte, yte)
my_moefs.show_model('median', inputs=inputs, outputs=outputs)

The program load the IRIS dataset from the built-in datasets and normalize the matrix X. Indeed, the input data should be in the form

X : real NumPy matrix NxM (number of samples x number of features), with 0 <= X[i,j] <= 1 for every i,j
y : integer Numpy vector Nx1. In particular, if number of classes is C then y[i] belongs to {1, 2, 3, ...., C} for every i

After the normalization, the script splits the dataset into training and testing. It defines and train an MPAES_RCS object with default parameters, variator and initializer. Finally, it shows the results by plotting the pareto for the training set, another pareto evaluated for the testing set, and RB/DB for the classifier with an accuracy that is the median among the values within the archive.

Citation

If you use SK-MOEFS in a scientific publication, we would appreciate citations to the following papers:

Gallo G., Ferrari V., Marcelloni F., Ducange P. (2020) SK-MOEFS: A Library in Python for Designing Accurate and Explainable Fuzzy Models. In: Lesot MJ. et al. (eds) Information Processing and Management of Uncertainty in Knowledge-Based Systems. IPMU 2020. Communications in Computer and Information Science, vol 1239. Springer, Cham
Antonelli, M., Ducange, P., Marcelloni, F.: A fast and efficient multi-objective evolutionary learning scheme for fuzzy rule-based classifiers. Inf. Sci. 283, 36-54 (2014)

License

SK-MOEFS is released under the GNU General Public License GPLv3 (See https://www.gnu.org/licenses/)

rambodrahmani / skmoefs