Artificial Immune Systems Package

A Python package for Artificial Immune Systems algorithms

Language

• Português

Documentation

• Official Docs
• Github Wiki

Introduction

AISP is a python package that implements artificial immune systems techniques, distributed under the GNU Lesser General Public License v3.0 (LGPLv3).

The package started in 2022 as a research package at the Federal Institute of Northern Minas Gerais - Salinas campus (IFNMG - Salinas).

Artificial Immune Systems (AIS) are inspired by the vertebrate immune system, creating metaphors that apply the ability to detect and catalog pathogens, among other features of this system.

What can you do with AISP?

AISP provides implementations of bio-inspired algorithms for:

• Anomaly detection: Identify abnormal patterns in data.
• Classification: Classify data with multiple classes.
• Optimization: Find optimal solutions for objective functions.
• Clustering: Group data without supervision.

Implemented Algorithms

Negative Selection (aisp.nsa)

• BNSA - Binary Negative Selection Algorithm
• RNSA - Real-Valued Negative Selection Algorithm

Clonal Selection (aisp.csa)

• AIRS - Artificial Immune Recognition System
• CLONALG - Clonal Selection Algorithm

Immune Network Theory (aisp.ina)

• AiNet - Artificial Immune Network for clustering and data compression

Module in Development

Danger Theory (aisp.dta)

• DCA - Dendritic Cell Algorithm (planned)

API overview

All algorithms follow a simple and consistent interface:

• fit(X, y, verbose: bool = True): trains the model for classification tasks.
• fit(X, verbose: bool = True): trains the model for clustering tasks.
• predict(X): makes predictions based on new data.
• optimize(max_iters: int =..., n_iter_no_change: int =..., verbose: bool = True): run the optimization algorithms

Installation

The module requires installation of python 3.10 or higher.

Dependencies

User installation

The simplest way to install AISP is using pip:

pip install aisp

Quick Start

Below are minimal examples demonstrating how to use AISP for different tasks.

Classification with RNSA

import numpy as np
from aisp.nsa import RNSA

# Generating training data
np.random.seed(1)
class_a = np.random.uniform(high=0.5, size=(50, 2))
class_b = np.random.uniform(low=0.51, size=(50, 2))
x_train = np.vstack((class_a, class_b))
y_train = ['a'] * 50 + ['b'] * 50

# Training the model
model = RNSA(N=150, r=0.3, seed=1)
model.fit(x_train, y_train, verbose=False)

# Predict
x_test = [
    [0.15, 0.45],  # Expected: 'a'
    [0.85, 0.65],  # Expected: 'b'
]

y_pred = model.predict(x_test)
print(y_pred)

Clustering with AiNet

import numpy as np
from aisp.ina import AiNet

np.random.seed(1)
# Generating training data
a = np.random.uniform(high=0.4, size=(50, 2))
b = np.random.uniform(low=0.6, size=(50, 2))
x_train = np.vstack((a, b))

# Training the model
model = AiNet(
    N=150,
    mst_inconsistency_factor=1,
    seed=1,
    affinity_threshold=0.85,
    suppression_threshold=0.7
)

model.fit(x_train, verbose=False)

# Predict cluster labels
x_test = [
    [0.15, 0.45],
    [0.85, 0.65],
]

y_pred = model.predict(x_test)
print(y_pred)

Optimization with CLONALG

import numpy as np
from aisp.csa import Clonalg

# Define search space
bounds = {'low': -5.12, 'high': 5.12}

# Objective function (Rastrigin)
def rastrigin(x):
    x = np.clip(x, bounds['low'], bounds['high'])
    return 10 * len(x) + np.sum(x**2 - 10 * np.cos(2 * np.pi * x))

# Initialize optimizer
model = Clonalg(problem_size=2, rate_hypermutation=0.5, bounds=bounds, seed=1)
model.register('affinity_function', rastrigin)

# Run optimization
population = model.optimize(100, 50, False)

print(model.best_solution, model.best_cost) # Best solution

Examples

Explore the example notebooks available in the AIS-Package/aisp repository. These notebooks demonstrate how to utilize the package's functionalities in various scenarios, including applications of the RNSA, BNSA and AIRS algorithms on datasets such as Iris, Geyser, and Mushrooms.

You can run the notebooks directly in your browser without any local installation using Binder:

💡 Tip: Binder may take a few minutes to load the environment, especially on the first launch.