Introduction

Genetic Algorithm in Python, which could be used for Sampling, Feature Select, Model Select, etc in Machine Learning

Install

sudo pip3 install genetics

Usage

API

Initialize the Object

GA(train, valid, estimator, groups=100, iter=200, r_sample=0.8, r_crossover=0.5, r_vary=0.01, r_keep_best=0.1, n_jobs=4, verbose=False)

• train: a 2D numpy matrix, the last column will be used as the labels
• valid: a 2D numpy matrix, which’s columns should be the same as the train
• estimator: a SKLearn estimator, such as RandomForestClassifier or SVR etc.
• groups: the groups in every generation, default 200
• iter: the number of iterations, the procedure will stop when reach the number. default 200
• r_sample: useful when doing sampling, the ratio for sampling, deault 0.8
• r_crossover: the ratio of crossover when generating a children from his parents, default 0.5
• r_vary: the ratio for varying when generating a child from his parents, default 0.01, suggest 0 - 0.1
• r_keep_best: the ratio for keeping the best groups in every generation, default 0.1
• n_jobs: the number for running procedure in parallel, default 4
• verbose: the flag for showing the verbose messages, default False

Sampling

# Example
from genetics import GA
# the sample_result is a 2D numpy matrix, which is the result after sampling
# the sample_genes is the gene used for selecting instances, just ignore it if you don't need it
# the sample_scores is the final score when doing validation in valid set
(sample_result, sample_genes, sample_scores) = GA(train, valid, RandomForestClassfier).select_instance()
# Or you can do sampling by calling this
(sample_result, sample_genes, sample_scores) = GA(train, valid, RandomForestClassfier).select(axis=0)

Feature Selection

# Example
from genetics import GA
# the sample_result is a 2D numpy matrix, which is the result after selecting feature
# the sample_genes is the gene used for selecting features, just ignore it if you don't need it
# the sample_scores is the final score when doing validation in valid set
(sample_result, sample_genes, sample_scores) = GA(train, valid, RandomForestClassfier).select_feature()
# Or you can select features by calling this
(sample_result, sample_genes, sample_scores) = GA(train, valid, RandomForestClassfier).select(axis=1)

Citation

Zeng X, Yuan S, Huang X, et al. 
Identification of cytokine via an improved genetic algorithm[J]. 
Frontiers of Computer Science, 2015, 9(4): 643-651.