https://github.com/yahiakr/kerasga

A simple and easy-to-use implementation of Genetic Algorithm for Keras neural network models.
https://github.com/yahiakr/kerasga

genetic-algorithm keras neural-network

Last synced: 4 months ago
JSON representation

A simple and easy-to-use implementation of Genetic Algorithm for Keras neural network models.

• Host: GitHub
• URL: https://github.com/yahiakr/kerasga
• Owner: yahiakr
• License: gpl-3.0
• Created: 2019-12-13T01:48:26.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2020-02-08T23:33:34.000Z (almost 5 years ago)
• Last Synced: 2023-11-20T16:20:17.407Z (about 1 year ago)
• Topics: genetic-algorithm, keras, neural-network
• Language: Python
• Homepage: https://pypi.org/project/KerasGA/
• Size: 24.4 KB
• Stars: 4
• Watchers: 3
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# KerasGA

A simple and easy-to-use implementation of Genetic Algorithm for Keras NN models in Python.

## Features

* Instead of using backpropagation, use Genetic Algorithm to update the parameters/weights of the neural network while training the model.

* Create an initial population (of size: `population_size`) of randomly initialized chromosomes (i.e model weights).

* You can adjust the `selection_rate` & the `mutation_rate`.

* Perform the different GA operations (i.e Selection, Crossover, & Mutation).

## Examples

Here are a few projects based on this package:

* [yahiakr/FlappyAI](https://github.com/yahiakr/FlappyAI)

* [yahiakr/SnakeAI](https://github.com/yahiakr/SnakeAI)

## Usage

* Install `KerasGA` :

```

$ pip install KerasGA

```

* import `GeneticAlgorithm` from `KerasGA` and initiate an object :

```python

from KerasGA import GeneticAlgorithm

population_size =  10

GA = GeneticAlgorithm(model, population_size = population_size, selection_rate = 0.1, mutation_rate = 0.2)

```

**PS:** `model` is a Keras model.

* Generate the initial population:

```python

population = GA.initial_population()

```

* To set the wights of a model you can use `.set_weights()` built-in function:

```python

for chromosome in population:

	model.set_weights(chromosome)

	# then evaluate the chromosome (i.e assign its final score)

```

* After calculating the scores for each chromosome, it's time to select the top-performers:

```python

# Selection:

# 'scores' is a list of length = population_size

# 'top_performers' is a list of tuples: (chromosome, it's score)

top_performers = GA.strongest_parents(population,scores)

# Make pairs:

# 'GA.pair' return a tuple of type: (chromosome, it's score)

pairs = []

while len(pairs) != GA.population_size:

	pairs.append( GA.pair(top_performers) )

# Crossover:

base_offsprings =  []

for pair in pairs:

	offsprings = GA.crossover(pair[0][0], pair[1][0])

	# 'offsprings' contains two chromosomes

	base_offsprings.append(offsprings[-1])

# Mutation:

new_population = GA.mutation(base_offsprings)

```

And that's it :)