DeepGProp 1.0.8

Train Multilayer Perceptrons with Genetic Algorithms.

Neural Networks optimization with Genetic Algorithms

Based on the BSc thesis by

• Author: Luis Liñán Villafranca

• Mentor: Juan Julián Merelo Guervós

---

Table of contents

• Installation

  • Virtual environment creation

  • Installing the DeepGProp CLI

  • Extra modules

• Tutorials

• Tests and formatting

• Frameworks

• Utilities

• Datasets

• Licence

---

Installation

The first prerequisite is to have Python 3.6, 3.7 or 3.8 and pip installed on the system. It is recommended to create a virtual environment to isolate the used package versions. For more information about pip and venv check the official tutorial.

If you are using pyenv, remember to compile your version with the –enable-shared configuration option:

First, you need to install a version of python that’s been compiled with -fPIC. pyenv versions by default are not, so you will need to issue something like this:

env PYTHON_CONFIGURE_OPTS="--enable-shared" pyenv install 3.8.6

Virtual environment creation

We can use a core module to create the virtual environment, it’s been working since version 3.3

python -m venv .venv

Please make sure when you do this that all __pycache__ directories have been deleted; otherwise, it might fail in some unexpected place.

This will create a virtual environment in the .venv directory. Once that’s been done, we need to activate it; use one of the following commands (depending on the interpreter) (obtained from the official venv documentation):

Platform	Shell	Command to activate virtual environment
POSIX	bash/zsh	$ source <venv>/bin/activate
	fish	$ . <venv>/bin/activate.fish
	csh/tcsh	$ source <venv>/bin/activate.csh
	PowerShell Core	$ <venv>/bin/Activate.ps1
Windows	cmd.exe	C:\> <venv>\Scripts\activate.bat
	PowerShell	PS C:\> <venv>\Scripts\Activate.ps1

Table 1.1: Activating the virtual environment.

You won’t need to create the virtual environment in the case you’re using global installation of modules via version managers such as pyenv.

Installing the DeepGProp CLI

To run DeepGProp first we need to install its cli. You can install it with pip:

pip install -U DeepGProp

Or downloading the repository with:

pip install .

On the other hand, if we want the code to be updated as we change it, we will need to install DeepGProp in editable mode. To do this, we need to add the option -e/--editable to the installation command:

pip install -e .

All the available options can be listed using:

dgp --help

Extra modules

I’ve divided all the used packages in different groups to avoid installing undesirable ones for specific use of the repository:

Purpose	File path	Description
Test	requirements/tests.txt	Necessary packages for tests. Nox installs them automaticly when running the tests.
Lint	requirements/lint.txt	Necessary packages for linting. Nox installs them automaticly when linting the code.
Format	requirements/format.txt	Necessary packages for formatting. Nox installs them automaticly when running format command.
Dev	requirements/dev.txt	All above packages.

To install any of these packages you can run:

pip install -r <file path>

If you are not using any virtual environment, make sure you install these packages so that they are available in the required Python version.

Tutorials

• Runing DGP on the MNIST dataset

Tests and formatting

Note

To be able to run the DeepGProp tests, you will need to install it in editable mode. checkout in Installing the DeepGProp CLI section how to do it.

First, we need to install the Nox tool:

pip install -U nox

To run all the tests:

nox -k test

To run the linters:

nox -k lint

You can check all the possible sessions with the following command:

nox -l

Frameworks

• Keras - base library to create and run the neural networks.

• DEAP - genetic algorithms library used to optimize the models hyper parametters.

Utilities

• Automation:

  • Nox - automation tool to run different tasks as the tests or the code formatting check.

• Tests:

  • pytest - Python test framework to run the tests.

Datasets

All datasets need to have a first row with the column names, and one of the columns needs to be named class. For the time being, it’s prepared to run only classification problems.

Licence

The original code can be found in the DeepGProp repo under GPLv3 License.