Neural

Table of Contents

• Neural
  • Table of Contents
  • Introduction
  • Features
  • Requirements
  • Installation
  • Usage
    • Network Class
      • Attributes
      • Methods
    • Layer Class
      • Attributes
      • Methods
    • Neuron Class
      • Attributes
      • Methods
    • Crossover Functions
      • Layer Crossover
      • Neuron Crossover
  • Roadmap
  • Contributing
  • Licensing

Introduction

The Neural project provides classes for building neural networks, including Network, Layer, and Neuron classes. It also includes crossover functions for evolving networks.

Features

• Flexible and customizable neural network architecture
• Support for various activation functions
• Crossover functions for evolving networks
• Enum classes for activation, random weight generation, and derivative functions

Requirements

• Python 3.x

Installation

Clone the repository:

git clone https://github.com/estevaopbs/Neural.git

Usage

Network Class

The Network class represents a neural network. It consists of layers, and each layer contains neurons.

Attributes

• layers: List of Layer objects.

Methods

• evaluate(inputs: List[Number]) -> List[Number]: Evaluates the network with the given inputs.
• enforce_weights(): Enforces consistent weights across layers.
• mutate_weight(): Mutates a random weight in the network.
• mutate_bias(): Mutates a random bias in the network.
• mutate_layer_weights(): Mutates weights in a random layer.
• mutate_layer_biases(): Mutates biases in a random layer.
• mutate_weights(): Mutates weights in all layers.
• mutate_biases(): Mutates biases in all layers.
• to_json(filename: Path): Saves the network to a JSON file.
• from_json(filename: Path): Loads the network from a JSON file.
• to_pickle(filename: Path): Saves the network to a pickle file.
• from_pickle(filename: Path): Loads the network from a pickle file.

Layer Class

The Layer class represents a layer in a neural network.

Attributes

• neurons: List of Neuron objects.

Methods

• evaluate(inputs: List[Number]) -> List[Number]: Evaluates the layer with the given inputs.
• mutate_weight(): Mutates a random weight in the layer.
• mutate_bias(): Mutates a random bias in the layer.
• mutate_weights(neurons: int): Mutates weights in a specified number of neurons.
• mutate_biases(neurons: int): Mutates biases in a specified number of neurons.
• enforce_weights(weights: int): Enforces a consistent number of weights across neurons.
• from_data(data, NeuronClass: Type[Neuron] = Neuron) -> Layer: Creates a layer from data.
• from_random(...) -> Layer: Creates a layer with random weights and biases.

Neuron Class

The Neuron class represents a single neuron in a neural network.

Attributes

• activation: Activation function.
• weights: List of weights.
• bias: Bias value.
• second_step: Secondary activation function.
• rand_weights_generator: Random weight generator function.
• rand_bias_generator: Random bias generator function.
• derivative: Derivative function.
• evaluate: Internal evaluation function.

Methods

• evaluate(inputs: List[Number]) -> Number: Evaluates the neuron with the given inputs.
• mutate_weights(): Mutates a random weight in the neuron.
• mutate_bias(): Mutates the bias of the neuron.
• mutate_weights_and_bias(): Mutates both weights and bias.
• mutate_weights_or_bias(): Mutates either weights or bias.
• from_random(...) -> Neuron: Creates a neuron with random parameters.
• enforce_weights(weights: int): Enforces a consistent number of weights.

Crossover Functions

Layer Crossover

layer_crossover(parent: Network, donor: Network, layers: int) -> Network

Performs crossover at the layer level between a parent and a donor network.

Neuron Crossover

neuron_crossover(parent: Network, donor: Network, neurons: int) -> Network

Performs crossover at the neuron level between a parent and a donor network.

Roadmap

• Implement backpropagation for training networks.
• Implement unit tests for code coverage.

Contributing

Contributions are welcome! Feel free to open issues or pull requests.

Licensing

This project is licensed under the MIT License.