manoloesparta / logicnet

🧠 Neural network form scratch

Logic Neural Network

This is a 2 hidden layer neural network made with love :)
This was original made for learning logical gates but you can use with any normalized data.

Requirements

git clone https://github.com/manoloesparta/logicnet
pip install -r requirements.txt

Usage

Instance of a neural network

from neuralnetwork import NeuralNetwork

input_nodes = 2
hidden_layer1_nodes = 100
hidden_layer2_nodes = 100
output_nodes = 1
learning_rate = 0.3

ann = NeuralNetwork(input_nodes, hidden_layer1_nodes, hidden_layer2_nodes, output_nodes, learning_rate)

Creating the data for the neural network

NOTE: The only data it creates is about logical gates using the functions AND or OR. If you want to use your own data, you need to your normalize data and feed it to the network

from data import DataGenerator

num_samples = 100

data =  DataGenerator(num_samples).logic_data()
x = data['train_data']
y = data['target_data']

Training neural network

epochs = 1000

ann.train(x, y, epochs)

Predicting

unknown_answer = [[0,1]]

ann.predict(unknown_answer)

Saving and loading weights

Saving Weights

When a full training is complete automatically weights are saved in:

./src/weights/

Loading Weights

You can use this method without and instance of the class, and create the full neural network so you can start predicting

from neuralnetwork import NeuralNetwork

ann = NeuralNetwork.load_weights()
ann.predict([[0,1]])

Dokcer

In order to build your docker image use the command

docker build -t ann .

Finally in order to run your container use

docker run --rm -v $(pwd):/usr/src/ -it ann bash

License

This project is licensed under the MIT License