daherk2 / neuralpy

neuralpy - neural network library written in python

neuralpy 1.3.0

Within this package is the most intuitive fully-connected multilayer neural network model. Data science shouldn't have a high barrier to entry. neuralpy handles the math and overhead while you focus on the data.

neuralpy is a neural network model written in python based on Michael Nielsen's neural networks and deep learning book.

• Visit the neuralpy website
• Get detailed examples and explanations in the Official Documentation
• Contribute on Github

Getting Started (quick start)

The following demonstrates how to download and install neuralpy and how to create and train a simple neural network. Run the following command to download and install:

$ pip install neuralpy

Create a neural network in your project by specifying the number of nodes in each layer. Random weights and biases will automatically be generated:

import neuralpy
net = neuralpy.Network([2, 3, 1])

The network feeds input vectors as python lists forward and returns the output vector as a list:

x = [1, 1]
output = net.forward(x)
print output
# ex: [0.11471727263613461]

Train the neural network by first generating training data in the form of a list of tuples. Each tuple has two components and each component is a list representing the input and output respectively. This training set represents the simple OR function and it can be generated for you to save typing:

training_data = neuralpy.load_or()
# [
#   ([1, 1], [1]),
#   ([1, 0], [1]),
#   ...
# ]

Then we must specify the remaining hyperparameters. Let's say we want to limit it to 100 epochs and give it a learning rate of 1:

epochs = 100
learning_rate = 1

Then run the train method with the parameters. We're telling the network to conform to training data:

net.train(training_data, epochs, learning_rate)

Now feed forward the input from earlier and the output should be closer to 1.0, which is what we trained the network to do:

output = net.forward(x)
print output            # ex: [0.9542129706170075]

There is more information about advanced options such as monitoring the cost in the official documentation.

Since, this is a multilayer feedforward neural network, it is a universal approximator (Hornik, Stinchcombe and White, 1989). Neural Networks can be used for a wide range of applications from image processing to time series prediction.

• "You abandoned me. You left me to die."
• "Well, I wouldn't have done it if I'd known you were going to hassle me about it."