Jujube

Jujube is a library for doing machine learning using neural networks.

For instructions on how to use the library see the wiki. If you find any bugs or if anything is unclear, do not hesitate to open an issue.

Features

Architecture

• Fully connected

Training

• Stochastic gradient descent
• Backpropagation
• Momentum technique
• Regularization
• Early stopping

Activation functions

• Sigmoid
• Tanh
• Linear

Cost functions

• Quadratic cost
• Cross-entropy

Regularization techniques

• L1
• L2

Data sets

• MNIST database (source: Dariel Dato-on)
• Abalone (source: UCI)
• Sinus

Third-party libraries

• numpy

Credits

• Michael Nielsen for his online book Neural Networks and Deep Learning