Welcome to nimblenet’s documentation!

nimblenet is a lightweight and efficient Numpy library for creating feed forward neural networks. The library was developed with PYPY in mind and should play nicely with their super-fast JIT compiler. The networks can be trained by a variety of learning algorithms: backpropagation, resilient backpropagation, adaptive learning rate backpropagation, scaled conjugate gradient and SciPy’s optimize function.

This is a list of handy links to get up and running.

• Getting Started
• Cost Functions
• Activation Functions
• Saving and loading a trained network

Installing

$ pip install nimblenet

Dependencies

• Python 2.7
• NumPy
• SciPy (optional). This is of course a required depedency if you intend to train the network using SciPy’s optimize function.

Content

• Getting Started
  • Installing
  • Creating a Network
  • Training the Network
  • Using the Network
  • Putting it all together
• Initializing a Network
  • Parameters
  • Example
• Saving and loading a trained network
  • Save
  • Load
• Preprocessing
  • Usage
  • Available preprocessors
• Gradient Checking
  • Usage
• Activation Functions
  • Usage
  • List of cost functions
  • Arbitrary Activation Functions
• Learning Algorithms
  • Backpropagation Variations
  • Additional Learning Algorithms
• Cost Functions
  • Usage
  • List of cost functions
  • Arbitrary Cost Functions
• Using the Network
• Support

Support

Have you spotted a bug, or run into inconsistencies in the documentation? Please report the issue at Github.