solarjoe / neurolab

Neurolab is a simple and powerful Neural Network Library for Python

Neurolab

Neurolab is a simple and powerful Neural Network Library for Python. Contains based neural networks, train algorithms and flexible framework to create and explore other neural network types.

Features

• Pure python + numpy
• API like Neural Network Toolbox (NNT) from MATLAB
• Interface to use train algorithms form scipy.optimize
• Flexible network configurations and learning algorithms. You may change: train, error, initializetion and activation functions
• Unlimited number of neural layers and number of neurons in layers
• Variety of supported types of Artificial Neural Network and learning algorithms

Example

    >>> import numpy as np
    >>> import neurolab as nl
    >>> # Create train samples
    >>> input = np.random.uniform(-0.5, 0.5, (10, 2))
    >>> target = (input[:, 0] + input[:, 1]).reshape(10, 1)
    >>> # Create network with 2 inputs, 5 neurons in input layer
    >>> # And 1 in output layer
    >>> net = nl.net.newff([[-0.5, 0.5], [-0.5, 0.5]], [5, 1])
    >>> # Train process
    >>> err = net.train(input, target, show=15)
    Epoch: 15; Error: 0.150308402918;
    Epoch: 30; Error: 0.072265865089;
    Epoch: 45; Error: 0.016931355131;
    The goal of learning is reached
    >>> # Test
    >>> net.sim([[0.2, 0.1]]) # 0.2 + 0.1
    array([[ 0.28757596]])

Links

• Home Page: http://code.google.com/p/neurolab/
• PyPI Page: http://pypi.python.org/pypi/neurolab
• Documentation: http://packages.python.org/neurolab/
• Examples: http://packages.python.org/neurolab/example.html

Install

Install neurolab using pip:

    $> pip install neurolab

Or, if you don't have setuptools/distribute installed, use the download link at right to download the source package, and install it in the normal fashion. Ungzip and untar the source package, cd to the new directory, and:

    $> python setup.py install

Support neural networks types

• Single layer perceptron
  • create function: neurolab.net.newp()
  • example of use: newp
  • default train function: neurolab.train.train_delta()
  • support train functions: train_gd, train_gda, train_gdm, train_gdx, train_rprop, train_bfgs, train_cg
• Multilayer feed forward perceptron
  • create function: neurolab.net.newff()
  • example of use: newff
  • default train function: neurolab.train.train_bfgs()
  • support train functions: train_gd, train_gda, train_gdm, train_rprop, train_bfgs, train_cg
• Competing layer (Kohonen Layer)
  • create function: neurolab.net.newc()
  • example of use: newc
  • default train function: neurolab.train.train_cwta()
  • support train functions: train_wta
• Learning Vector Quantization (LVQ)
  • create function: neurolab.net.newlvq()
  • example of use: newlvq
  • default train function: neurolab.train.train_lvq()
• Elman Recurrent network
  • create function: neurolab.net.newelm()
  • example of use: newelm
  • default train function: neurolab.train.train_gdx()
  • support train functions: train_gd, train_gda, train_gdm, train_rprop, train_bfgs, train_cg
• Hopfield Recurrent network
  • create function: neurolab.net.newhop()
  • example of use: newhop
• Hemming Recurrent network
  • create function: neurolab.net.newhem()
  • example of use: newhem