This is a Python extension module written in C and Cython for implementing artificial neural networks and performing supervised machine learning. This is still in the experimental/prototyping stage of development and not at all tested. Feedback is greatly appreciated.
###Install The package source comes with a standard distutils setup script. The basic install command is:
python setup.py install
from the root directory.
#####Dependencies
- python >= 2.7
- cython >= 0.17
- numpy >= 1.8.2
- glibc >= 2.13
###Usage
The following program creates and performs backpropagation on a neural network to generate outputs equal to .5 when all it's inputs equal 1:
from net import Network
import numpy as np
# Backpropagation estimates partial derivatives by averaging the
# estimates of a certain number of test cases, known as the batch size.
batch_size = 20
# We will run backpropagation on a hundred batches.
train_size = batch_size * 100
# Data is typically modeled as a 2-d numpy array, where rows designate
# observations and columns designate elements of a vector. The input
# and output data we create has the same value at each observation.
in_v = np.array([[1.0, 1.0, 1.0, 1.0, 1.0]] * train_size)
out_v = np.array([[0.5, 0.5, 0.5]] * train_size)
# Create a new layered network with 5 inputs,
# 10 hidden neurons and 3 output neurons.
n = Network.Layered([10, 3], in_v)
n.backprop(expect=out_v, batch=batch_size)