CarbonComputed / neural

A neural networking library.

Neural

A neural networking library written in Python

Example:

from neural.connections.full import FullConnection
from neural.connections.output import OutputConnection
from neural.layers.sigmoid import SigmoidLayer
from neural.layers.linear import LinearLayer
from neural.networks.feedforward import FeedForwardNetwork
from neural.datasets.dataset import Dataset
from neural.datasets.xor import XORDataset
import neural.utilities

class Test():
        def __init__(self):
            self.input = LinearLayer(2,True)
            self.hidden = SigmoidLayer(50,True)
            self.output = SigmoidLayer(1,False)
            self.output2 = LinearLayer(1,False)
            self.in_hidden = FullConnection(self.input,self.hidden)
            self.hidden_out = FullConnection(self.hidden,self.output)
            self.out_out2 = OutputConnection(self.output,self.output2)
            self.network = FeedForwardNetwork(self.out_out2)
            self.network.add_layer(self.input)
            self.network.add_layer(self.hidden)
            self.network.add_layer(self.output)
            self.network.add_layer(self.output2)
            self.network.add_connection(self.in_hidden)
            self.network.add_connection(self.hidden_out)
        
        def test(self):
            dataset = XORDataset()
            self.network.train(dataset, alpha = 0.7, momentum=0)
            test_data = Dataset()
            test_data.add([1,0],[0])
            self.network.test(test_data)
            for n in self.network.layers[-1]:
                print n