A connection works similar to a layer; an only difference is that it shifts the data from one node to the other in a network.

In this chapter, we are going to learn about −

• Understanding Connections
• Creating Connections

Understanding Connections

Here is a working example of connections used while creating a network.

Example

ffy.py

from pybrain.structure import FeedForwardNetwork
from pybrain.structure import LinearLayer, SigmoidLayer
from pybrain.structure import FullConnection

network = FeedForwardNetwork()

#creating layer for input => 2 , hidden=> 3 and output=>1
inputLayer = LinearLayer(2)
hiddenLayer = SigmoidLayer(3)
outputLayer = LinearLayer(1)

#adding the layer to feedforward network
network.addInputModule(inputLayer)
network.addModule(hiddenLayer)
network.addOutputModule(outputLayer)

#Create connection between input ,hidden and output
input_to_hidden = FullConnection(inputLayer, hiddenLayer)
hidden_to_output = FullConnection(hiddenLayer, outputLayer)

#add connection to the network
network.addConnection(input_to_hidden)
network.addConnection(hidden_to_output)
network.sortModules()

print(network)

Output

C:pybrainpybrainsrc>python ffn.py
FeedForwardNetwork-6
Modules:
[<LinearLayer ''LinearLayer-3''>, <SigmoidLayer ''SigmoidLayer-7''>, 
   <LinearLayer ''LinearLayer-8''>]
Connections:
[<FullConnection ''FullConnection-4'': ''SigmoidLayer-7'' -> ''LinearLayer-8''>, 
   <FullConnection ''FullConnection-5'': ''LinearLayer-3'' -> ''SigmoidLayer-7''>]

Creating Connections

In Pybrain, we can create connections by using the connection module as shown below −

Example

connect.py

from pybrain.structure.connections.connection import Connection
class YourConnection(Connection):
   def __init__(self, *args, **kwargs):
      Connection.__init__(self, *args, **kwargs)
   def _forwardImplementation(self, inbuf, outbuf):
      outbuf += inbuf
   def _backwardImplementation(self, outerr, inerr, inbuf):
      inerr += outer

To create a connection, there are 2 methods — _forwardImplementation() and _backwardImplementation().

The _forwardImplementation() is called with the output buffer of the incoming module which is inbuf, and the input buffer of the outgoing module called outbuf. The inbuf is added to the outgoing module outbuf.

The _backwardImplementation() is called with outerr, inerr, and inbuf. The outgoing module error is added to the incoming module error in _backwardImplementation().

Let us now use the YourConnection in a network.

testconnection.py

from pybrain.structure import FeedForwardNetwork
from pybrain.structure import LinearLayer, SigmoidLayer
from connect import YourConnection

network = FeedForwardNetwork()

#creating layer for input => 2 , hidden=> 3 and output=>1
inputLayer = LinearLayer(2)
hiddenLayer = SigmoidLayer(3)
outputLayer = LinearLayer(1)

#adding the layer to feedforward network
network.addInputModule(inputLayer)
network.addModule(hiddenLayer)
network.addOutputModule(outputLayer)

#Create connection between input ,hidden and output
input_to_hidden = YourConnection(inputLayer, hiddenLayer)
hidden_to_output = YourConnection(hiddenLayer, outputLayer)

#add connection to the network
network.addConnection(input_to_hidden)
network.addConnection(hidden_to_output)
network.sortModules()

print(network)

Output

C:pybrainpybrainsrc>python testconnection.py
FeedForwardNetwork-6
Modules:
[<LinearLayer ''LinearLayer-3''>, <SigmoidLayer ''SigmoidLayer-7''>, 
   <LinearLayer ''LinearLayer-8''>]
Connections:
[<YourConnection ''YourConnection-4'': ''LinearLayer-3'' -> ''SigmoidLayer-7''>, 
   <YourConnection ''YourConnection-5'': ''SigmoidLayer-7'' -> ''LinearLayer-8''>]