How to build a Recurrent Network using PyBrain

Recipe Objective - How to build a Recurrent Network using PyBrain?

Recurrent networks are similar to the feed-through network; the only difference is that you must remember the data at each step. The history of each step must be saved.

For more related projects -

https://www.projectpro.io/projects/data-science-projects/deep-learning-projects
https://www.projectpro.io/projects/data-science-projects/keras-deep-learning-projects

Let's try to build a recurrent network -

# Importing libraries
from pybrain.structure import RecurrentNetwork
from pybrain.structure import LinearLayer, SigmoidLayer
from pybrain.structure import FullConnection

# Building feed forward network model
recurrent_network = RecurrentNetwork()

# Making layer for input => 3 , hidden=> 3 and output=>1
input_layer = LinearLayer(3)
hidden_layer = SigmoidLayer(3)
output_layer = LinearLayer(1)

# Adding the layer to feedforward network
recurrent_network.addInputModule(input_layer)
recurrent_network.addModule(hidden_layer)
recurrent_network.addOutputModule(output_layer)

# Making connection between input ,hidden and output
input_to_hidden = FullConnection(input_layer, hidden_layer)
hidden_to_output = FullConnection(hidden_layer, output_layer)

# Adding connection to the network
recurrent_network.addConnection(input_to_hidden)
recurrent_network.addConnection(hidden_to_output)

# Add recurrent connection from hidden to hidden layers
recurrent_network.addRecurrentConnection(FullConnection(hidden_layer, hidden_layer))

# Sorting modules
recurrent_network.sortModules()

# Printing network
print(recurrent_network)

# Activating network by passing 3 input values
print("Output => ",recurrent_network.activate((1, 1, 1)))

# Clearing the history of the network by calling reset method
recurrent_network.reset()


Output -
RecurrentNetwork-7
   Modules:
    [<LinearLayer 'LinearLayer-4'>, <SigmoidLayer 'SigmoidLayer-8'>, <LinearLayer 'LinearLayer-9'>]
   Connections:
    [<FullConnection 'FullConnection-5': 'LinearLayer-4' -> 'SigmoidLayer-8'>, <FullConnection 'FullConnection-6': 'SigmoidLayer-8' -> 'LinearLayer-9'>]
   Recurrent Connections:
    [<FullConnection 'FullConnection-3': 'SigmoidLayer-8' -> 'SigmoidLayer-8'>]
Output =>  [0.1918294]

In this way, we can build a recurrent network using pybrain.

​