Recurrent Neural Network library for Ruby which uses the tlearn neural network simulator(http://crl.ucsd.edu/innate/tlearn.html)

To see an example run:

rake example:train
rake example:fitness

TLearn supports recurrent networks (http://en.wikipedia.org/wiki/Recurrent_neural_network), FANN does not. Recurrent networks maintain state, enabling the context of previous inputs to effect further outputs.

While there have been attempts to add recurrent networks to FANN (http://leenissen.dk/fann/forum/viewtopic.php?t=47) these are still sitting on a dead branch that was never merged into master.

gem install tlearn

#Usage

require 'tlearn'

tlearn = TLearn::Run.new(:number_of_nodes => 86,
                         :output_nodes    => 41..46,
                         :linear          => 47..86,
                         :weight_limit    => 1.00,
                         :connections     => [{1..81   => 0},
                                              {1..40   => :i1..:i77},
                                              {41..46  => 1..40},
                                              {1..40   => 47..86},
                                              {47..86  => [1..40, {:max => 1.0, :min => 1.0}, :fixed, :one_to_one]}])
                 
  
training_data = [{[0] * 77 => [1, 0, 0, 0, 0, 0]}],
                [{[1] * 77 => [0, 0, 0, 0, 0, 1]}]
  
tlearn.train(training_data, sweeps = 200)

tlearn.fitness([0] * 77, sweeps = 200)
# => ["0.016", "0.013", "0.022", "0.020", "0.463", "0.467"]

You will often seperate the training and fitness process. By specifying a working directory when training you can reuse the training data for later fitness evaulations.

tlearn.train(training_data, sweeps = 200, working_dir='/training_session/')

tlearn.fitness([0] * 77, sweeps = 200, working_dir = '/training_session/')
# => ["0.016", "0.013", "0.022", "0.020", "0.463", "0.467"]

Yes, its complicated configuring this thing. Lets work through the different configuration options:

#Total number of nodes in the network (not counting input).
:number_of_nodes => 86

#The nodes that are used for output.
:'output_nodes' => 41..46

nodes 1-10 are linear. Linear nodes ouput the inner-product of the input and weight vectors
:linear_nodes => 1..10

#nodes 1-10 are bipolar. Bipolar nodes output ranges continuously from -1 to +1.
:bipolar_nodes => 1..10

#weights between nodes will not exceed 1.00
:weight_limit => 1.00

We use ranges to specify connections between nodes:

1..3 => 4..6

Indicates connections:

node 1 <- node 4 
node 1 <- node 5
node 1 <- node 6

node 2 <- node 4
node 2 <- node 5
node 2 <- node 6

node 3 <- node 4
node 3 <- node 5
node 3 <- node 6

Note that the nodes specified first (1..3) are the destination nodes, the second nodes (4..6) are the source nodes. The sources nodes feed into the destination nodes.

Make sure you feed the input nodes into the network.

#1-6 nodes are fed input nodes 1-10.
1..6 => i1..i10

We can also add constraints to the different connections:

#nodes 1-6 are fed from node 0 (node 0 is always the bias node).
1..6 => 0

#nodes 1-6 connections with nodes 7-10 are fixed at initiation values and will not change throughout learning.
1..6 => [7..10, :fixed]

#nodes 1-6 connections with nodes 7-10 are fixed at 2 and will not change throughout learning.
1..6 => [7..9, {:min => 2, :max => 2}]

#1-6 nodes connections with nodes from 7-9 are fixed at weight 1. 
1..3 => [7..9, {:min => 1.0, :max => 1.0}, :fixed, :'one_to_one']

one_to_one changes the way connections are mapped. Instead of one node mapping to every other node we have a 1-1 mapping between nodes:

For example:

node 1 -> node 7
mode 2 -> node 8
node 3 -> node 9

There is also the TLearn manual if you want read more:

http://blog.josephwilk.net/uploads/tlearn.pdf

#C TLearn Library

The Original TLearn C library was written by:

• Kim Plunkett
• Jeffrey L. Elman

#Contributors

• Joseph Wilk blog | twitter

#License

(The MIT License)

Copyright (c) 2012 Joseph Wilk

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the 'Software'), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED 'AS IS', WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.