rann 0.1.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA1:
+  metadata.gz: b474ce98e49bb5067d2fc8ddba98a9c6f887c618
+  data.tar.gz: feb50a3a4804494c0a4c2551e7a2b3807b89dd69
+SHA512:
+  metadata.gz: fc4f352df3b64af7d6b4c513d0aefd343ec35b4b6b599e7154cdbb7eefd06365023c2192c1736e3d7298a6895b388db3047387cfb4e52e7136f97eb09a320e5d
+  data.tar.gz: eec1f19c738594b4b29c5de6da0268d249c4eaaddec7f243919f96ee73331025e571a52d215e894255c79f2451b55de5eb43caaa53a91d898048548190ee355e

data/.gitignore

@@ -0,0 +1,8 @@
+/.bundle/
+/.yardoc
+/_yardoc/
+/coverage/
+/doc/
+/pkg/
+/spec/reports/
+/tmp/

data/.travis.yml

@@ -0,0 +1,5 @@
+sudo: false
+language: ruby
+rvm:
+  - 2.3.1
+before_install: gem install bundler -v 1.16.0

data/CHANGES.md

@@ -0,0 +1,3 @@
+- Basic classes required for most neural network designs and backprop.
+
+  *Michael Campbell*

data/Gemfile

@@ -0,0 +1,5 @@
+source "https://rubygems.org"
+
+git_source(:github){ |repo_name| "https://github.com/#{repo_name}" }
+
+gemspec

data/Gemfile.lock

@@ -0,0 +1,26 @@
+PATH
+  remote: .
+  specs:
+    rann (0.1.0)
+      parallel (~> 1.12.0)
+      ruby-graphviz (~> 1.2.3)
+
+GEM
+  remote: https://rubygems.org/
+  specs:
+    minitest (5.10.3)
+    parallel (1.12.0)
+    rake (10.5.0)
+    ruby-graphviz (1.2.3)
+
+PLATFORMS
+  ruby
+
+DEPENDENCIES
+  bundler (~> 1.16)
+  minitest (~> 5.0)
+  rake (~> 10.0)
+  rann!
+
+BUNDLED WITH
+   1.16.0

data/LICENCE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2017 Michael Campbell
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

data/README.md

@@ -0,0 +1,59 @@
+# RANN
+
+This library provides objects and algorithms for designing, processing and
+training Artificial Neural Networks in Ruby.
+
+## Installation
+
+Add this line to your application's Gemfile:
+
+```ruby
+gem 'rann'
+```
+
+And then execute:
+
+    $ bundle
+
+Or install it yourself as:
+
+    $ gem install rann
+
+## Usage
+
+See examples/
+
+To run an example:
+
+```
+git clone https://github.com/mikecmpbll/rann.git
+cd rann
+bin/setup
+ruby examples/xor.rb
+```
+
+## TODO
+
+So much. So much.
+
+- Convenience methods for setting up standard network topologies, crucially,
+  layers
+- Batch normalization/drop out/early stopping
+- Hyperparameter optimisation
+- Other adaptive learning rate algorithms (Adadelta, Adam, etc?)
+- Explore matrix operations and other ways to optimise performance of algorithms
+- RPROP?
+- Use enumerable-statistics gem?
+- Speed up by adding a reduce step to the parallel gem?
+- More examples
+- Tests
+
+## Development
+
+After checking out the repo, run `bin/setup` to install dependencies. Then, run `rake test` to run the tests. You can also run `bin/console` for an interactive prompt that will allow you to experiment.
+
+To install this gem onto your local machine, run `bundle exec rake install`. To release a new version, update the version number in `version.rb`, and then run `bundle exec rake release`, which will create a git tag for the version, push git commits and tags, and push the `.gem` file to [rubygems.org](https://rubygems.org).
+
+## Contributing
+
+Bug reports and pull requests are welcome on GitHub at https://github.com/mikecmpbll/rann.

data/Rakefile

@@ -0,0 +1,10 @@
+require "bundler/gem_tasks"
+require "rake/testtask"
+
+Rake::TestTask.new :test do |t|
+  t.libs << "test"
+  t.libs << "lib"
+  t.test_files = FileList["test/**/*_test.rb"]
+end
+
+task default: :test

data/bin/console

@@ -0,0 +1,14 @@
+#!/usr/bin/env ruby
+
+require "bundler/setup"
+require "rann"
+
+# You can add fixtures and/or initialization code here to make experimenting
+# with your gem easier. You can also use a different console, if you like.
+
+# (If you use this, don't forget to add pry to your Gemfile!)
+# require "pry"
+# Pry.start
+
+require "irb"
+IRB.start(__FILE__)

data/bin/setup

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+set -euo pipefail
+IFS=$'\n\t'
+set -vx
+
+bundle install
+
+# Do any other automated setup that you need to do here

data/examples/xor.rb

@@ -0,0 +1,46 @@
+require "bundler/setup"
+require "rann"
+
+# inputs
+inputs = Array.new(2){ |i| RANN::Neuron.new "input #{i}", 0, :input }
+
+# hidden layer
+hiddens = Array.new(3){ |i| RANN::Neuron.new "hidden #{i}", 3 }
+bias = RANN::Neuron.new "bias", 0, :bias
+
+# output layer
+output = RANN::Neuron.new "output", 3, :output, :sig
+
+# connect it all
+connections = []
+hiddens.each do |h|
+  inputs.each do |i|
+    connections.push RANN::Connection.new i, h
+  end
+  connections.push RANN::Connection.new bias, h
+  connections.push RANN::Connection.new h, output
+end
+
+network = RANN::Network.new connections
+backprop = RANN::Backprop.new network
+
+inputs = [[0,0],[0,1],[1,0],[1,1]]
+targets = [[0],[1],[1],[0]]
+
+i = 0
+loop do
+  i += 1
+  sample_index = (rand * inputs.size).to_i
+
+  avg_error =
+    backprop.run_batch(
+      [inputs[sample_index].map(&:to_d)],
+      [targets[sample_index].map(&:to_d)],
+      processes: 0,
+      checking: false
+    )
+
+  puts "iteration #{i} error: #{avg_error.to_f}"
+
+  break if avg_error < 0.0001
+end

data/lib/rann/backprop.rb

@@ -0,0 +1,222 @@
+require "parallel"
+require "rann/gradient_checker"
+require "rann/util/array_ext"
+
+module RANN
+  class Backprop
+    include Util::ArrayExt
+
+    ACTIVATION_DERIVATIVES = {
+      relu:   ->(x){ x > 0 ? 1.to_d : 0.to_d },
+      sig:    ->(x){ x.mult(1 - x, 10) },
+      linear: ->(_){ 1.to_d },
+      tanh:   ->(x){ 1 - x.power(2, 10) },
+      step:   ->(_){ 0.to_d },
+    }
+
+    DECAY = BigDecimal.new('0.9')
+    MASTER_STEP_SIZE = BigDecimal.new('0.01')
+    FUDGE_FACTOR = BigDecimal.new('0.00000001')
+    LEARNING_RATE = BigDecimal.new('0.01')
+    FRICTION = BigDecimal.new('0.8')
+    NUM_ITERATIONS_BEFORE_LR_ANNEALING = BigDecimal.new('10')
+
+    attr_accessor :network, :lr, :velocities
+
+    def initialize network, restore = {}
+      @network             = network
+      @connections_hash    = network.connections.each.with_object({}){ |c, h| h[c.id] = c }
+      @lr                  = LEARNING_RATE
+      @friction            = FRICTION
+      @velocities          = Hash.new(BigDecimal.new('0'))
+      @historical_gradient = (restore[:historical_gradient] || {}).tap{ |h| h.default = 0.to_d }
+      @historical_update   = Hash.new(MASTER_STEP_SIZE)
+      @batch_count         = BigDecimal.new('0')
+    end
+
+    def run_batch(inputs, targets, opts = {})
+      @batch_count += 1
+
+      batch_size      = inputs.size
+      avg_gradients   = Hash.new{ |h, k| h[k] = 0 }
+      avg_batch_error = 0
+
+      # force longer bits of work per iteration, to maximise CPU usage
+      # less marshalling data etc, more work.
+      grouped_inputs  = in_groups inputs, [1, opts[:processes]].max * 10, false
+      grouped_results =
+        Parallel.map_with_index grouped_inputs, in_processes: opts[:processes] do |inputs, i|
+          group_avg_gradients = Hash.new{ |h, k| h[k] = 0.to_d }
+          group_avg_error     = 0.to_d
+
+          inputs.each do |input|
+            gradients, error = Backprop.run_single(network, input, targets[i])
+
+            gradients.each do |cid, g|
+              group_avg_gradients[cid] += g.div batch_size, 10
+            end
+            group_avg_error += error.div batch_size, 10
+          end
+
+          group_avg_gradients.default_proc = nil
+          [group_avg_gradients, group_avg_error]
+        end
+
+      grouped_results.each do |group_avg_gradients, group_avg_error|
+        avg_gradients.merge!(group_avg_gradients){ |_, o, n| o + n }
+        avg_batch_error += group_avg_error
+      end
+
+      if opts[:checking]
+        # check assumes batchsize 1 for now
+        sorted_gradients = avg_gradients.values_at *network.connections.map(&:id)
+        if GradientChecker.check network, inputs.first, targets.first, sorted_gradients
+          puts "gradient valid"
+        else
+          puts "gradient INVALID"
+        end
+      end
+
+      avg_gradients.each do |con_id, gradient|
+        con = @connections_hash[con_id]
+        next if con.locked?
+
+        update = adagrad gradient, con.id
+
+        con.weight += update
+      end
+
+      avg_batch_error
+    end
+
+    def self.run_single network, inputs, targets
+      states = []
+      inputs = [inputs] if inputs.flatten == inputs
+
+      # run the data into the network. (feed forward)
+      # all but last
+      (inputs.size - 1).times do |timestep|
+        network.evaluate inputs[timestep]
+        states[timestep] = network.reset!
+      end
+      # last
+      outputs = network.evaluate inputs.last
+      states[inputs.size - 1] = network.reset!
+
+      # calculate error
+      error = mse targets, outputs
+
+      # backward pass with unravelling for recurrent networks
+      deltas = Hash.new{ |h, k| h[k] = Hash.new(0.to_d) }
+
+      # outputs first
+      network.output_neurons.each.with_index do |o, i|
+        activation_derivative = ACTIVATION_DERIVATIVES[o.activation_function]
+
+        deltas[0][o.id] = mse_delta(targets[i], outputs[i], activation_derivative)
+      end
+
+      # remove this push mechanism, shouldn't be necessary and uses extra memory.
+      incoming_deltas = Hash.new{ |h, k| h[k] = Hash.new{ |h, k| h[k] = [] } }
+      # each timestep backwards through time
+      (inputs.size - 1).downto 0 do |t|
+        network.output_neurons.each do |o|
+          traverse from: o, network: network, timestep: t, deltas: deltas do |other, con|
+            if other.context?
+              this_t = t - 1
+              other  = o
+            else
+              this_t = t
+            end
+
+            incoming_deltas[this_t][other.id] <<
+              deltas[t][o.id].mult(con.weight, 10)
+
+            if incoming_deltas[this_t][other.id].size == network.connections_from(other).size
+              sum_of_deltas = incoming_deltas[this_t][other.id].reduce(:+)
+
+              deltas[this_t][other.id] =
+                ACTIVATION_DERIVATIVES[other.activation_function]
+                  .call(states[this_t][other.id])
+                  .mult(sum_of_deltas, 10)
+            end
+          end
+        end
+      end
+
+      gradients = {}
+
+      network.connections.each_with_index do |con, i|
+        gradients[con.id] = 0.to_d
+        next if con.output_neuron.context?
+
+        (inputs.size - 1).downto 0 do |t|
+          if nd = deltas[t][con.output_neuron.id]
+            gradient =
+              if con.input_neuron.context?
+                t == 0 ? 0.to_d : nd.mult(states[t - 1][con.input_neuron.id], 10)
+              else
+                nd.mult states[t][con.input_neuron.id], 10
+              end
+
+            gradients[con.id] += gradient
+          end
+        end
+      end
+
+      reset! network
+      [gradients, error]
+    end
+
+    def state
+      { historical_gradient: @historical_gradient }
+    end
+
+    def self.reset! network
+      network.reset!
+      network.neurons.select(&:context?).each{ |n| n.value = 0.to_d }
+    end
+
+    def adagrad avg_grad, cid
+      @historical_gradient[cid] = DECAY.mult(@historical_gradient[cid], 10) + (1 - DECAY).mult(avg_grad.power(2, 10), 10)
+
+      avg_grad.mult(- @lr.div((FUDGE_FACTOR + @historical_gradient[cid]).sqrt(10), 10), 10)
+    end
+
+    def self.mse targets, outputs
+      total_squared_error = 0.to_d
+
+      targets.size.times do |i|
+        total_squared_error += (targets[i] - outputs[i]).power(2, 10).div(2, 10)
+      end
+
+      total_squared_error
+    end
+
+    def self.mse_delta target, actual, activation_derivative
+      step_one = actual - target
+      step_two = activation_derivative.call actual
+
+      step_one.mult step_two, 10
+    end
+
+    def self.traverse from:, network:, timestep:, deltas:, &block
+      # halt traversal if reached next timestep.
+      return if from.context?
+
+      bptt_connecting_to(from, network, timestep, deltas).each do |n, c|
+        yield n, c
+
+        traverse from: n, network: network, timestep: timestep, deltas: deltas, &block
+      end
+    end
+
+    def self.bptt_connecting_to neuron, network, timestep, deltas
+      network.connections_to(neuron).each.with_object [] do |c, a|
+        unless c.input_neuron.input? || deltas[timestep].key?(c.input_neuron.id)
+          a << [c.input_neuron, c]
+        end
+      end
+    end
+  end
+end

data/lib/rann/connection.rb

@@ -0,0 +1,67 @@
+require "securerandom"
+require "bigdecimal"
+require "bigdecimal/util"
+
+module RANN
+  class Connection
+    attr_accessor *%i(
+      output_neuron
+      input_neuron
+      weight
+      processed
+      enabled
+      id
+    )
+
+    def initialize input_neuron, output_neuron, weight = nil
+      @id            = SecureRandom.hex
+      @output_neuron = output_neuron
+      @input_neuron  = input_neuron
+      @weight        = weight || initial_weight
+      @processed     = false
+      @enabled       = true
+      @locked        = false
+    end
+
+    def process
+      if processable? && !processed?
+        out_value = input_neuron.value.mult weight, 10
+        output_neuron.push_value! out_value
+        @processed = true
+      end
+    end
+
+    def neurons
+      [output_neuron, input_neuron]
+    end
+
+    def processable?
+      input_neuron.value
+    end
+
+    def enabled?
+      enabled
+    end
+
+    def processed?
+      processed
+    end
+
+    def locked?
+      @locked
+    end
+
+    def reset!
+      @processed = false
+    end
+
+  private
+    def initial_weight
+      if output_neuron.context?
+        1.to_d
+      else
+        rand.to_d 10
+      end
+    end
+  end
+end

data/lib/rann/gradient_checker.rb

@@ -0,0 +1,55 @@
+module RANN
+  class GradientChecker
+    EPSILON = 10.to_d.power -4, 10
+
+    def self.check network, inputs, targets, dvec
+      gradapprox = []
+
+      network.params.size.times do |i|
+        thetaplus = network.params.dup
+        thetaplus[i] = thetaplus[i] + EPSILON
+        thetaminus = network.params.dup
+        thetaminus[i] = thetaminus[i] - EPSILON
+
+        network.impose thetaplus
+        outputs = network.evaluate inputs
+        error_thetaplus = error outputs, targets
+        network.reset!
+
+        network.impose thetaminus
+        outputs = network.evaluate inputs
+        error_thetaminus = error outputs, targets
+        network.reset!
+
+        gradapprox[i] = (error_thetaplus - error_thetaminus).div(EPSILON.mult(2, 10), 10)
+      end
+
+      gradapprox.each.with_index.all?{ |ga, i| in_epsilon? ga, dvec[i] }
+    end
+
+    def self.error outputs, targets
+      total_squared_error = 0.to_d
+
+      targets.size.times do |i|
+        total_squared_error += (targets[i] - outputs[i]).power(2, 10).div(2, 10)
+      end
+
+      total_squared_error
+    end
+
+    def self.in_epsilon? exp, act, epsilon = 0.001
+      # delta = [exp.abs, act.abs].min * epsilon
+      delta = epsilon
+      n = (exp - act).abs
+      msg = "Expected |#{exp} - #{act}| (#{n}) to be <= #{delta}"
+
+      if delta >= n
+        true
+      else
+        puts msg
+
+        false
+      end
+    end
+  end
+end

data/lib/rann/locked_connection.rb

@@ -0,0 +1,10 @@
+require "rann/connection"
+
+module RANN
+  class LockedConnection < Connection
+    def initialize(*)
+      super
+      @locked = true
+    end
+  end
+end

data/lib/rann/lstm.rb

@@ -0,0 +1,86 @@
+require "rann/network"
+require "rann/neuron"
+require "rann/product_neuron"
+require "rann/connection"
+require "rann/locked_connection"
+
+module RANN
+  class LSTM
+    attr_reader :network, :inputs, :outputs, :name
+
+    def initialize name
+      @name    = name
+      @network = RANN::Network.new
+      @inputs  = []
+      @outputs = []
+    end
+
+    def init
+      @inputs.each.with_index do |input, i|
+        f = RANN::Neuron.new("LSTM #{name} F #{i}", 3, :standard, :sig).tap{ |n| @network.add n }
+        i = RANN::Neuron.new("LSTM #{name} I #{i}", 4, :standard, :sig).tap{ |n| @network.add n }
+        g = RANN::Neuron.new("LSTM #{name} G #{i}", 3, :standard, :tanh).tap{ |n| @network.add n }
+        o = RANN::Neuron.new("LSTM #{name} O #{i}", 3, :standard, :sig).tap{ |n| @network.add n }
+        bias_f = RANN::Neuron.new("LSTM #{name} Bias F #{i}", 0, :bias).tap do |n|
+          @network.add n
+          n.value = 1.to_d
+        end
+        bias_i = RANN::Neuron.new("LSTM #{name} Bias I #{i}", 0, :bias).tap do |n|
+          @network.add n
+          n.value = 1.to_d
+        end
+        bias_g = RANN::Neuron.new("LSTM #{name} Bias G #{i}", 0, :bias).tap do |n|
+          @network.add n
+          n.value = 1.to_d
+        end
+        bias_o = RANN::Neuron.new("LSTM #{name} Bias O #{i}", 0, :bias).tap do |n|
+          @network.add n
+          n.value = 1.to_d
+        end
+        memory_product = RANN::ProductNeuron.new("LSTM #{name} Mem Product #{i}", 2, :standard, :linear).tap{ |n| @network.add n }
+        i_g_product = RANN::ProductNeuron.new("LSTM #{name} Hidden 2/3 Product #{i}", 2, :standard, :linear).tap{ |n| @network.add n }
+        memory_standard = RANN::Neuron.new("LSTM #{name} Mem Standard #{i}", 2, :standard, :linear).tap{ |n| @network.add n }
+        memory_tanh = RANN::Neuron.new("LSTM #{name} Mem Tanh #{i}", 1, :standard, :tanh).tap{ |n| @network.add n }
+        memory_o_product = RANN::ProductNeuron.new("LSTM #{name} Mem/Hidden 4 Product #{i}", 2, :standard, :linear).tap{ |n| @network.add n }
+        output = RANN::Neuron.new("LSTM #{name} Output #{i}", 1, :standard, :linear).tap{ |n| @network.add n }
+        @outputs << output
+        memory_context = RANN::Neuron.new("LSTM #{name} Mem Context #{i}", 1, :context).tap{ |n| @network.add n }
+        output_context = RANN::Neuron.new("LSTM #{name} Output Context #{i}", 1, :context).tap{ |n| @network.add n }
+
+        @network.add RANN::LockedConnection.new input, f, 1
+        @network.add RANN::LockedConnection.new input, i, 1
+        @network.add RANN::LockedConnection.new input, g, 1
+        @network.add RANN::LockedConnection.new input, o, 1
+        @network.add RANN::LockedConnection.new f, memory_product, 1
+        @network.add RANN::LockedConnection.new i, i_g_product, 1
+        @network.add RANN::LockedConnection.new g, i_g_product, 1
+        @network.add RANN::LockedConnection.new i_g_product, memory_standard, 1
+        @network.add RANN::LockedConnection.new memory_product, memory_standard, 1
+        @network.add RANN::LockedConnection.new memory_standard, memory_tanh, 1
+        @network.add RANN::LockedConnection.new o, memory_o_product, 1
+        @network.add RANN::LockedConnection.new memory_tanh, memory_o_product, 1
+        @network.add RANN::LockedConnection.new memory_o_product, output, 1
+        @network.add RANN::LockedConnection.new memory_standard, memory_context, 1
+        @network.add RANN::Connection.new memory_context, memory_product
+        @network.add RANN::Connection.new memory_context, i
+        @network.add RANN::LockedConnection.new memory_o_product, output_context, 1
+        @network.add RANN::Connection.new output_context, f
+        @network.add RANN::Connection.new output_context, i
+        @network.add RANN::Connection.new output_context, g
+        @network.add RANN::Connection.new output_context, o
+        @network.add RANN::Connection.new bias_f, f
+        @network.add RANN::Connection.new bias_i, i
+        @network.add RANN::Connection.new bias_g, g
+        @network.add RANN::Connection.new bias_o, o
+      end
+    end
+
+    def add_input neuron
+      input = RANN::Neuron.new "LSTM #{name} Input #{neuron.name}", 0, :standard, :linear
+      @network.add input
+      @inputs << input
+      connection = RANN::Connection.new neuron, input
+      @network.add connection
+    end
+  end
+end

data/lib/rann/network.rb

@@ -0,0 +1,171 @@
+require "graphviz"
+require "yaml"
+
+module RANN
+  class Network
+    UnconnectedNetworkError = Class.new(StandardError)
+
+    attr_accessor *%i(
+      neurons
+      input_neurons
+      hidden_neurons
+      output_neurons
+      connections
+      structure
+    )
+
+    def initialize connections = []
+      @connections    = connections
+      @neurons        = connections.flat_map(&:neurons).uniq
+      @input_neurons  = @neurons.select &:input?
+      @output_neurons = @neurons.select &:output?
+      @hidden_neurons = @neurons - @input_neurons - @output_neurons
+    end
+
+    def impose weights
+      connections.each.with_index do |c, i|
+        c.weight = weights[i]
+      end
+    end
+
+    def params
+      connections.map(&:weight)
+    end
+
+    def evaluate input
+      input_neurons.each.with_index do |neuron, i|
+        neuron.value = input[i]
+      end
+
+      # use some proper graph traversal, rather than this crude blanketing?
+      # would probably be easier to detect circular dependency this way too?
+      begin
+        i = 0
+        until output_neurons.all?{ |neuron| neuron.value }
+          i += 1
+          connections.each do |connection|
+            next if !connection.enabled?
+
+            connection.process
+          end
+          raise UnconnectedNetworkError if i > 5_000
+        end
+      rescue UnconnectedNetworkError
+        visualise
+        raise
+      end
+
+      outputs
+    end
+
+    def visualise
+      # Create a new graph
+      g = GraphViz.new(:G, type: :digraph)
+
+      # Create nodes
+
+      missing_nodes = connections.each.with_object([]) do |c, o|
+        o << c.output_neuron unless neurons.include? c.output_neuron
+        o << c.input_neuron unless neurons.include? c.input_neuron
+      end
+
+      graph_nodes = neurons.each.with_object({}) do |n, h|
+        h[n] = g.add_nodes("#{n.name}: #{n.value&.to_f&.round(5)}")
+      end
+
+      # Create edges between the nodes
+      connections.each do |c|
+        g.add_edges(
+          graph_nodes[c.input_neuron],
+          graph_nodes[c.output_neuron],
+          color: c.processed? ? "#ff0000" : "#000000",
+          label: c.weight.to_f.round(5)
+        )
+      end
+
+      # Generate output image
+      g.output png: "nnet.png"
+      `open nnet.png`
+    end
+
+    def dump_weights
+      File.write "nn_weights_dump_#{DateTime.now.strftime('%Y-%m-%d-%H-%M-%S')}.yml", params.to_yaml
+    end
+
+    def outputs
+      output_neurons.map &:value
+    end
+
+    def state
+      neurons.each.with_object({}){ |n, s| s[n.id] = n.value }
+    end
+
+    def connections_to neuron
+      @connections_to = {} unless defined? @connections_to
+
+      @connections_to[neuron] ||= connections.select{ |con| con.output_neuron == neuron }
+    end
+
+    def connections_from neuron
+      @connections_from = {} unless defined? @connections_from
+
+      @connections_from[neuron] ||= connections.select{ |con| con.input_neuron == neuron }
+    end
+
+    def add *features
+      features.each do |feature|
+        case feature
+        when Neuron
+          case feature.type
+          when :input
+            @input_neurons  << feature
+          when :output
+            @output_neurons << feature
+          else
+            @hidden_neurons << feature
+          end
+
+          @neurons << feature
+        when Connection
+          @connections << feature
+        when Network
+          add *feature.neurons
+          add *feature.connections
+        end
+      end
+    end
+
+    def remove *features
+      features.each do |feature|
+        case feature
+        when Neuron
+          case feature.type
+          when :input
+            raise "trying to remove an input neuron ..."
+          when :output
+            raise "trying to remove an output neuron ..."
+          else
+            @hidden_neurons.delete feature
+          end
+
+          @neurons.delete feature
+        when Connection
+          @connections.delete feature
+        end
+      end
+    end
+
+    def reset!
+      state.tap do
+        neurons.each{ |neuron| neuron.reset! }
+        connections.each{ |connection| connection.reset! }
+      end
+    end
+
+    def recalculate_neuron_connection_counts!
+      neurons.each do |neuron|
+        neuron.connection_count = connections.count{ |c| c.output_neuron == neuron }
+      end
+    end
+  end
+end

data/lib/rann/neuron.rb

@@ -0,0 +1,83 @@
+require "securerandom"
+require "bigdecimal"
+require "bigdecimal/util"
+
+module RANN
+  class Neuron
+    ACTIVATION_FUNCTIONS = {
+      sig:    ->(v){ 1.to_d.div(1 + (Math::E ** -v), 10) },
+      tanh:   ->(v){ Math.tanh(v).to_d(10) },
+      relu:   ->(v){ [0.to_d, v].max },
+      linear: ->(v){ v },
+      step:   ->(v){ v > 0.5 ? 1.to_d : 0.to_d },
+    }
+
+    attr_accessor *%i(
+      activation_function
+      value
+      incoming
+      connection_count
+      type
+      name
+      id
+    )
+
+    def initialize name, connection_count, type = :standard, af = nil
+      @id                  = SecureRandom.hex
+      @connection_count    = connection_count
+      @type                = type
+      @incoming            = []
+      @activation_function = af || initial_activation_function
+      @name                = name
+
+      set_default_value!
+    end
+
+    def push_value! value
+      incoming << value
+      set_value! if incoming.size == connection_count
+    end
+
+    def set_value!
+      intermediate = incoming.reduce :+
+      self.value   = ACTIVATION_FUNCTIONS[activation_function].call intermediate
+    end
+
+    def reset!
+      set_default_value!
+      @incoming.clear
+    end
+
+    def increment_connection_count!
+      @connection_count += 1
+    end
+
+    def decrement_connection_count!
+      @connection_count -= 1
+    end
+
+    %i(input output context bias standard).each do |t|
+      define_method "#{t}?" do
+        type == t
+      end
+    end
+
+  private
+    def set_default_value!
+      self.value =
+        if context?
+          value || 0.to_d
+        elsif bias?
+          1.to_d
+        end
+    end
+
+    def initial_activation_function
+      if standard? || context?
+        :relu
+      else
+        :linear
+      end
+    end
+  end
+end

data/lib/rann/product_neuron.rb

@@ -0,0 +1,10 @@
+require "rann/neuron"
+
+module RANN
+  class ProductNeuron < Neuron
+    def set_value!
+      intermediate = incoming.reduce{ |i, m| m.mult i, 10 }
+      self.value   = ACTIVATION_FUNCTIONS[activation_function].call intermediate
+    end
+  end
+end

data/lib/rann/util/array_ext.rb

@@ -0,0 +1,56 @@
+module RANN
+  module Util
+    module ArrayExt
+      # Method `in_groups` from
+      # activesupport/lib/active_support/core_ext/array/grouping.rb under MIT
+      # licence. Original licence printed below.
+      #
+      # Copyright (c) 2005-2017 David Heinemeier Hansson
+
+      # 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.
+
+      def in_groups array, number, fill_with = nil
+        # size.div number gives minor group size;
+        # size % number gives how many objects need extra accommodation;
+        # each group hold either division or division + 1 items.
+        division = array.size.div number
+        modulo = array.size % number
+
+        # create a new array avoiding dup
+        groups = []
+        start = 0
+
+        number.times do |index|
+          length = division + (modulo > 0 && modulo > index ? 1 : 0)
+          groups << last_group = array.slice(start, length)
+          last_group << fill_with if fill_with != false &&
+            modulo > 0 && length == division
+          start += length
+        end
+
+        if block_given?
+          groups.each{ |g| yield(g) }
+        else
+          groups
+        end
+      end
+    end
+  end
+end

data/lib/rann/version.rb

@@ -0,0 +1,3 @@
+module RANN
+  VERSION = "0.1.0"
+end

data/lib/rann.rb

@@ -0,0 +1,9 @@
+require "bigdecimal"
+require "bigdecimal/util"
+require "rann/version"
+require "rann/network"
+require "rann/neuron"
+require "rann/product_neuron"
+require "rann/connection"
+require "rann/locked_connection"
+require "rann/backprop"

data/rann.gemspec

@@ -0,0 +1,29 @@
+
+lib = File.expand_path("../lib", __FILE__)
+$LOAD_PATH.unshift(lib) unless $LOAD_PATH.include?(lib)
+require "rann/version"
+
+Gem::Specification.new do |spec|
+  spec.name          = "rann"
+  spec.version       = RANN::VERSION
+  spec.authors       = ["Michael Campbell"]
+  spec.email         = ["mike@ydd.io"]
+
+  spec.summary       = %q{Ruby Artificial Neural Networks}
+  spec.description   = %q{Libary for working with neural networks in Ruby.}
+  spec.homepage      = "https://github.com/mikecmpbll/rann"
+
+  spec.files         = `git ls-files -z`.split("\x0").reject do |f|
+    f.match(%r{^(test|spec|features)/})
+  end
+  spec.bindir        = "exe"
+  spec.executables   = spec.files.grep(%r{^exe/}){ |f| File.basename(f) }
+  spec.require_paths = ["lib"]
+
+  spec.add_runtime_dependency "parallel", "~> 1.12.0"
+  spec.add_runtime_dependency "ruby-graphviz", "~> 1.2.3"
+
+  spec.add_development_dependency "bundler", "~> 1.16"
+  spec.add_development_dependency "rake", "~> 10.0"
+  spec.add_development_dependency "minitest", "~> 5.0"
+end

metadata

@@ -0,0 +1,136 @@
+--- !ruby/object:Gem::Specification
+name: rann
+version: !ruby/object:Gem::Version
+  version: 0.1.0
+platform: ruby
+authors:
+- Michael Campbell
+autorequire: 
+bindir: exe
+cert_chain: []
+date: 2017-11-08 00:00:00.000000000 Z
+dependencies:
+- !ruby/object:Gem::Dependency
+  name: parallel
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: 1.12.0
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: 1.12.0
+- !ruby/object:Gem::Dependency
+  name: ruby-graphviz
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: 1.2.3
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: 1.2.3
+- !ruby/object:Gem::Dependency
+  name: bundler
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '1.16'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '1.16'
+- !ruby/object:Gem::Dependency
+  name: rake
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '10.0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '10.0'
+- !ruby/object:Gem::Dependency
+  name: minitest
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '5.0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '5.0'
+description: Libary for working with neural networks in Ruby.
+email:
+- mike@ydd.io
+executables: []
+extensions: []
+extra_rdoc_files: []
+files:
+- ".gitignore"
+- ".travis.yml"
+- CHANGES.md
+- Gemfile
+- Gemfile.lock
+- LICENCE
+- README.md
+- Rakefile
+- bin/console
+- bin/setup
+- examples/xor.rb
+- lib/rann.rb
+- lib/rann/backprop.rb
+- lib/rann/connection.rb
+- lib/rann/gradient_checker.rb
+- lib/rann/locked_connection.rb
+- lib/rann/lstm.rb
+- lib/rann/network.rb
+- lib/rann/neuron.rb
+- lib/rann/product_neuron.rb
+- lib/rann/util/array_ext.rb
+- lib/rann/version.rb
+- rann.gemspec
+homepage: https://github.com/mikecmpbll/rann
+licenses: []
+metadata: {}
+post_install_message: 
+rdoc_options: []
+require_paths:
+- lib
+required_ruby_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: '0'
+required_rubygems_version: !ruby/object:Gem::Requirement
+  requirements:
+  - - ">="
+    - !ruby/object:Gem::Version
+      version: '0'
+requirements: []
+rubyforge_project: 
+rubygems_version: 2.5.1
+signing_key: 
+specification_version: 4
+summary: Ruby Artificial Neural Networks
+test_files: []