bp_net 0.0.2

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: 4ae78900480ba50a21932b089dfb61c8008e4094255da4c76f1b8897ee3a9b8b
+  data.tar.gz: ea012ac44327f5e91df5da58c131c0df5a637aa27a613200ae21b1a2f2ed92f7
+SHA512:
+  metadata.gz: 9fe9e7a877f4dbc80c403bcb072bf9051dc9c77a7ad182ce5a4f737beb3d0e082e6fa7b40af6bd4ce954c919d17d0b11e7dd9b49ccac0b1ec01b0e37910a5fbe
+  data.tar.gz: e4b2b7e6a1269ea7659e0de5bb8ade20e247172490a74432e6aa3a0a219ba3c0ba1f037fa4b5d15ef2b1d40d92cc5b23e4047037f7a0286499de2476874b7f8f

data/lib/bp_net.rb

@@ -0,0 +1,108 @@
+require 'matrix'
+
+class Matrix
+  def disp(line_feed: 1)
+    row_vectors.each do |row| # self.row_vectors...
+      row.each do |ele| # print elements by row
+        print "#{ele}\t"
+      end
+      line_feed.times { puts } # Line Feed
+    end
+    self
+  end
+end
+
+class BpNet
+  # accessable attributes
+  ACCESSABLE_ATTRS = %i[inodes hnodes onodes lr wih who]
+  attr_accessor(*ACCESSABLE_ATTRS)
+
+  # initialization
+  def initialize(inputnodes:, hiddennodes:, outputnodes:, learningrate:)
+    @inodes = inputnodes  # input nodes
+    @hnodes = hiddennodes # hidden nodes
+    @onodes = outputnodes # output nodes
+    @lr = learningrate    # learning rate
+
+    # weight matrixes
+    @wih = Matrix.build(hiddennodes, inputnodes) { rand - 0.5 }
+    @who = Matrix.build(outputnodes, hiddennodes) { rand - 0.5 }
+  end
+
+  # train
+  def train(inputs_mat, targets_mat)
+    inputs_mat = arr2mat(inputs_mat)    # convert array to matrix
+    targets_mat = arr2mat(targets_mat)  # convert arr 2 mat
+
+    hidden_inputs = @wih * inputs_mat                   # Hi = Wih * I
+    hidden_outputs = BpNet.sigmoid(hidden_inputs)    # H = sigmoid(Hi)
+    outputs_inputs = @who * hidden_outputs              # Oi = Who * H
+    outputs_outputs = BpNet.sigmoid(outputs_inputs)  # O = sigmoid(Oi)
+
+    output_errors = targets_mat - outputs_outputs
+    @who += update(output_errors, outputs_inputs, hidden_outputs)
+
+    hidden_errors = @who.t * output_errors
+    @wih += update(hidden_errors, hidden_inputs, inputs_mat)
+  end
+
+  # query
+  # Hi = Wih * I
+  # H = sigmoid(Hi)
+  # Oi = Who * H
+  # O = sigmoid(Oi)
+  def query(inputs_mat)
+    inputs_mat = arr2mat(inputs_mat) # convert array to matrix
+
+    hidden_inputs = @wih * inputs_mat                   # Hi = Wih * I
+    hidden_outputs = BpNet.sigmoid(hidden_inputs)    # H = sigmoid(Hi)
+    outputs_inputs = @who * hidden_outputs              # Oi = Who * H
+    outputs_outputs = BpNet.sigmoid(outputs_inputs)  # O = sigmoid(Oi)
+  end
+
+  def arr2mat(arr)
+    if arr.is_a? Array
+      Matrix[arr].t
+    else
+      arr
+    end
+  end
+
+  ## calculations goes here
+  # singular sigmoid
+  @singular_sigmoid = ->(x) { 1 / (1 + Math::E**-x) }
+
+  # singular sigmoid differentiation
+  @singular_sigmoid_diff = lambda do |x|
+    @singular_sigmoid.call(x) * (1 - @singular_sigmoid.call(x))
+  end
+
+  def update(errors, next_inputs, prev_outputs)
+    @lr *
+      errors.hadamard_product(BpNet.sigmoid_diff(next_inputs)) *
+      prev_outputs.t
+  end
+
+  ## packed enumerable calculations
+  # sigmoid
+  def self.sigmoid(x)
+    enum_calc(x, @singular_sigmoid) { |param| sigmoid(param) }
+  end
+
+  # sigmoid differentiation
+  def self.sigmoid_diff(x)
+    enum_calc(x, @singular_sigmoid_diff) { |param| sigmoid_diff(param) }
+  end
+
+  ## enumerable calculation
+  def self.enum_calc(x, calc, &block)
+    if x.is_a? Enumerable
+      x.map(&block) # return values recursively
+    else
+      calc.call(x) # return value
+    end
+  end
+
+  # private class methods
+  private_class_method :enum_calc
+end

metadata

@@ -0,0 +1,43 @@
+--- !ruby/object:Gem::Specification
+name: bp_net
+version: !ruby/object:Gem::Version
+  version: 0.0.2
+platform: ruby
+authors:
+- Rurika Misaka
+autorequire:
+bindir: bin
+cert_chain: []
+date: 2022-05-08 00:00:00.000000000 Z
+dependencies: []
+description: A sipmle BP Neural Network gem
+email: goblinmaiden1931@gmail.com
+executables: []
+extensions: []
+extra_rdoc_files: []
+files:
+- lib/bp_net.rb
+homepage:
+licenses:
+- MIT
+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: []
+rubygems_version: 3.2.3
+signing_key:
+specification_version: 4
+summary: Simple BP Neural Network
+test_files: []