differentiation 0.2.0 → 0.2.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ac6ecb62757c2673747b141af828086090e2cc42addb0ae5445503908f108f47
-  data.tar.gz: 39b075aca09020073e28b34aced45567b41a92bd9ddd755e2de60515ecd4299e
+  metadata.gz: 56dda61629d3925285ac54fafbf2ba9dea87171ad88062fced1c655bc40bb174
+  data.tar.gz: 6c67cbe07765683e1b2cda12d3e0217497035ef2a2fda50a3e62e981c37304ed
 SHA512:
-  metadata.gz: 6971cae699fbcdb58f86d36cb1e92908dec436d270d4f99d04c1480826674dbc016955510bf3172b96a924691c2fb9cdee2f9c895939742f239465ce24100fb3
-  data.tar.gz: e7d079380efccc25f93be4331f9e5ee18df5bb78b3f70cfc678ec43720439c77f76639bdff178627f8c99a0557c17a2e80a5a6078af8e6616c769c757c6a3db7
+  metadata.gz: e1fad5d9d8f66808baf9f5477fd24706c73cbf6a61d978775ec1d5e1f2bad514b85b4340536780ade19aeff940e92cc454903b9b0bfe0a35635dacb3aff8bc1c
+  data.tar.gz: 837a1274dbabb47e2ffcc203f05a673a815b716235f578ef21171e76fc8029427783991b4d3e619bfb98eb18a81576e30f559152d0da68cf7864c29ed3cb0b98

data/examples/xor_perceptron.rb

@@ -0,0 +1,57 @@
+require "matrix"
+require "differentiation"
+
+Dataset = {
+  inputs: [
+    Matrix[[0.0], [0.0]],
+    Matrix[[0.0], [1.0]],
+    Matrix[[1.0], [0.0]],
+    Matrix[[1.0], [1.0]],
+  ],
+  targets: [
+    0.0,
+    1.0,
+    1.0,
+    0.0
+  ]
+}
+
+def sigmoid(x, a=1.0)
+  1.0 / (1.0 + Math.exp(-a * x))
+end
+
+def mlp(input, w1, b1, w2, b2)
+  x = w1 * input + b1
+  x = x.map{|i| sigmoid(i) }
+  x = w2 * x + b2
+  sigmoid(x[0, 0])
+end
+
+differential def loss(w1, b1, w2, b2)
+  l = 0.0
+  Dataset[:inputs].zip(Dataset[:targets]) do |x, y|
+    l += (y - mlp(x, w1, b1, w2, b2)) ** 2
+  end
+  l
+end
+
+weights1 = Matrix.build(2, 2){ (rand() - 0.5) * 0.5 }
+bias1 = Matrix.build(2, 1) { 0.0 }
+weights2 = Matrix.build(1, 2){ (rand() - 0.5) * 0.5 }
+bias2 = Matrix.build(1, 1) { 0.0 }
+
+rate = 2.0
+
+10001.times do |step|
+  _loss = loss(weights1, bias1, weights2, bias2)
+  puts "%5d\t%.05f" % [step, _loss] if step % 1000 == 0
+  gw1, gb1, gw2, gb2 = _loss.gradients(:w1, :b1, :w2, :b2)
+  weights1 -= gw1.map{|v| v * rate }
+  bias1 -= gb1.map{|v| v * rate }
+  weights2 -= gw2.map{|v| v * rate }
+  bias2 -= gb2.map{|v| v * rate }
+end
+
+Dataset[:inputs].each do |x|
+  puts "#{[x[0,0], x[1, 0]]} => #{mlp(x, weights1, bias1, weights2, bias2)}"
+end

data/lib/differentiation/dual_number.rb

@@ -92,6 +92,14 @@ module Differentiation
       end
     end
 
+    def +@
+      self
+    end
+
+    def -@
+      self * -1
+    end
+
     def +(other)
       if other.is_a?(DualNumber)
         n = @n + other.n
@@ -134,7 +142,7 @@ module Differentiation
     def /(other)
       if other.is_a?(DualNumber)
         n = @n / other.n
-        diff = ->(var) { (self.derivative(var) / other) + (@n * other.derivative(var)) / (other.n ** 2) }
+        diff = ->(var) { (self.derivative(var) / other.n) - (@n * other.derivative(var)) / (other.n ** 2) }
         named_variables = @named_variables.merge(other.named_variables)
       else
         n = @n / other

data/lib/differentiation/ext/math.rb

@@ -0,0 +1,40 @@
+# frozen_string_literal: true
+
+module DualMath
+  def sin(x)
+    if x.is_a?(Differentiation::DualNumber)
+      Differentiation::DualNumber.new(super(x.n), ->(var){ cos(x.n) * x.derivative(var) }, named_variables: x.named_variables)
+    else
+      super(x)
+    end
+  end
+
+  def cos(x)
+    if x.is_a?(Differentiation::DualNumber)
+      Differentiation::DualNumber.new(super(x.n), ->(var){ -1.0 * sin(x.n) * x.derivative(var) }, named_variables: x.named_variables)
+    else
+      super(x)
+    end
+  end
+
+  def tan(x)
+    if x.is_a?(Differentiation::DualNumber)
+      Differentiation::DualNumber.new(super(x.n), ->(var){ ((1.0 / cos(x.n)) ** 2) * x.derivative(var) }, named_variables: x.named_variables)
+    else
+      super(x)
+    end
+  end
+
+  def exp(x)
+    if x.is_a?(Differentiation::DualNumber)
+      Differentiation::DualNumber.new(super(x.n), ->(var){ exp(x.n) * x.derivative(var) }, named_variables: x.named_variables)
+    else
+      super(x)
+    end
+  end
+end
+
+module Math
+  prepend DualMath
+end
+Math.singleton_class.prepend(DualMath)

data/lib/differentiation/version.rb

@@ -1,3 +1,3 @@
 module Differentiation
-  VERSION = "0.2.0"
+  VERSION = "0.2.1"
 end

data/lib/differentiation.rb

@@ -73,4 +73,4 @@ end
 require "differentiation/ext/kernel"
 require "differentiation/ext/integer"
 require "differentiation/ext/float"
-
+require "differentiation/ext/math"

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: differentiation
 version: !ruby/object:Gem::Version
-  version: 0.2.0
+  version: 0.2.1
 platform: ruby
 authors:
 - nagachika
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2019-04-12 00:00:00.000000000 Z
+date: 2019-04-13 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: bundler
@@ -68,11 +68,13 @@ files:
 - bin/console
 - bin/setup
 - differentiation.gemspec
+- examples/xor_perceptron.rb
 - lib/differentiation.rb
 - lib/differentiation/dual_number.rb
 - lib/differentiation/ext/float.rb
 - lib/differentiation/ext/integer.rb
 - lib/differentiation/ext/kernel.rb
+- lib/differentiation/ext/math.rb
 - lib/differentiation/version.rb
 homepage: https://github.com/nagachika/differentiation
 licenses: []
@@ -93,7 +95,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
 requirements: []
 rubyforge_project: 
-rubygems_version: 2.7.6
+rubygems_version: 2.7.6.2
 signing_key: 
 specification_version: 4
 summary: Make Ruby Differentiable.