pa_learner 0.1.1

data/Gemfile

@@ -0,0 +1,10 @@
+$LOAD_PATH << "."
+
+source "http://rubygems.org"
+
+# Specify your gem's dependencies in pa_learner.gemspec
+gemspec
+
+gem 'growl'
+gem 'guard-minitest'
+gem 'rb-fsevent'

data/README.md

@@ -0,0 +1,45 @@
+#The what?
+Online learning algorithm: Lightweight, that-works
+ 
+Implementation of a ``Passive-Aggressive'' online learning algorithm. The Passive-Aggressive online learning framework defines a family of online margin based linear learners. For further details see: 'Online Passive-Aggressive Algorithms' by Crammer et al. JMLR, 2006.
+- - - 
+#The how?
+
+##Usage
+>x is a vector (Array) of dim elements
+>y is its corresponding value (label -or- functional value)
+###Init
+>pal = PaLearner::DistRegressor.new( dim )
+###Update
+>pal.update!(x, y)
+###Estimate
+>pal.estimate( x ) 
+
+##Examples
+>d = [0.25, 0.75]
+>@data = Array.new(100)
+>(0..@data.size).each do |i|
+>  x_1 = rand
+>  x_2 = 1 - x_1
+>  @data[i] = {:y=>d[0]*x_1+d[1]*x_2 , :x=>[x_1, x_2]}
+>end
+>puts "data--size = #{@data.size}"
+>
+>pal = PaLearner::DistRegressor.new( 2 )
+>
+>@data.shuffle.inject(pal) { |pal, yx| pal.update!(yx[:x], yx[:y]); pal; }
+>@data.shuffle.inject(pal) { |pal, yx| pal.update!(yx[:x], yx[:y]); pal; }
+>
+>puts "pal.estimate([1,0]) = #{pal.estimate( [1, 0] )}"
+>puts "pal.estimate([0,1]) = #{pal.estimate( [0, 1] )}"
+
+#License
+##The MIT License
+``
+Copyright (c) 2010 ronbee.github@gmail.com
+
+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.''

data/Rakefile

@@ -0,0 +1,2 @@
+require 'bundler'
+Bundler::GemHelper.install_tasks

data/lib/pa_learner.rb

@@ -0,0 +1,97 @@
+module PaLearner
+  EPS_TO_AVOID_ZERO_DIV = 0.000000001
+  attr :w
+  
+  def estimate(x)
+    inner_product( x, @w)
+  end
+  
+  def bin_classify(x)
+    sign( estimate(x) )
+  end
+  
+  class PA
+    include PaLearner
+    
+    def initialize(n, type=:pa, cost = 0)
+      pa_kickstart( n )
+      @type = type
+      @cost = cost
+      self
+    end
+    
+    def update!(x, y)
+      l = hinge_loss( x, y )
+      etta = calc_update_delta( l, x)
+      @w.each_index { |i| @w[i] += y*etta*x[i] }
+      self
+    end
+    
+    private
+    def calc_update_delta(l,x)
+      delta = l / ( inner_product( x, x ) + EPS_TO_AVOID_ZERO_DIV )  if @type == :pa
+      delta = [ @cost, l / ( inner_product( x, x ) + EPS_TO_AVOID_ZERO_DIV ) ].min if @type == :pa_I
+      delta = l / ( inner_product( x, x ) + 1/(2+@cost) ) if @type == :pa_II
+      delta
+    end
+  end
+  
+  class DistRegressor
+    include PaLearner
+    def initialize( n, eps=0.001)
+      pa_kickstart( n )
+      @eps = eps
+    end
+
+    def update!(x,y)
+      rate = sign( estimate(x) - y ) * (   loss( estimate(x), y ) / ( inner_product(x, x) + EPS_TO_AVOID_ZERO_DIV )   ) 
+      rate_x = x.map {|v| -Math::log(v) * rate }
+      @w = @w.zip( rate_x ).map{ |v| v.first + v.last }
+      project_to_simplex!(@w)
+      self
+    end
+
+    #------
+    private
+
+    def loss(y_e, y)
+      [ 0, (y_e - y).abs - @eps ].max
+    end    
+
+    def project_to_simplex!( w )
+      working_ind = w.inject( {} ){|m,v| m[m.size]=v; m }.keys
+      last_size = working_ind.size + 1
+      while working_ind.size < last_size && working_ind.size > 0
+        last_size = working_ind.size
+        update = (working_ind.inject(0){|s,i| s+w[i]} -1) / working_ind.size.to_f
+        working_ind.reject! do |i|
+          w[i] -= update
+          w[i] = 0 if w[i] < 0
+          w[i] == 0
+        end
+      end
+      w
+    end
+  end
+  
+  #------
+  private
+
+  def pa_kickstart(n)
+    @w   = Array.new(n, 0)
+  end
+
+  def inner_product(a, b)
+    a.zip(b).map {|v| v.first*v.last}.inject(0){|sum,v| sum+v }
+  end
+
+  def hinge_loss(x,y)
+    [0, 1 - inner_product(@w, x) * y].max
+  end
+
+  def sign( v )
+    v >=0 ? 1 : -1
+  end
+
+end
+

data/lib/pa_learner/version.rb

@@ -0,0 +1,3 @@
+module PaLearner
+  VERSION = "0.1.1"
+end

data/pa_learner.gemspec

@@ -0,0 +1,21 @@
+# -*- encoding: utf-8 -*-
+$:.push File.expand_path("../lib", __FILE__)
+require "pa_learner/version"
+
+Gem::Specification.new do |s|
+  s.name        = "pa_learner"
+  s.version     = PaLearner::VERSION
+  s.platform    = Gem::Platform::RUBY
+  s.authors     = ["ronbee"]
+  s.email       = ["ronbee.github@gmail.com"]
+  s.homepage    = "https://github.com/ronbee"
+  s.summary     = %q{Implementing Online Passive-Aggressive Algorithms}
+  s.description = %q{Passive-Aggressive algorithms are a family of online margin based linear lerners. For further details see: 'Online Passive-Aggressive Algorithms' by Crammer et al. JMLR, 2006.}
+
+  s.rubyforge_project = "pa_learner"
+
+  s.files         = `git ls-files`.split("\n")
+  s.test_files    = `git ls-files -- {test,spec,features}/*`.split("\n")
+  s.executables   = `git ls-files -- bin/*`.split("\n").map{ |f| File.basename(f) }
+  s.require_paths = ["lib"]
+end

data/test/pa_learner_test.rb

@@ -0,0 +1,49 @@
+require 'minitest/unit'
+
+require File.dirname(__FILE__) + '/../lib/pa_learner'
+
+class MatchTest < MiniTest::Unit::TestCase
+
+  def setup
+    d = [0.25, 0.75] # ground truth -- we will use it for testing both distribution tracking, and classification
+    @data = Array.new(100)
+    (0..@data.size).each do |i|
+      x_1 = rand
+      x_2 = 1 - x_1
+      @data[i] = {:y=>d[0]*x_1+d[1]*x_2 , :x=>[x_1, x_2]}
+    end        
+  end
+
+  def test_sanity_distribution_tracking
+    pal = PaLearner::DistRegressor.new( 2 )
+
+    @data.shuffle.inject(pal) { |pal, yx| pal.update!(yx[:x], yx[:y]); pal; }
+    @data.shuffle.inject(pal) { |pal, yx| pal.update!(yx[:x], yx[:y]); pal; }
+
+    puts "pal.estimate([1,0]) = #{pal.estimate( [1, 0] )}"
+    puts "pal.estimate([0,1]) = #{pal.estimate( [0, 1] )}"
+
+    assert( ( (pal.estimate( [1, 0] )-0.25).abs - 0.01) < 0, "epsilon error larger than 0.01" )
+  end
+  
+  def test_sanity_classify_pa_0
+    pal = PaLearner::PA.new( 2 )
+    @data.shuffle.inject(pal) { |pal, yx| pal.update!(yx[:x], yx[:y]>=0 ? 1 : -1); pal; }
+    err_rate = ( @data.inject(0) {|errs,yx| errs += 1 unless pal.bin_classify(yx[:x]) == (yx[:y]>=0 ? 1 : -1); errs } ) / @data.size.to_f
+    assert( err_rate < 0.0001, "PA-0 error rate is too high: #{err_rate}" )
+  end
+  
+  def test_sanity_classify_pa_I
+    pal = PaLearner::PA.new( 2, :pa_I, 10 )
+    @data.shuffle.inject(pal) { |pal, yx| pal.update!(yx[:x], yx[:y]>=0 ? 1 : -1); pal; }
+    err_rate = ( @data.inject(0) {|errs,yx| errs += 1 unless pal.bin_classify(yx[:x]) == (yx[:y]>=0 ? 1 : -1); errs } ) / @data.size.to_f
+    assert( err_rate < 0.0001, "PA-I error rate is too high: #{err_rate}" )
+  end
+  
+  def test_sanity_classify_pa_II
+    pal = PaLearner::PA.new( 2, :pa_II, 10 )
+    @data.shuffle.inject(pal) { |pal, yx| pal.update!(yx[:x], yx[:y]>=0 ? 1 : -1); pal; }
+    err_rate = ( @data.inject(0) {|errs,yx| errs += 1 unless pal.bin_classify(yx[:x]) == (yx[:y]>=0 ? 1 : -1); errs } ) / @data.size.to_f
+    assert( err_rate < 0.0001, "PA-II error rate is too high: #{err_rate}" )
+  end
+end

metadata

@@ -0,0 +1,61 @@
+--- !ruby/object:Gem::Specification 
+name: pa_learner
+version: !ruby/object:Gem::Version 
+  prerelease: 
+  version: 0.1.1
+platform: ruby
+authors: 
+- ronbee
+autorequire: 
+bindir: bin
+cert_chain: []
+
+date: 2011-08-13 00:00:00 Z
+dependencies: []
+
+description: "Passive-Aggressive algorithms are a family of online margin based linear lerners. For further details see: 'Online Passive-Aggressive Algorithms' by Crammer et al. JMLR, 2006."
+email: 
+- ronbee.github@gmail.com
+executables: []
+
+extensions: []
+
+extra_rdoc_files: []
+
+files: 
+- Gemfile
+- README.md
+- Rakefile
+- lib/pa_learner.rb
+- lib/pa_learner/version.rb
+- pa_learner.gemspec
+- test/pa_learner_test.rb
+homepage: https://github.com/ronbee
+licenses: []
+
+post_install_message: 
+rdoc_options: []
+
+require_paths: 
+- lib
+required_ruby_version: !ruby/object:Gem::Requirement 
+  none: false
+  requirements: 
+  - - ">="
+    - !ruby/object:Gem::Version 
+      version: "0"
+required_rubygems_version: !ruby/object:Gem::Requirement 
+  none: false
+  requirements: 
+  - - ">="
+    - !ruby/object:Gem::Version 
+      version: "0"
+requirements: []
+
+rubyforge_project: pa_learner
+rubygems_version: 1.8.8
+signing_key: 
+specification_version: 3
+summary: Implementing Online Passive-Aggressive Algorithms
+test_files: 
+- test/pa_learner_test.rb