distribution 0.5.0 → 0.6.0

data/History.txt

@@ -1,3 +1,15 @@
+=== 0.6.0 / 2011-09-23
+* Incomplete Beta functions on math renamed to Regularized beta, because MathExtension::IncompleteBeta calculates regularized beta function, not Incomplete Beta. 
+* Corrected documention on F distribution and added comments on gamma and beta[Claudio Bustos]
+* Moved ported methods from GSL to lib/math_extension. Updated spec for gamma and beta distributions[Claudio Bustos]
+* Added beta distribution functions. p_value does not seem to work yet.[John Woods]
+* Added most incomplete beta function GSL translations, also log_beta from GSL.[John Woods]
+* Added header information to the incomplete gamma files translated from GSL[John Woods]
+* Removed left-over invgammp function.[John Woods]
+* Fixed lots of bugs, translated most GSL error tests into rspec.[John Woods]
+* Added Gamma distribution, spec. No statistics2 functions for Gamma, so only implemented pure Ruby and GSL.[John Woods]
+* Added console task to rakefile[John Woods]
+
 === 0.5.0 / 2011-05-03
 
 * Exception raises on calculation of T's cdf with ruby engine. For now, stick to gsl implementation
@@ -35,4 +47,4 @@
 
 === 0.1.0 / 2011-01-26
 
-* Basic set (pdf, cdf, p_value) for Normal, Chi Square, F and T distributions 
+* Basic set (pdf, cdf, p_value) for Normal, Chi Square, F and T distributions 

data/Manifest.txt

@@ -17,6 +17,10 @@ data/template/distribution/gsl.erb
 data/template/distribution/ruby.erb
 data/template/spec.erb
 lib/distribution.rb
+lib/distribution/beta.rb
+lib/distribution/beta/gsl.rb
+lib/distribution/beta/java.rb
+lib/distribution/beta/ruby.rb
 lib/distribution/binomial.rb
 lib/distribution/binomial/gsl.rb
 lib/distribution/binomial/java.rb
@@ -39,6 +43,10 @@ lib/distribution/f/gsl.rb
 lib/distribution/f/java.rb
 lib/distribution/f/ruby.rb
 lib/distribution/f/statistics2.rb
+lib/distribution/gamma.rb
+lib/distribution/gamma/gsl.rb
+lib/distribution/gamma/java.rb
+lib/distribution/gamma/ruby.rb
 lib/distribution/hypergeometric.rb
 lib/distribution/hypergeometric/gsl.rb
 lib/distribution/hypergeometric/java.rb
@@ -46,6 +54,14 @@ lib/distribution/hypergeometric/ruby.rb
 lib/distribution/logistic.rb
 lib/distribution/logistic/ruby.rb
 lib/distribution/math_extension.rb
+lib/distribution/math_extension/chebyshev_series.rb
+lib/distribution/math_extension/erfc.rb
+lib/distribution/math_extension/exponential_integral.rb
+lib/distribution/math_extension/gammastar.rb
+lib/distribution/math_extension/gsl_utilities.rb
+lib/distribution/math_extension/incomplete_beta.rb
+lib/distribution/math_extension/incomplete_gamma.rb
+lib/distribution/math_extension/log_utilities.rb
 lib/distribution/normal.rb
 lib/distribution/normal/gsl.rb
 lib/distribution/normal/java.rb
@@ -60,12 +76,14 @@ lib/distribution/t/gsl.rb
 lib/distribution/t/java.rb
 lib/distribution/t/ruby.rb
 lib/distribution/t/statistics2.rb
+spec/beta_spec.rb
 spec/binomial_spec.rb
 spec/bivariatenormal_spec.rb
 spec/chisquare_spec.rb
 spec/distribution_spec.rb
 spec/exponential_spec.rb
 spec/f_spec.rb
+spec/gamma_spec.rb
 spec/hypergeometric_spec.rb
 spec/logistic_spec.rb
 spec/math_extension_spec.rb

data/README.txt

@@ -4,7 +4,7 @@
 
 == DESCRIPTION:
 
-Statistical Distributions library. Includes Normal univariate and bivariate, T, F, Chi Square, Binomial, Hypergeometric, Exponential and Poisson.
+Statistical Distributions library. Includes Normal univariate and bivariate, T, F, Chi Square, Binomial, Hypergeometric, Exponential, Poisson, Beta and Gamma.
 
 Uses Ruby by default and C (statistics2/GSL) or Java extensions where available.
 
@@ -44,6 +44,8 @@ Shortnames for distributions:
 * Hypergeometric: hypg
 * Exponential: expo
 * Poisson: pois
+* Beta: beta
+* Gamma: gamma
 
 For example
 

data/Rakefile

@@ -17,5 +17,10 @@ Hoe.spec 'distribution' do
   self.extra_dev_deps << ["rspec",">=2.0"] << ["rubyforge",">=0"]
 
 end
+# git log --pretty=format:"*%s[%cn]" v0.5.0..HEAD >> History.txt
+desc "Open an irb session preloaded with distribution"
+task :console do
+  sh "irb -rubygems -I lib -r distribution.rb"
+end
 
 # vim: syntax=ruby

data/lib/distribution.rb

@@ -22,7 +22,8 @@
 # * Code of several Ruby engines came from statistics2.rb, 
 #   created by Shin-ichiro HARA(sinara@blade.nagaokaut.ac.jp).
 #   Retrieve from http://blade.nagaokaut.ac.jp/~sinara/ruby/math/statistics2/
-# 
+# * Code of Beta and Gamma distribution came from GSL project.
+#   Ported by John O. Woods
 # Specific notices will be placed where there are appropiate
 # 
 if !respond_to? :define_singleton_method
@@ -49,7 +50,7 @@ require 'distribution/math_extension'
 #    Distribution::Normal.p_value(0.95)
 #    => 1.64485364660836
 module Distribution
-  VERSION="0.5.0"
+  VERSION="0.6.0"
   
   module Shorthand
     EQUIVALENCES={:p_value=>:p, :cdf=>:cdf, :pdf=>:pdf, :rng=>:r, :exact_pdf=>:epdf, :exact_cdf=>:ecdf, :exact_p_value=>:ep}
@@ -133,13 +134,15 @@ module Distribution
             
     end
     # create alias for common methods
-    alias_method :inverse_cdf,:p_value if singleton_methods.include? :p_value
+    alias_method :inverse_cdf, :p_value if singleton_methods.include? :p_value
     end
 
   end
 
   autoload(:Normal, 'distribution/normal')  
   autoload(:ChiSquare, 'distribution/chisquare')
+  autoload(:Gamma, 'distribution/gamma')
+  autoload(:Beta, 'distribution/beta')
   autoload(:T, 'distribution/t')
   autoload(:F, 'distribution/f')
   autoload(:BivariateNormal, 'distribution/bivariatenormal')

data/lib/distribution/beta.rb

@@ -0,0 +1,34 @@
+require 'distribution/beta/ruby'
+require 'distribution/beta/gsl'
+# no statistics2 functions for beta.
+require 'distribution/beta/java'
+
+module Distribution
+  # From Wikipedia:
+  #    In probability theory and statistics, the beta distribution 
+  #    is a family of continuous probability distributions defined 
+  #    on the interval (0, 1) parameterized by two positive shape 
+  #    parameters, typically denoted by alpha and beta. 
+  # This module calculate cdf and inverse cdf for Beta Distribution.
+  #
+  module Beta
+    extend Distributable
+    SHORTHAND='beta'
+    create_distribution_methods
+
+    ##
+    # :singleton-method: pdf(x,a,b)
+    # Returns PDF of of Beta distribution with parameters a and b
+
+
+    ##
+    # :singleton-method: cdf(x,a,b)
+    # Returns the integral of Beta distribution with parameters a and b
+
+    ##
+    # :singleton-method: p_value(qn,a,b)
+    # Return the quantile of the corresponding integral +qn+
+    # on a beta distribution's cdf with parameters a and b 
+
+  end
+end

data/lib/distribution/beta/gsl.rb

@@ -0,0 +1,24 @@
+module Distribution
+  module Beta
+    module GSL_
+      class << self
+        def pdf(x,a,b)
+          GSL::Ran::beta_pdf(x.to_f, a.to_f, b.to_f)
+        end
+        # Return the P-value of the corresponding integral with
+        # k degrees of freedom
+        def p_value(pr,a,b)
+          GSL::Cdf::beta_Pinv(pr.to_f, a.to_f, b.to_f)
+        end
+        # Beta cumulative distribution function (cdf).
+        #
+        # Returns the integral of Beta distribution
+        # with parameters +a+ and +b+ over [0, x]
+        #
+        def cdf(x,a,b)
+          GSL::Cdf::beta_P(x.to_f, a.to_f, b.to_f)
+        end
+      end
+    end
+  end
+end

data/lib/distribution/beta/java.rb

@@ -0,0 +1,9 @@
+module Distribution
+  module Beta
+    # TODO
+    module Java_
+      class << self
+      end
+    end
+  end
+end

data/lib/distribution/beta/ruby.rb

@@ -0,0 +1,42 @@
+# Added by John O. Woods, SciRuby project.
+module Distribution
+  module Beta
+    module Ruby_
+      class << self
+
+        include Math
+        # Beta distribution probability density function
+        #
+        # Adapted from GSL-1.9 (apparently by Knuth originally), found in randist/beta.c
+        #
+        # Form: p(x) dx = (Gamma(a + b)/(Gamma(a) Gamma(b))) x^(a-1) (1-x)^(b-1) dx
+        #
+        # == References
+        # * http://www.gnu.org/s/gsl/manual/html_node/The-Gamma-Distribution.html
+        def pdf(x,a,b)
+          return 0 if x < 0 || x > 1
+
+          gab = Math.lgamma(a+b).first
+          ga  = Math.lgamma(a).first
+          gb  = Math.lgamma(b).first
+
+          if x == 0.0 || x == 1.0
+            Math.exp(gab - ga - gb) * x**(a-1) * (1-x)**(b-1)
+          else
+            Math.exp(gab - ga - gb + Math.log(x)*(a-1) + Math::Log.log1p(-x)*(b-1))
+          end
+        end
+
+        # Gamma cumulative distribution function
+        # Translated from GSL-1.9: cdf/beta.c gsl_cdf_beta_P
+        def cdf(x,a,b)
+          return 0.0 if x <= 0.0
+          return 1.0 if x >= 1.0
+          Math::IncompleteBeta.axpy(1.0, 0.0, a,b,x)
+        end
+
+
+      end
+    end
+  end
+end

data/lib/distribution/binomial.rb

@@ -3,7 +3,7 @@ require 'distribution/binomial/gsl'
 require 'distribution/binomial/java'
 module Distribution
 
-  # Calculate statisticals for T Distribution.
+  # Calculate statisticals for Binomial Distribution.
   module Binomial
     SHORTHAND = 'bino'  
     

data/lib/distribution/binomial/ruby.rb

@@ -6,9 +6,11 @@ module Distribution
           raise "k>n" if k>n
 	  Math.binomial_coefficient(n,k)*(pr**k)*(1-pr)**(n-k)          
         end
+        # TODO: Use exact_regularized_beta for 
+        # small values and regularized_beta for bigger ones.
         def cdf(k,n,pr)
           #(0..x.floor).inject(0) {|ac,i| ac+pdf(i,n,pr)}
-          Math.regularized_beta_function(1-pr,n - k,k+1)
+          Math.regularized_beta(1-pr,n - k,k+1)
         end
         def exact_cdf(k,n,pr)
           (0..k).inject(0) {|ac,i| ac+pdf(i,n,pr)}

data/lib/distribution/f.rb

@@ -3,7 +3,7 @@ require 'distribution/f/gsl'
 require 'distribution/f/statistics2'
 require 'distribution/f/java'
 module Distribution
-  # Calculate cdf and inverse cdf for Chi Square Distribution.
+  # Calculate cdf and inverse cdf for F Distribution.
   # 
   module F
     SHORTHAND='fdist'  

data/lib/distribution/gamma.rb

@@ -0,0 +1,37 @@
+require 'distribution/gamma/ruby'
+require 'distribution/gamma/gsl'
+# no statistics2 functions for gamma.
+require 'distribution/gamma/java'
+
+module Distribution
+  # From Wikipedia:
+  #    The gamma distribution is a two-parameter family of 
+  #    continuous probability distributions. It has a scale parameter a
+  #    and a shape parameter b.
+  # 
+  # Calculate pdf, cdf and inverse cdf for Gamma Distribution.
+  #
+  module Gamma
+    extend Distributable
+    SHORTHAND='gamma'
+    create_distribution_methods
+
+    ##
+    # :singleton-method: pdf(x,a,b)
+    # Returns PDF of of Gamma distribution with +a+ as scale
+    # parameter and +b+ as shape parameter
+
+
+    ##
+    # :singleton-method: cdf(x,a,b)
+    # Returns the integral of Gamma distribution with +a+ as scale
+    # parameter and +b+ as shape parameter
+
+    ##
+    # :singleton-method: p_value(qn,a,b)
+    # Return the upper limit for the integral of a 
+    # gamma distribution which returns +qn+ 
+    # with scale +a+ and shape +b+
+
+  end
+end

data/lib/distribution/gamma/gsl.rb

@@ -0,0 +1,24 @@
+module Distribution
+  module Gamma
+    module GSL_
+      class << self
+        def pdf(x,a,b)
+          GSL::Ran::gamma_pdf(x.to_f, a.to_f, b.to_f)
+        end
+        # Return the P-value of the corresponding integral with
+        # k degrees of freedom
+        def p_value(pr,a,b)
+          GSL::Cdf::gamma_Pinv(pr.to_f, a.to_f, b.to_f)
+        end
+        # Chi-square cumulative distribution function (cdf).
+        #
+        # Returns the integral of Chi-squared distribution
+        # with k degrees of freedom over [0, x]
+        #
+        def cdf(x,a,b)
+          GSL::Cdf::gamma_P(x.to_f, a.to_f, b.to_f)
+        end
+      end
+    end
+  end
+end

data/lib/distribution/gamma/java.rb

@@ -0,0 +1,9 @@
+module Distribution
+  module Gamma
+    # TODO
+    module Java_
+      class << self
+      end
+    end
+  end
+end

data/lib/distribution/gamma/ruby.rb

@@ -0,0 +1,53 @@
+# Added by John O. Woods, SciRuby project.
+module Distribution
+  module Gamma
+    module Ruby_
+      class << self
+
+        include Math
+        # Gamma distribution probability density function
+        #
+        # If you're looking at Wikipedia's Gamma distribution page, the arguments for this pdf function correspond
+        # as follows:
+        #
+        # * +x+: same
+        # * +a+: alpha or k
+        # + +b+: theta or 1/beta
+        #
+        # This is confusing! But we're trying to most closely mirror the GSL function for the gamma distribution
+        # (see references).
+        #
+        # Adapted the function itself from GSL-1.9 in rng/gamma.c: gsl_ran_gamma_pdf
+        #
+        # ==References
+        # * http://www.gnu.org/software/gsl/manual/html_node/The-Gamma-Distribution.html
+        # * http://en.wikipedia.org/wiki/Gamma_distribution
+        def pdf(x,a,b)
+          return 0 if x < 0
+          if x == 0
+            return 1.quo(b) if a == 1
+            return 0
+          elsif a == 1
+            Math.exp(-x.quo(b)).quo(b)
+          else
+            Math.exp((a-1)*Math.log(x.quo(b)) - x.quo(b) - Math.lgamma(a).first).quo(b)
+          end
+        end
+
+        # Gamma cumulative distribution function
+        def cdf(x,a,b)
+          return 0.0 if x <= 0.0
+
+          y = x.quo(b)
+          return (1-Math::IncompleteGamma.q(a, y)) if y > a
+          return (Math::IncompleteGamma.p(a, y))
+        end
+
+        #def p_value(pr,a,b)
+        #  cdf(1.0-pr,a,b)
+        #end
+
+      end
+    end
+  end
+end

data/lib/distribution/math_extension.rb

@@ -1,3 +1,15 @@
+# The next few requires eventually probably need to go in their own gem. They're all functions and constants used by
+# GSL-adapted pure Ruby math functions.
+require "distribution/math_extension/chebyshev_series"
+require "distribution/math_extension/erfc"
+require "distribution/math_extension/exponential_integral"
+require "distribution/math_extension/gammastar"
+require "distribution/math_extension/gsl_utilities"
+require "distribution/math_extension/incomplete_gamma"
+require "distribution/math_extension/incomplete_beta"
+require "distribution/math_extension/log_utilities"
+
+
 if RUBY_VERSION<"1.9"
   require 'mathn'
   def Prime.each(upper,&block) 
@@ -30,6 +42,8 @@ module Distribution
     B12 = -691.0 / 2730.0
     B14 =  7.0 / 6.0
     B16 = -3617.0 / 510.0
+
+
     # From statistics2
     def loggamma(x)
       v = 1.0
@@ -203,28 +217,44 @@ module Distribution
     def beta(x,y)
       (gamma(x)*gamma(y)).quo(gamma(x+y))
     end
+
+    # Get pure-Ruby logbeta
     def logbeta(x,y)
-      (loggamma(x)+loggamma(y))-loggamma(x+y)
+      Beta.log_beta(x,y).first
+    end
+
+    # Log beta function conforming to style of lgamma (returns sign in second array index)
+    def lbeta(x,y)
+      Beta.log_beta(x,y)
+    end
+
+    # I_x(a,b): Regularized incomplete beta function
+    # Fast version. For a exact calculation, based on factorial
+    # use exact_regularized_beta_function
+    def regularized_beta(x,a,b)
+      return 1 if x==1
+      IncompleteBeta.evaluate(a,b,x)
     end
     # I_x(a,b): Regularized incomplete beta function
     # TODO: Find a faster version.
     # Source:
     # * http://dlmf.nist.gov/8.17
-    def regularized_beta_function(x,a,b)
+    def exact_regularized_beta(x,a,b)
       return 1 if x==1
-      #incomplete_beta(x,a,b).quo(beta(a,b))
       m=a.to_i
       n=(b+a-1).to_i
       (m..n).inject(0) {|sum,j|
         sum+(binomial_coefficient(n,j)* x**j * (1-x)**(n-j))
       }
 
-    end
-    # B_x(a,b) : Incomplete beta function
-    # Should be replaced by
-    # http://lib.stat.cmu.edu/apstat/63
+     end
+    # 
+    # Incomplete beta function: B(x;a,b)
+    # +a+ and +b+ are parameters and +x+ is 
+    # integration upper limit.
     def incomplete_beta(x,a,b)
-      raise "Doesn't work"
+      IncompleteBeta.evaluate(a,b,x)*beta(a,b)
+      #Math::IncompleteBeta.axpy(1.0, 0.0, a,b,x)
     end
     
     # Rising factorial
@@ -234,10 +264,26 @@ module Distribution
    
     # Ln of gamma
     def loggamma(x)
-      lg=Math.lgamma(x)
-      lg[0]*lg[1]
+      Math.lgamma(x).first
+    end
+
+    def incomplete_gamma(a, x = 0, with_error = false)
+      IncompleteGamma.p(a,x, with_error)
+    end
+    alias :gammp :incomplete_gamma
+
+    def gammq(a, x, with_error = false)
+      IncompleteGamma.q(a,x,with_error)
+    end
+
+    def unnormalized_incomplete_gamma(a, x, with_error = false)
+      IncompleteGamma.unnormalized(a, x, with_error)
+    end
+
+    # Not the same as erfc. This is the GSL version, which may have slightly different results.
+    def erfc_e x, with_error = false
+      Erfc.evaluate(x, with_error)
     end
-    
     
     # Sequences without repetition. n^k'
     # Also called 'failing factorial'
@@ -306,13 +352,13 @@ end
 
 module Math
   include Distribution::MathExtension
-  module_function :factorial, :beta, :loggamma, :binomial_coefficient, :binomial_coefficient_gamma, :regularized_beta_function, :incomplete_beta, :permutations, :rising_factorial , :fast_factorial, :combinations, :logbeta
+  module_function :factorial, :beta, :loggamma, :erfc_e, :unnormalized_incomplete_gamma, :incomplete_gamma, :gammp, :gammq, :binomial_coefficient, :binomial_coefficient_gamma, :exact_regularized_beta, :incomplete_beta, :regularized_beta, :permutations, :rising_factorial , :fast_factorial, :combinations, :logbeta, :lbeta
 end
 
 # Necessary on Ruby 1.9
 module CMath # :nodoc:
   include Distribution::MathExtension
-  module_function :factorial, :beta, :loggamma,  :binomial_coefficient, :binomial_coefficient_gamma, :regularized_beta_function, :incomplete_beta, :permutations, :rising_factorial, :fast_factorial, :combinations, :logbeta
+  module_function :factorial, :beta, :loggamma, :unnormalized_incomplete_gamma, :incomplete_gamma, :gammp, :gammq, :erfc_e, :binomial_coefficient, :binomial_coefficient_gamma,  :incomplete_beta, :exact_regularized_beta, :regularized_beta, :permutations, :rising_factorial, :fast_factorial, :combinations, :logbeta, :lbeta
 end
 
 if RUBY_VERSION<"1.9"