statistics2 0.54

data/lib/statistics2.rb

@@ -0,0 +1,617 @@
+$:.unshift File.dirname(__FILE__)
+$:.unshift File.join(File.dirname(__FILE__), '..', 'ext')
+
+require 'statistics2/version'
+
+#   distributions of statistics             
+#     by Shin-ichiro HARA
+#
+#   Ref:
+#     [1] http://www.matsusaka-u.ac.jp/~okumura/algo/
+#     [2] http://www5.airnet.ne.jp/tomy/cpro/sslib11.htm
+
+module Statistics2
+  SQ2PI = Math.sqrt(2 * Math::PI)
+
+  # Easy access to our singleton
+  def self.metaclass; class << self; self; end; end
+
+  # Remove existing methods.
+  ["normaldist", "normalxXX_", "normal__X_", "normal___x", "normalx__x",
+   "pnormaldist", "pnormalxXX_", "pnormal__X_", "pnormal___x", "pnormalx__x",
+   "chi2dist", "chi2X_", "chi2_x", "pchi2dist", "pchi2X_", "pchi2_x",
+   "tdist", "txXX_", "t__X_", "t___x", "tx__x", "ptdist", "ptxXX_", "pt__X_", "pt___x", "ptx__x",
+   "fdist", "fX_", "f_x", "pfdist", "pfX_", "pf_x",
+   "bindens", "bindist", "binX_", "bin_x",
+   "poissondens", "poissondist", "poissonX_", "poisson_x"].each do |m|
+    undef_method(m) if self.private_method_defined?(m)
+    self.metaclass.instance_eval do
+      undef_method(m) if self.method_defined?(m)
+    end
+  end
+  
+  # Newton approximation
+  def newton_a(y, ini, epsilon = 1.0e-6, limit = 30)
+    x = ini
+    limit.times do |i|
+      prev = x
+      f, df = yield(prev)
+      x = (y - f)/df + prev
+      if (x - prev).abs < epsilon
+        return x
+      end
+    end
+    $stderr.puts("Warning(newton approximation): over limit")
+    x
+  end
+
+  module_function :newton_a
+  private :newton_a
+  private_class_method :newton_a
+
+  # Gamma function
+  LOG_2PI = Math.log(2 * Math::PI)# log(2PI)
+  N = 8
+  B0  = 1.0
+  B1  = -1.0 / 2.0
+  B2  = 1.0 / 6.0
+  B4  = -1.0 / 30.0
+  B6  =  1.0 / 42.0
+  B8  = -1.0 / 30.0
+  B10 =  5.0 / 66.0
+  B12 = -691.0 / 2730.0
+  B14 =  7.0 / 6.0
+  B16 = -3617.0 / 510.0
+  
+  def loggamma(x)
+    v = 1.0
+    while (x < N)
+      v *= x
+      x += 1.0
+    end
+    w = 1.0 / (x * x)
+    ret = B16 / (16 * 15)
+    ret = ret * w + B14 / (14 * 13)
+    ret = ret * w + B12 / (12 * 11)
+    ret = ret * w + B10 / (10 *  9)
+    ret = ret * w + B8  / ( 8 *  7)
+    ret = ret * w + B6  / ( 6 *  5)
+    ret = ret * w + B4  / ( 4 *  3)
+    ret = ret * w + B2  / ( 2 *  1)
+    ret = ret / x + 0.5 * LOG_2PI - Math.log(v) - x + (x - 0.5) * Math.log(x)
+    ret
+  end
+  
+  def gamma(x)
+    if (x < 0.0)
+      return Math::PI / (Math.sin(Math.PI * x) * Math.exp(loggamma(1 - x))) #/
+    end
+    Math.exp(loggamma(x))
+  end
+
+  module_function :loggamma, :gamma
+  private :loggamma, :gamma
+  private_class_method :loggamma, :gamma
+
+  #normal-distribution
+  # (-\infty, z]
+  def p_nor(z)
+    if z < -12 then return 0.0 end
+    if z > 12 then return 1.0 end
+    if z == 0.0 then return 0.5 end
+
+    if z > 0.0
+      e = true
+    else
+      e = false
+      z = -z
+    end
+    z = z.to_f
+    z2 = z * z
+    t = q = z * Math.exp(-0.5 * z2) / SQ2PI
+
+    3.step(199, 2) do |i|
+      prev = q
+      t *= z2 / i
+      q += t
+      if q <= prev
+        return(e ? 0.5 + q : 0.5 - q)
+      end
+    end
+    e ? 1.0 : 0.0
+  end
+
+  # inverse of normal distribution ([2])
+  # Pr( (-\infty, x] ) = qn -> x
+  def pnorm(qn)
+    b = [1.570796288, 0.03706987906, -0.8364353589e-3,
+         -0.2250947176e-3, 0.6841218299e-5, 0.5824238515e-5,
+         -0.104527497e-5, 0.8360937017e-7, -0.3231081277e-8,
+         0.3657763036e-10, 0.6936233982e-12]
+    
+    if(qn < 0.0 || 1.0 < qn)
+      $stderr.printf("Error : qn <= 0 or qn >= 1  in pnorm()!\n")
+      return 0.0;
+    end
+    qn == 0.5 and return 0.0
+    
+    w1 = qn
+    qn > 0.5 and w1 = 1.0 - w1
+    w3 = -Math.log(4.0 * w1 * (1.0 - w1))
+    w1 = b[0]
+    1.upto 10 do |i|
+      w1 += b[i] * w3**i;
+    end
+    qn > 0.5 and return Math.sqrt(w1 * w3)
+    -Math.sqrt(w1 * w3)
+  end
+
+  private :p_nor, :pnorm
+  module_function :p_nor, :pnorm
+  private_class_method :p_nor, :pnorm
+
+  # Returns the integral of normal distribution over (-Infty, x].
+  def normaldist(z)
+    p_nor(z)
+  end
+
+  # Returns the P-value of normaldist(x).
+  def pnormaldist(y)
+    pnorm(y)
+  end
+
+  #chi-square distribution ([1])
+  #[x, \infty)
+  def q_chi2(df, chi2)
+    chi2 = chi2.to_f
+    if (df & 1) != 0
+      chi = Math.sqrt(chi2)
+      if (df == 1) then return 2 * normal___x(chi); end
+      s = t = chi * Math.exp(-0.5 * chi2) / SQ2PI
+      k = 3
+      while k < df
+        t *= chi2 / k;  s += t;
+        k += 2
+      end
+      2 * (normal___x(chi) + s)
+    else
+      s = t = Math.exp(-0.5 * chi2)
+      k = 2
+      while k < df
+        t *= chi2 / k;  s += t;
+        k += 2
+      end
+      s
+    end
+  end
+
+  def chi2dens(n, x)
+    if n == 1
+      1.0/Math.sqrt(2 * Math::PI * x) * Math::E**(-x/2.0)
+    elsif n == 2
+      0.5 * Math::E**(-x/2.0)
+    else
+      n = n.to_f
+      n2 = n/2
+      x = x.to_f
+      1.0 / 2**n2 / gamma(n2) * x**(n2 - 1.0) * Math.exp(-x/2.0)
+    end
+  end
+
+  # [x, \infty)
+  # Pr([x, \infty)) = y -> x
+  def pchi2(n, y)
+    if n == 1
+      w = pnorm(1 - y/2) # = pnormal___x(y/2)
+      w * w
+    elsif n == 2
+#      v = (1.0 / y - 1.0) / 33.0
+#      newton_a(y, v) {|x| [q_chi2(n, x), -chi2dens(n, x)] }
+      -2.0 * Math.log(y)
+    else
+      eps = 1.0e-5
+      v = 0.0
+      s = 10.0
+      loop do
+        v += s
+        if s <= eps then break end
+        if (qe = q_chi2(n, v) - y) == 0.0 then break end
+        if qe < 0.0
+          v -= s
+          s /= 10.0 #/
+        end
+      end
+      v
+    end
+  end
+
+  private :q_chi2, :pchi2, :chi2dens
+  module_function :q_chi2, :pchi2, :chi2dens
+  private_class_method :q_chi2, :pchi2, :chi2dens
+  
+  # Returns the integral of Chi-squared distribution with n degrees of freedom over [0, x].
+  def chi2dist(n, x); 1.0 - q_chi2(n, x); end
+  
+  # Returns the P-value of chi2dist().
+  def pchi2dist(n, y); pchi2(n, 1.0 - y); end
+
+
+  # t-distribution ([1])
+  # (-\infty, x]
+  def p_t(df, t)
+    c2 = df.to_f / (df + t * t);
+    s = Math.sqrt(1.0 - c2)
+    s = -s if t < 0.0
+    p = 0.0;
+    i = df % 2 + 2
+    while i <= df
+      p += s
+      s *= (i - 1) * c2 / i
+      i += 2
+    end
+    if df & 1 != 0
+      0.5+(p*Math.sqrt(c2)+Math.atan(t/Math.sqrt(df)))/Math::PI
+    else
+      (1.0 + p) / 2.0
+    end
+  end
+
+  # inverse of t-distribution ([2])
+  # (-\infty, -q/2] + [q/2, \infty)
+  def ptsub(q, n)
+    q = q.to_f
+    if(n == 1 && 0.001 < q && q < 0.01)
+      eps = 1.0e-4
+    elsif (n == 2 && q < 0.0001)
+      eps = 1.0e-4
+    elsif (n == 1 && q < 0.001)
+      eps = 1.0e-2
+    else
+      eps = 1.0e-5
+    end
+    s = 10000.0
+    w = 0.0
+    loop do
+      w += s
+      if(s <= eps) then return w end
+      if((qe = 2.0 - p_t(n, w)*2.0 - q) == 0.0) then return w end
+      if(qe < 0.0)
+        w -= s
+        s /= 10.0 #/
+      end
+    end
+  end
+
+  def pt(q, n)
+    q = q.to_f
+    if(q < 1.0e-5 || q > 1.0 || n < 1)
+      $stderr.printf("Error : Illigal parameter in pt()!\n")
+      return 0.0
+    end
+    
+    if(n <= 5) then return ptsub(q, n) end
+    if(q <= 5.0e-3 && n <= 13) then return ptsub(q, n) end
+
+    f1 = 4.0 * (f = n.to_f)
+    f5 = (f4 = (f3 = (f2 = f * f) * f) * f) * f
+    f2 *= 96.0
+    f3 *= 384.0
+    f4 *= 92160.0
+    f5 *= 368640.0
+    u = pnormaldist(1.0 - q / 2.0)
+
+    w0 = (u2 = u * u) * u
+    w1 = w0 * u2
+    w2 = w1 * u2
+    w3 = w2 * u2
+    w4 = w3 * u2
+    w = (w0 + u) / f1
+    w += (5.0 * w1 + 16.0 * w0 + 3.0 * u) / f2
+    w += (3.0 * w2 + 19.0 * w1 + 17.0 * w0 - 15.0 * u) / f3
+    w += (79.0 * w3 + 776.0 * w2 + 1482.0 * w1 - 1920.0 * w0 - 9450.0 * u) / f4
+    w += (27.0 * w4 + 339.0 * w3 + 930.0 * w2 - 1782.0 * w1 - 765.0 * w0 + 17955.0 * u) / f5
+    u + w
+  end
+  
+  private  :p_t, :pt, :ptsub
+  module_function :p_t, :pt, :ptsub
+  private_class_method :p_t, :pt, :ptsub
+
+  # Returns the integral of t-distribution with n degrees of freedom over (-Infty, x].
+  def tdist(n, t); p_t(n, t); end
+  
+  # Returns the P-value of tdist().
+  def ptdist(n, y)
+    if y > 0.5
+      pt(2.0 - y*2.0, n)
+    else
+      - pt(y*2.0, n)
+    end
+  end
+
+  # F-distribution ([1])
+  # [x, \infty) 
+  def q_f(df1, df2, f)
+    if (f <= 0.0) then return 1.0; end
+    if (df1 % 2 != 0 && df2 % 2 == 0)
+      return 1.0 - q_f(df2, df1, 1.0 / f)
+    end
+    cos2 = 1.0 / (1.0 + df1.to_f * f / df2.to_f)
+    sin2 = 1.0 - cos2
+
+    if (df1 % 2 == 0)
+        prob = cos2 ** (df2.to_f / 2.0)
+        temp = prob
+        i = 2
+        while i < df1
+          temp *= (df2.to_f + i - 2) * sin2 / i
+          prob += temp
+          i += 2
+        end
+        return prob
+    end
+    prob = Math.atan(Math.sqrt(df2.to_f / (df1.to_f * f)))
+    temp = Math.sqrt(sin2 * cos2)
+    i = 3
+    while i <= df1
+      prob += temp
+      temp *= (i - 1).to_f * sin2 / i.to_f;
+      i += 2.0
+    end
+    temp *= df1.to_f
+    i = 3
+    while i <= df2
+      prob -= temp
+      temp *= (df1.to_f + i - 2) * cos2 / i.to_f
+      i += 2
+    end
+    prob * 2.0 / Math::PI
+  end
+
+  # inverse of F-distribution ([2])
+  def pfsub(x, y, z)
+    (Math.sqrt(z) - y) / x / 2.0
+  end
+
+  # [x, \infty)
+  def pf(q, n1, n2)
+    if(q < 0.0 || q > 1.0 || n1 < 1 || n2 < 1)
+      $stderr.printf("Error : Illegal parameter in pf()!\n")
+      return 0.0
+    end
+    
+    if n1 <= 240 || n2 <= 240
+      eps = 1.0e-5
+      if(n2 == 1) then eps = 1.0e-4 end
+      fw = 0.0
+      s = 1000.0
+      loop do
+        fw += s
+        if s <= eps  then return fw end
+        if (qe = q_f(n1, n2, fw) - q) == 0.0 then return fw end
+        if qe < 0.0
+          fw -= s
+          s /= 10.0 #/
+        end
+      end
+    end
+
+    eps = 1.0e-6
+    qn = q
+    if q < 0.5 then qn = 1.0 - q
+      u = pnorm(qn)
+      w1 = 2.0 / n1 / 9.0
+      w2 = 2.0 / n2 / 9.0
+      w3 = 1.0 - w1
+      w4 = 1.0 - w2
+      u2 = u * u
+      a = w4 * w4 - u2 * w2
+      b = -2. * w3 * w4
+      c = w3 * w3 - u2 * w1
+      d = b * b - 4 * a * c
+      if(d < 0.0)
+        fw = pfsub(a, b, 0.0)
+      else
+        if(a.abs > eps)
+          fw = pfsub(a, b, d)
+        else
+          if(b.abs > eps) then return -c / b end
+          fw = pfsub(a, b, 0.0)
+        end
+      end
+      fw * fw * fw
+    end
+  end
+
+  private  :q_f, :pf, :pfsub
+  module_function :q_f, :pf, :pfsub
+  private_class_method :q_f, :pf, :pfsub
+
+  # Returns the integral of F-distribution with n1 and n2 degrees of freedom over [0, x].
+  def fdist(n1, n2, f); 1.0 - q_f(n1, n2, f); end
+
+  # Returns the P-value of fdist().
+  def pfdist(n1, n2, y); pf(1.0 - y, n1, n2); end
+
+  ############################################################################
+  # discrete distributions
+
+  def perm(n, x = n)
+    raise RangeError if n < 0 || x < 0
+    r = 1
+    while x >= 1
+      r *= n
+      n -= 1
+      x -= 1
+    end
+    r
+  end
+  
+  def combi(n, x)
+    raise RangeError if n < 0 || x < 0
+    x = n - x if x*2 > n
+    perm(n, x) / perm(x, x)
+  end
+
+  module_function :perm, :combi
+  private_class_method :perm, :combi
+
+  def bindens(n, p, x)
+    p = p.to_f
+    q = 1.0 - p
+    combi(n, x) * p**x * q**(n - x)
+  end
+  
+  def bindist(n, p, x)
+    (0..x).inject(0.0) do |s, k|
+      s + bindens(n, p, k)
+    end
+  end
+
+  def poissondens(m, x)
+    return 0.0 if x < 0
+    m = m.to_f
+    m ** x * Math::E ** (-m) / perm(x)
+  end
+
+  def poissondist(m, x)
+    (0..x).inject(0.0) do |s, k|
+      s + poissondens(m, k)
+    end
+  end
+
+  ############################################################################
+  # normal-distribution
+  
+  # Returns the integral of normal distribution over (-Infty, x].
+  def normalxXX_(z); normaldist(z); end
+  
+  # Returns the integral of normal distribution over [0, x].
+  def normal__X_(z); normaldist(z) - 0.5; end
+  
+  # Returns the integral of normal distribution over [x, Infty).
+  def normal___x(z); 1.0 - normaldist(z); end
+
+  # Returns the integral of normal distribution over (-Infty, -x] + [x, Infty).
+  def normalx__x(z); 2.0 - normaldist(z) * 2.0; end
+  
+  module_function :normaldist, :normalxXX_, :normal__X_, :normal___x, :normalx__x
+
+  
+  # inverse of normal-distribution
+
+  # Return the P-value of the corresponding integral.
+  def pnormalxXX_(z); pnormaldist(z); end
+  
+  # Return the P-value of the corresponding integral.
+  def pnormal__X_(y); pnormalxXX_(y + 0.5); end
+  
+  # Return the P-value of the corresponding integral.
+  def pnormal___x(y); pnormalxXX_(1.0 - y); end
+  
+  # Return the P-value of the corresponding integral.
+  def pnormalx__x(y); pnormalxXX_(1.0 - y/2.0); end
+  
+  module_function :pnormaldist, :pnormalxXX_, :pnormal__X_, :pnormal___x, :pnormalx__x
+
+
+  # chi2-distribution
+  
+  # Returns the integral of Chi-squared distribution with n degrees of freedom over [x, Infty).
+  def chi2_x(n, x); 1.0 - chi2dist(n, x); end
+  
+  # Returns the integral of Chi-squared distribution with n degrees of freedom over [0, x].
+  def chi2X_(n, x); chi2dist(n, x); end
+  
+  module_function :chi2dist, :chi2X_, :chi2_x
+
+  
+  # inverse of chi2-distribution
+
+  # Return the P-value of the corresponding integral.
+  def pchi2_x(n, y); pchi2dist(n, 1.0 - y); end
+  
+  # Return the P-value of the corresponding integral.
+  def pchi2X_(n, y); pchi2dist(n, y); end
+  
+  module_function :pchi2dist, :pchi2X_, :pchi2_x
+
+
+  # t-distribution
+
+  # Returns the integral of normal distribution with n degrees of freedom over (-Infty, -x] + [x, Infty).
+  def tx__x(n, x); 2.0 - tdist(n, x) * 2.0; end
+  
+  # Returns the integral of t-distribution with n degrees of freedom over (-Infty, x].
+  def txXX_(n, x); tdist(n, x); end
+  
+  # Returns the integral of t-distribution with n degrees of freedom over [0, x].
+  def t__X_(n, x); tdist(n, x) - 0.5; end
+
+  # Returns the integral of t-distribution with n degrees of freedom over [x, Infty).
+  def t___x(n, x); 1.0 - tdist(n, x); end
+  
+  module_function :tdist, :txXX_, :t__X_, :t___x, :tx__x
+
+  
+  # inverse of t-distribution
+
+  # Return the P-value of the corresponding integral.
+  def ptx__x(n, y); ptdist(n, 1.0 - y / 2.0); end
+
+  # Return the P-value of the corresponding integral.
+  def ptxXX_(n, y); ptdist(n, y); end
+
+  # Return the P-value of the corresponding integral.
+  def pt__X_(n, y); ptdist(n, 0.5 + y); end
+
+  # Return the P-value of the corresponding integral.
+  def pt___x(n, y); ptdist(n, 1.0 - y); end
+  
+  module_function :ptdist, :ptxXX_, :pt__X_, :pt___x, :ptx__x
+
+
+  # F-distribution
+
+  # Returns the integral of F-distribution with n1 and n2 degrees of freedom over [x, Infty).
+  def f_x(n1, n2, x); 1.0 - fdist(n1, n2, x); end
+
+  # Returns the integral of F-distribution with n1 and n2 degrees of freedom over [0, x].
+  def fX_(n1, n2, x); fdist(n1, n2, x); end
+  module_function :fdist, :fX_, :f_x
+
+  
+  # inverse of F-distribution
+
+  # Return the P-value of the corresponding integral.
+  def pf_x(n1, n2, x); pfdist(n1, n2, 1.0 - x); end
+
+  # Return the P-value of the corresponding integral.
+  def pfX_(n1, n2, x); pfdist(n1, n2, x); end
+  
+  module_function :pfdist, :pfX_, :pf_x
+
+  
+  # discrete distributions
+  def binX_(n, p, x); bindist(n, p, x); end
+  def bin_x(n, p, x); bindist(n, 1.0 - p, n - x);  end
+  module_function :bindens, :bindist, :binX_, :bin_x
+
+  def poissonX_(m, x); poissondist(m, x); end
+  def poisson_x(m, x); 1.0 - poissondist(m, x-1); end
+  module_function :poissondens, :poissondist, :poissonX_, :poisson_x
+end
+
+if !defined?(Statistics2::NO_EXT) || !Statistics2::NO_EXT
+  require 'statistics2.so'
+end
+
+if $0 == __FILE__
+  if ARGV.empty?
+    puts "Example:"
+    puts "        #$0 normaldist 0.01"
+    puts "        #$0 pf_x 2 3 0.01"
+    exit
+  end
+  p Statistics2.send(ARGV[0], *ARGV[1..-1].map{|x| eval(x)})
+end