1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
|
# Added by John O. Woods, SciRuby project.
# Derived from GSL-1.9 source files in the specfunc/ dir.
module Distribution
module MathExtension
module Beta
class << self
# Based on gsl_sf_lnbeta_e and gsl_sf_lnbeta_sgn_e
# Returns result and sign in an array. If with_error is specified, also returns the error.
def log_beta(x,y, with_error=false)
sign = nil
raise(ArgumentError, "x and y must be nonzero") if x == 0.0 || y == 0.0
raise(ArgumentError, "not defined for negative integers") if [x,y].any? { |v| (v.is_a?(Fixnum) && v < 0) }
# See if we can handle the positive case with min/max < 0.2
if x > 0 && y > 0
min, max = [x,y].minmax
ratio = min.quo(max)
if ratio < 0.2
gsx = Gammastar.evaluate(x, with_error)
gsy = Gammastar.evaluate(y, with_error)
gsxy = Gammastar.evaluate(x+y, with_error)
lnopr = Log::log_1plusx(ratio, with_error)
gsx, gsx_err, gsy, gsy_err, gsxy, gsxy_err, lnopr, lnopr_err = [gsx,gsy,gsxy,lnopr].flatten if with_error
lnpre = Math.log((gsx*gsy).quo(gsxy) * Math::SQRT2 * Math::SQRTPI)
lnpre_err = gsx_err.quo(gsx) + gsy_err(gsy) + gsxy_err.quo(gsxy) if with_error
t1 = min * Math.log(ratio)
t2 = 0.5 * Math.log(min)
t3 = (x+y-0.5)*lnopr
lnpow = t1 - t2 - t3
lnpow_err = Float::EPSILON * (t1.abs + t2.abs + t3.abs) + (x+y-0.5).abs * lnopr_err if with_error
result = lnpre + lnpow
error = lnpre_err + lnpow_err + 2.0*Float::EPSILON*result.abs if with_error
return with_error ? [result, 1.0, error] : [result, 1.0]
end
end
# General case: fallback
lgx, sgx = Math.lgamma(x)
lgy, sgy = Math.lgamma(y)
lgxy, sgxy = Math.lgamma(x+y)
sgn = sgx * sgy * sgxy
raise("Domain error: sign is -") if sgn == -1
result = lgx + lgy - lgxy
if with_error
lgx_err, lgy_err, lgxy_err = begin
STDERR.puts("Warning: Error is unknown for Math::lgamma, guessing.")
[Math::EPSILON, Math::EPSILON, Math::EPSILON]
end
error = lgx_err + lgy_err + lgxy_err + Float::EPSILON*(lgx.abs+lgy.abs+lgxy.abs) + 2.0*(Float::EPSILON)*result.abs
return [result, sgn, error]
else
return [result, sgn]
end
end
end
end
# Calculate regularized incomplete beta function
module IncompleteBeta
MAX_ITER = 512
CUTOFF = 2.0 * Float::MIN
class << self
# Evaluate aa * beta_inc(a,b,x) + yy
#
# No error mode available.
#
# From GSL-1.9: cdf/beta_inc.c, beta_inc_AXPY
def axpy(aa,yy,a,b,x)
return aa*0 + yy if x == 0.0
return aa*1 + yy if x == 1.0
ln_beta = Math.logbeta(a, b)
ln_pre = -ln_beta + a * Math.log(x) + b * Math::Log.log1p(-x)
prefactor = Math.exp(ln_pre)
if x < (a+1).quo(a+b+2)
# Apply continued fraction directly
epsabs = yy.quo((aa * prefactor).quo(a)).abs * Float::EPSILON
cf = continued_fraction(a, b, x, epsabs)
return aa * (prefactor * cf).quo(a) + yy
else
# Apply continued fraction after hypergeometric transformation
epsabs = (aa + yy).quo( (aa*prefactor).quo(b) ) * Float::EPSILON
cf = continued_fraction(b, a, 1-x, epsabs)
term = (prefactor * cf).quo(b)
return aa == -yy ? -aa*term : aa*(1-term)+yy
end
end
# Evaluate the incomplete beta function
# gsl_sf_beta_inc_e
def evaluate(a,b,x,with_error=false)
raise(ArgumentError, "Domain error: a(#{a}), b(#{b}) must be positive; x(#{x}) must be between 0 and 1, inclusive") if a <= 0 || b <= 0 || x < 0 || x > 1
if x == 0
return with_error ? [0.0,0.0] : 0.0
elsif x == 1
return with_error ? [1.0,0.0] : 1.0
else
ln_beta = Beta.log_beta(a,b, with_error)
ln_1mx = Log.log_1plusx(-x, with_error)
ln_x = Math.log(x)
ln_beta, ln_beta_err, ln_1mx, ln_1mx_err, ln_x_err = begin
#STDERR.puts("Warning: Error is unknown for Math::log, guessing.")
[ln_beta,ln_1mx,Float::EPSILON].flatten
end
ln_pre = -ln_beta + a*ln_x + b*ln_1mx
ln_pre_err = ln_beta_err + (a*ln_x_err).abs + (b*ln_1mx_err).abs if with_error
prefactor, prefactor_err = begin
if with_error
exp_err(ln_pre, ln_pre_err)
else
[Math.exp(ln_pre), nil]
end
end
if x < (a+1).quo(a+b+2)
# Apply continued fraction directly
cf = continued_fraction(a,b,x, nil, with_error)
cf,cf_err = cf if with_error
result = (prefactor * cf).quo(a)
return with_error ? [result, ((prefactor_err*cf).abs + (prefactor*cf_err).abs).quo(a)] : result
else
# Apply continued fraction after hypergeometric transformation
cf = continued_fraction(b, a, 1-x, nil)
cf,cf_err = cf if with_error
term = (prefactor * cf).quo(b)
result = 1 - term
return with_error ? [result, (prefactor_err*cf).quo(b) + (prefactor*cf_err).quo(b) + 2.0*Float::EPSILON*(1+term.abs)] : result
end
end
end
def continued_fraction_cutoff(epsabs)
return CUTOFF if epsabs.nil?
0.0/0 # NaN
end
# Continued fraction calculation of incomplete beta
# beta_cont_frac from GSL-1.9
#
# If epsabs is set, will execute the version of the GSL function in the cdf folder. Otherwise, does the
# basic one in specfunc.
def continued_fraction(a,b,x,epsabs=nil,with_error=false)
num_term = 1
den_term = 1 - (a+b)*x.quo(a+1)
k = 0
den_term = continued_fraction_cutoff(epsabs) if den_term.abs < CUTOFF
den_term = 1.quo(den_term)
cf = den_term
1.upto(MAX_ITER) do |k|
coeff = k *(b-k)*x.quo(((a-1)+2*k)*(a+2*k)) # coefficient for step 1
delta_frac = nil
2.times do
den_term = 1 + coeff*den_term
num_term = 1 + coeff.quo(num_term)
den_term = continued_fraction_cutoff(epsabs) if den_term.abs < CUTOFF
num_term = continued_fraction_cutoff(epsabs) if num_term.abs < CUTOFF
den_term = 1.quo(den_term)
delta_frac = den_term * num_term
cf *= delta_frac
coeff = -(a+k)*(a+b+k)*x.quo((a+2*k)*(a+2*k+1)) # coefficient for step 2
end
break if (delta_frac-1).abs < 2.0*Float::EPSILON
break if !epsabs.nil? && (cf * (delta_frac-1).abs < epsabs)
end
if k > MAX_ITER
raise("Exceeded maximum number of iterations") if epsabs.nil?
return with_error ? [0.0/0, 0] : 0.0/0 # NaN if epsabs is set
end
with_error ? [cf, k * 4 * Float::EPSILON * cf.abs] : cf
end
end
end
end
end
|
