Skip to content

Commit 6220584

Browse files
devmotiondevmotion
committed
Revert "Use julia implementations for pdfs and some cdf-like functions (#113)"
This reverts commit 28d732a.
1 parent 7185ee5 commit 6220584

File tree

9 files changed

+90
-278
lines changed

9 files changed

+90
-278
lines changed

‎Project.toml

-2
Original file line numberDiff line numberDiff line change
@@ -4,7 +4,6 @@ version = "0.9.16"
44

55
[deps]
66
ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"
7-
HypergeometricFunctions = "34004b35-14d8-5ef3-9330-4cdb6864b03a"
87
InverseFunctions = "3587e190-3f89-42d0-90ee-14403ec27112"
98
IrrationalConstants = "92d709cd-6900-40b7-9082-c6be49f344b6"
109
LogExpFunctions = "2ab3a3ac-af41-5b50-aa03-7779005ae688"
@@ -14,7 +13,6 @@ SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
1413

1514
[compat]
1615
ChainRulesCore = "1"
17-
HypergeometricFunctions = "0.3"
1816
InverseFunctions = "0.1"
1917
IrrationalConstants = "0.1"
2018
LogExpFunctions = "0.3.2"

‎src/distrs/beta.jl

+7-67
Original file line numberDiff line numberDiff line change
@@ -1,17 +1,14 @@
11
# functions related to beta distributions
22

3-
using HypergeometricFunctions: _₂F₁
4-
53
# R implementations
4+
# For pdf and logpdf we use the Julia implementation
65
using .RFunctions:
7-
# betapdf,
8-
# betalogpdf,
9-
# betacdf,
10-
# betaccdf,
11-
# betalogcdf,
12-
# betalogccdf,
13-
# betainvcdf,
14-
# betainvccdf,
6+
betacdf,
7+
betaccdf,
8+
betalogcdf,
9+
betalogccdf,
10+
betainvcdf,
11+
betainvccdf,
1512
betainvlogcdf,
1613
betainvlogccdf
1714

@@ -25,60 +22,3 @@ function betalogpdf(α::T, β::T, x::T) where {T<:Real}
2522
val = xlogy(α - 1, y) + xlog1py(β - 1, -y) - logbeta(α, β)
2623
return x < 0 || x > 1 ? oftype(val, -Inf) : val
2724
end
28-
29-
function betacdf(α::Real, β::Real, x::Real)
30-
# Handle a degenerate case
31-
if iszero(α) && β > 0
32-
return float(last(promote(α, β, x, x >= 0)))
33-
end
34-
35-
return first(beta_inc(α, β, clamp(x, 0, 1)))
36-
end
37-
38-
function betaccdf(α::Real, β::Real, x::Real)
39-
# Handle a degenerate case
40-
if iszero(α) && β > 0
41-
return float(last(promote(α, β, x, x < 0)))
42-
end
43-
44-
last(beta_inc(α, β, clamp(x, 0, 1)))
45-
end
46-
47-
# The log version is currently based on non-log version. When the cdf is very small we shift
48-
# to an implementation based on the hypergeometric function ₂F₁ to avoid underflow.
49-
function betalogcdf(α::T, β::T, x::T) where {T<:Real}
50-
# Handle a degenerate case
51-
if iszero(α) && β > 0
52-
return log(last(promote(x, x >= 0)))
53-
end
54-
55-
_x = clamp(x, 0, 1)
56-
p, q = beta_inc(α, β, _x)
57-
if p < floatmin(p)
58-
# see https://dlmf.nist.gov/8.17#E7
59-
return -log(α) + xlogy(α, _x) + log(_₂F₁(promote(α, 1 - β, α + 1, _x)...)) - logbeta(α, β)
60-
elseif p <= 0.7
61-
return log(p)
62-
else
63-
return log1p(-q)
64-
end
65-
end
66-
betalogcdf(α::Real, β::Real, x::Real) = betalogcdf(promote(α, β, x)...)
67-
68-
function betalogccdf(α::Real, β::Real, x::Real)
69-
# Handle a degenerate case
70-
if iszero(α) && β > 0
71-
return log(last(promote(α, β, x, x < 0)))
72-
end
73-
74-
p, q = beta_inc(α, β, clamp(x, 0, 1))
75-
if q < 0.7
76-
return log(q)
77-
else
78-
return log1p(-p)
79-
end
80-
end
81-
82-
betainvcdf(α::Real, β::Real, p::Real) = first(beta_inc_inv(α, β, p))
83-
84-
betainvccdf(α::Real, β::Real, p::Real) = last(beta_inc_inv(β, α, p))

‎src/distrs/binom.jl

+6-26
Original file line numberDiff line numberDiff line change
@@ -1,18 +1,18 @@
11
# functions related to binomial distribution
22

33
# R implementations
4+
# For pdf and logpdf we use the Julia implementation
45
using .RFunctions:
5-
# binompdf,
6-
# binomlogpdf,
7-
# binomcdf,
8-
# binomccdf,
9-
# binomlogcdf,
10-
# binomlogccdf,
6+
binomcdf,
7+
binomccdf,
8+
binomlogcdf,
9+
binomlogccdf,
1110
binominvcdf,
1211
binominvccdf,
1312
binominvlogcdf,
1413
binominvlogccdf
1514

15+
1616
# Julia implementations
1717
binompdf(n::Real, p::Real, k::Real) = exp(binomlogpdf(n, p, k))
1818

@@ -22,23 +22,3 @@ function binomlogpdf(n::T, p::T, k::T) where {T<:Real}
2222
val = min(0, betalogpdf(m + 1, n - m + 1, p) - log(n + 1))
2323
return 0 <= k <= n && isinteger(k) ? val : oftype(val, -Inf)
2424
end
25-
26-
for l in ("", "log"), compl in (false, true)
27-
fbinom = Symbol(string("binom", l, ifelse(compl, "c", "" ), "cdf"))
28-
fbeta = Symbol(string("beta" , l, ifelse(compl, "", "c"), "cdf"))
29-
@eval function ($fbinom)(n::Real, p::Real, k::Real)
30-
if isnan(k)
31-
return last(promote(n, p, k))
32-
end
33-
res = ($fbeta)(max(0, floor(k) + 1), max(0, n - floor(k)), p)
34-
35-
# When p == 1 the betaccdf doesn't return the correct result
36-
# so these cases have to be special cased
37-
if isone(p)
38-
newres = oftype(res, $compl ? k < n : k >= n)
39-
return $(l === "" ? :newres : :(log(newres)))
40-
else
41-
return res
42-
end
43-
end
44-
end

‎src/distrs/chisq.jl

+20-9
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,22 @@
11
# functions related to chi-square distribution
22

3-
# Just use the Gamma definitions
4-
for f in ("pdf", "logpdf", "cdf", "ccdf", "logcdf", "logccdf", "invcdf", "invccdf", "invlogcdf", "invlogccdf")
5-
_chisqf = Symbol("chisq"*f)
6-
_gammaf = Symbol("gamma"*f)
7-
@eval begin
8-
$(_chisqf)(k::Real, x::Real) = $(_chisqf)(promote(k, x)...)
9-
$(_chisqf)(k::T, x::T) where {T<:Real} = $(_gammaf)(k/2, 2, x)
10-
end
11-
end
3+
# R implementations
4+
# For pdf and logpdf we use the Julia implementation
5+
using .RFunctions:
6+
chisqcdf,
7+
chisqccdf,
8+
chisqlogcdf,
9+
chisqlogccdf,
10+
chisqinvcdf,
11+
chisqinvccdf,
12+
chisqinvlogcdf,
13+
chisqinvlogccdf
14+
15+
# Julia implementations
16+
# promotion ensures that we do forward e.g. `chisqpdf(::Int, ::Float32)` to
17+
# `gammapdf(::Float32, ::Int, ::Float32)` but not `gammapdf(::Float64, ::Int, ::Float32)`
18+
chisqpdf(k::Real, x::Real) = chisqpdf(promote(k, x)...)
19+
chisqpdf(k::T, x::T) where {T<:Real} = gammapdf(k / 2, 2, x)
20+
21+
chisqlogpdf(k::Real, x::Real) = chisqlogpdf(promote(k, x)...)
22+
chisqlogpdf(k::T, x::T) where {T<:Real} = gammalogpdf(k / 2, 2, x)

‎src/distrs/fdist.jl

+12-16
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,17 @@
11
# functions related to F distribution
22

3+
# R implementations
4+
# For pdf and logpdf we use the Julia implementation
5+
using .RFunctions:
6+
fdistcdf,
7+
fdistccdf,
8+
fdistlogcdf,
9+
fdistlogccdf,
10+
fdistinvcdf,
11+
fdistinvccdf,
12+
fdistinvlogcdf,
13+
fdistinvlogccdf
14+
315
# Julia implementations
416
fdistpdf(ν1::Real, ν2::Real, x::Real) = exp(fdistlogpdf(ν1, ν2, x))
517

@@ -11,19 +23,3 @@ function fdistlogpdf(ν1::T, ν2::T, x::T) where {T<:Real}
1123
val = (xlogy(ν1, ν1ν2) + xlogy(ν1 - 2, y) - xlogy(ν1 + ν2, 1 + ν1ν2 * y)) / 2 - logbeta(ν1 / 2, ν2 / 2)
1224
return x < 0 ? oftype(val, -Inf) : val
1325
end
14-
15-
for f in ("cdf", "ccdf", "logcdf", "logccdf")
16-
ff = Symbol("fdist"*f)
17-
bf = Symbol("beta"*f)
18-
@eval $ff(ν1::T, ν2::T, x::T) where {T<:Real} = $bf(ν1/2, ν2/2, inv(1 + ν2/(ν1*max(0, x))))
19-
@eval $ff(ν1::Real, ν2::Real, x::Real) = $ff(promote(ν1, ν2, x)...)
20-
end
21-
for f in ("invcdf", "invccdf", "invlogcdf", "invlogccdf")
22-
ff = Symbol("fdist"*f)
23-
bf = Symbol("beta"*f)
24-
@eval function $ff(ν1::T, ν2::T, y::T) where {T<:Real}
25-
x = $bf(ν1/2, ν2/2, y)
26-
return x/(1 - x)*ν2/ν1
27-
end
28-
@eval $ff(ν1::Real, ν2::Real, y::Real) = $ff(promote(ν1, ν2, y)...)
29-
end

‎src/distrs/gamma.jl

+7-87
Original file line numberDiff line numberDiff line change
@@ -1,17 +1,14 @@
11
# functions related to gamma distribution
22

3-
using HypergeometricFunctions: drummond1F1
4-
53
# R implementations
4+
# For pdf and logpdf we use the Julia implementation
65
using .RFunctions:
7-
# gammapdf,
8-
# gammalogpdf,
9-
# gammacdf,
10-
# gammaccdf,
11-
# gammalogcdf,
12-
# gammalogccdf,
13-
# gammainvcdf,
14-
# gammainvccdf,
6+
gammacdf,
7+
gammaccdf,
8+
gammalogcdf,
9+
gammalogccdf,
10+
gammainvcdf,
11+
gammainvccdf,
1512
gammainvlogcdf,
1613
gammainvlogccdf
1714

@@ -25,80 +22,3 @@ function gammalogpdf(k::T, θ::T, x::T) where {T<:Real}
2522
val = -loggamma(k) + xlogy(k - 1, xθ) - log(θ) - xθ
2623
return x < 0 ? oftype(val, -Inf) : val
2724
end
28-
29-
function gammacdf(k::T, θ::T, x::T) where {T<:Real}
30-
# Handle the degenerate case
31-
if iszero(k)
32-
return float(last(promote(x, x >= 0)))
33-
end
34-
return first(gamma_inc(k, max(0, x)/θ))
35-
end
36-
gammacdf(k::Real, θ::Real, x::Real) = gammacdf(map(float, promote(k, θ, x))...)
37-
38-
function gammaccdf(k::T, θ::T, x::T) where {T<:Real}
39-
# Handle the degenerate case
40-
if iszero(k)
41-
return float(last(promote(x, x < 0)))
42-
end
43-
return last(gamma_inc(k, max(0, x)/θ))
44-
end
45-
gammaccdf(k::Real, θ::Real, x::Real) = gammaccdf(map(float, promote(k, θ, x))...)
46-
47-
gammalogcdf(k::Real, θ::Real, x::Real) = _gammalogcdf(map(float, promote(k, θ, x))...)
48-
49-
# Implemented via the non-log version. For tiny values, we recompute the result with
50-
# loggamma. In that situation, there is little risk of significant cancellation.
51-
function _gammalogcdf(k::Float64, θ::Float64, x::Float64)
52-
# Handle the degenerate case
53-
if iszero(k)
54-
return log(x >= 0)
55-
end
56-
57-
xdθ = max(0, x)/θ
58-
l, u = gamma_inc(k, xdθ)
59-
if l < floatmin(Float64) && isfinite(k) && isfinite(xdθ)
60-
return -log(k) + k*log(xdθ) - xdθ + log(drummond1F1(1.0, 1 + k, xdθ)) - loggamma(k)
61-
elseif l < 0.7
62-
return log(l)
63-
else
64-
return log1p(-u)
65-
end
66-
end
67-
function _gammalogcdf(k::T, θ::T, x::T) where {T<:Union{Float16,Float32}}
68-
return T(_gammalogcdf(Float64(k), Float64(θ), Float64(x)))
69-
end
70-
71-
gammalogccdf(k::Real, θ::Real, x::Real) = _gammalogccdf(map(float, promote(k, θ, x))...)
72-
73-
# Implemented via the non-log version. For tiny values, we recompute the result with
74-
# loggamma. In that situation, there is little risk of significant cancellation.
75-
function _gammalogccdf(k::Float64, θ::Float64, x::Float64)
76-
# Handle the degenerate case
77-
if iszero(k)
78-
return log(x < 0)
79-
end
80-
81-
xdθ = max(0, x)/θ
82-
l, u = gamma_inc(k, xdθ)
83-
if u < floatmin(Float64)
84-
return loggamma(k, xdθ) - loggamma(k)
85-
elseif u < 0.7
86-
return log(u)
87-
else
88-
return log1p(-l)
89-
end
90-
end
91-
92-
function _gammalogccdf(k::T, θ::T, x::T) where {T<:Union{Float16,Float32}}
93-
return T(_gammalogccdf(Float64(k), Float64(θ), Float64(x)))
94-
end
95-
96-
function gammainvcdf(k::Real, θ::Real, p::Real)
97-
_k, _θ, _p = promote(k, θ, p)
98-
return _θ*gamma_inc_inv(_k, _p, 1 - _p)
99-
end
100-
101-
function gammainvccdf(k::Real, θ::Real, p::Real)
102-
_k, _θ, _p = promote(k, θ, p)
103-
return _θ*gamma_inc_inv(_k, 1 - _p, _p)
104-
end

‎src/distrs/pois.jl

+5-15
Original file line numberDiff line numberDiff line change
@@ -1,13 +1,12 @@
11
# functions related to Poisson distribution
22

33
# R implementations
4+
# For pdf and logpdf we use the Julia implementation
45
using .RFunctions:
5-
# poispdf,
6-
# poislogpdf,
7-
# poiscdf,
8-
# poisccdf,
9-
# poislogcdf,
10-
# poislogccdf,
6+
poiscdf,
7+
poisccdf,
8+
poislogcdf,
9+
poislogccdf,
1110
poisinvcdf,
1211
poisinvccdf,
1312
poisinvlogcdf,
@@ -21,12 +20,3 @@ function poislogpdf(λ::T, x::T) where {T <: Real}
2120
val = xlogy(x, λ) - λ - loggamma(x + 1)
2221
return x >= 0 && isinteger(x) ? val : oftype(val, -Inf)
2322
end
24-
25-
# Just use the Gamma definitions
26-
poiscdf(λ::Real, x::Real) = gammaccdf(max(0, floor(x + 1)), 1, λ)
27-
28-
poisccdf(λ::Real, x::Real) = gammacdf(max(0, floor(x + 1)), 1, λ)
29-
30-
poislogcdf(λ::Real, x::Real) = gammalogccdf(max(0, floor(x + 1)), 1, λ)
31-
32-
poislogccdf(λ::Real, x::Real) = gammalogcdf(max(0, floor(x + 1)), 1, λ)

‎src/distrs/tdist.jl

+1-2
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,8 @@
11
# functions related to student's T distribution
22

33
# R implementations
4+
# For pdf and logpdf we use the Julia implementation
45
using .RFunctions:
5-
# tdistpdf,
6-
# tdistlogpdf,
76
tdistcdf,
87
tdistccdf,
98
tdistlogcdf,

0 commit comments

Comments
 (0)